diff --git a/README.md b/README.md index c26baa93..2e6ecbd3 100644 --- a/README.md +++ b/README.md @@ -1,8 +1,10 @@ -# Human: 3D Face Detection, Body Pose, Hand & Finger Tracking, Iris Tracking, Age & Gender Prediction & Emotion Prediction +# Human Library + +## 3D Face Detection, Body Pose, Hand & Finger Tracking, Iris Tracking, Age & Gender Prediction & Emotion Prediction - [**Documentation**](https://github.com/vladmandic/human#readme) - [**Code Repository**](https://github.com/vladmandic/human) -- [**Package**](https://www.npmjs.com/package/@vladmandic/human) +- [**NPM Package**](https://www.npmjs.com/package/@vladmandic/human) - [**Issues Tracker**](https://github.com/vladmandic/human/issues) - [**Live Demo**](https://vladmandic.github.io/human/demo/demo-esm.html) @@ -30,6 +32,25 @@ Compatible with Browser, WebWorker and NodeJS execution! There are multiple ways to use `Human` library, pick one that suits you: +### Included + +- `dist/human.js`: IIFE format minified bundle with TFJS for Browsers +- `dist/human.esm.js`: ESM format minified bundle with TFJS for Browsers +- `dist/human.esm-nobundle.js`: ESM format non-minified bundle without TFJS for Browsers +- `dist/human.cjs`: CommonJS format minified bundle with TFJS for NodeJS +- `dist/human-nobundle.cjs`: CommonJS format non-minified bundle without TFJS for NodeJS + +All versions include `sourcemap` + +Defaults: +```json + { + "main": "dist/human.cjs", + "module": "dist/human.esm.js", + "browser": "dist/human.esm.js", + } +``` + ### 1. [IIFE](https://developer.mozilla.org/en-US/docs/Glossary/IIFE) script *Simplest way for usage within Browser* @@ -43,8 +64,6 @@ Simply download `dist/human.js`, include it in your `HTML` file & it's ready to IIFE script auto-registers global namespace `human` within global `Window` object This way you can also use `Human` library within embbedded ` + + + diff --git a/demo/demo-esm.js b/demo/demo-esm.js index f8bdf669..7d0521b2 100644 --- a/demo/demo-esm.js +++ b/demo/demo-esm.js @@ -1,8 +1,9 @@ -/* global QuickSettings */ +/* global tf, QuickSettings */ import human from '../dist/human.esm.js'; const ui = { + backend: 'wasm', baseColor: 'rgba(255, 200, 255, 0.3)', baseLabel: 'rgba(255, 200, 255, 0.8)', baseFont: 'small-caps 1.2rem "Segoe UI"', @@ -34,6 +35,15 @@ function str(...msg) { return line; } +async function setupTF() { + if (ui.backend === 'wasm') { + tf.env().set('WASM_HAS_SIMD_SUPPORT', false); + tf.env().set('WASM_HAS_MULTITHREAD_SUPPORT', true); + } + await human.tf.setBackend(ui.backend); + await human.tf.ready(); +} + async function drawFace(result, canvas) { const ctx = canvas.getContext('2d'); ctx.strokeStyle = ui.baseColor; @@ -217,9 +227,10 @@ async function runHumanDetect(input, canvas) { drawHand(result.hand, canvas); // update log const engine = await human.tf.engine(); - const memory = `Memory: ${engine.state.numBytes.toLocaleString()} bytes ${engine.state.numDataBuffers.toLocaleString()} buffers ${engine.state.numTensors.toLocaleString()} tensors`; + const memory = `${engine.state.numBytes.toLocaleString()} bytes ${engine.state.numDataBuffers.toLocaleString()} buffers ${engine.state.numTensors.toLocaleString()} tensors`; + const gpu = engine.backendInstance.numBytesInGPU ? `GPU: ${engine.backendInstance.numBytesInGPU.toLocaleString()} bytes` : ''; log.innerText = ` - TFJS Version: ${human.tf.version_core} | ${memory} | GPU: ${engine.backendInstance.numBytesInGPU.toLocaleString()} bytes + TFJS Version: ${human.tf.version_core} | Backend: ${human.tf.getBackend()} | Memory: ${memory} ${gpu} Performance: ${str(result.performance)} | Object size: ${(str(result)).length.toLocaleString()} bytes `; // rinse & repeate @@ -255,6 +266,11 @@ function setupGUI() { } runHumanDetect(video, canvas); }); + settings.addDropDown('Backend', ['webgl', 'wasm', 'cpu'], (val) => { + ui.backend = val.value; + setupTF(); + }); + settings.addHTML('title', 'Enabled Models'); settings.hideTitle('title'); settings.addBoolean('Face Detect', config.face.enabled, (val) => config.face.enabled = val); settings.addBoolean('Face Mesh', config.face.mesh.enabled, (val) => config.face.mesh.enabled = val); settings.addBoolean('Face Iris', config.face.iris.enabled, (val) => config.face.iris.enabled = val); @@ -263,7 +279,7 @@ function setupGUI() { settings.addBoolean('Face Emotion', config.face.emotion.enabled, (val) => config.face.emotion.enabled = val); settings.addBoolean('Body Pose', config.body.enabled, (val) => config.body.enabled = val); settings.addBoolean('Hand Pose', config.hand.enabled, (val) => config.hand.enabled = val); - settings.addHTML('line3', '
'); settings.hideTitle('line3'); + settings.addHTML('title', 'Model Parameters'); settings.hideTitle('title'); settings.addRange('Max Objects', 1, 20, 5, 1, (val) => { config.face.detector.maxFaces = parseInt(val); config.body.maxDetections = parseInt(val); @@ -288,7 +304,7 @@ function setupGUI() { config.face.detector.iouThreshold = parseFloat(val); config.hand.iouThreshold = parseFloat(val); }); - settings.addHTML('line1', '
'); settings.hideTitle('line1'); + settings.addHTML('title', 'UI Options'); settings.hideTitle('title'); settings.addBoolean('Draw Boxes', true); settings.addBoolean('Draw Points', true); settings.addBoolean('Draw Polygons', true); @@ -342,8 +358,7 @@ async function setupImage() { async function main() { // initialize tensorflow - await human.tf.setBackend('webgl'); - await human.tf.ready(); + await setupTF(); // setup ui control panel await setupGUI(); // setup webcam diff --git a/dist/human.node.js b/dist/human-nobundle.cjs similarity index 99% rename from dist/human.node.js rename to dist/human-nobundle.cjs index feeac566..295a5f71 100644 --- a/dist/human.node.js +++ b/dist/human-nobundle.cjs @@ -5236,4 +5236,4 @@ exports.ssrnet = ssrnet; exports.posenet = posenet; exports.handpose = handpose; exports.tf = tf; -//# sourceMappingURL=human.node.js.map +//# sourceMappingURL=human-nobundle.cjs.map diff --git a/dist/human.node.js.map b/dist/human-nobundle.cjs.map similarity index 100% rename from dist/human.node.js.map rename to dist/human-nobundle.cjs.map diff --git a/dist/human.cjs b/dist/human.cjs new file mode 100644 index 00000000..47757f19 --- /dev/null +++ b/dist/human.cjs @@ -0,0 +1,4006 @@ +var Rs=Object.defineProperty,s2=Object.prototype.hasOwnProperty;var Ae=(n,t)=>()=>(t||(t={exports:{}},n(t.exports,t)),t.exports),ef=n=>Rs(n,"__esModule",{value:!0}),vl=(n,t)=>{ef(n);for(var e in t)Rs(n,e,{get:t[e],enumerable:!0})},o2=(n,t)=>{if(ef(n),typeof t=="object"||typeof t=="function")for(let e in t)!s2.call(n,e)&&e!=="default"&&Rs(n,e,{get:()=>t[e],enumerable:!0});return n},Aa=n=>n&&n.__esModule?n:o2(Rs({},"default",{value:n,enumerable:!0}),n);var hf=Ae(l2=>{vl(l2,{FetchError:()=>Ft,Headers:()=>fn,Request:()=>Mi,Response:()=>Rn,default:()=>w2});const ln=Aa(require("stream")),yl=Aa(require("http")),Os=Aa(require("url")),tf=Aa(require("https")),oi=Aa(require("zlib")),u2=ln.default.Readable,Yn=Symbol("buffer"),bl=Symbol("type");class Ta{constructor(){this[bl]="";const n=arguments[0],t=arguments[1],e=[];let i=0;if(n){const a=n,s=Number(a.length);for(let o=0;o1&&arguments[1]!==void 0?arguments[1]:{},i=e.size;let r=i===void 0?0:i;var a=e.timeout;let s=a===void 0?0:a;n==null?n=null:rf(n)?n=Buffer.from(n.toString()):Na(n)||(Buffer.isBuffer(n)||(Object.prototype.toString.call(n)==="[object ArrayBuffer]"?n=Buffer.from(n):ArrayBuffer.isView(n)?n=Buffer.from(n.buffer,n.byteOffset,n.byteLength):n instanceof ln.default||(n=Buffer.from(String(n))))),this[Kn]={body:n,disturbed:!1,error:null},this.size=r,this.timeout=s,n instanceof ln.default&&n.on("error",function(o){const l=o.name==="AbortError"?o:new Ft(`Invalid response body while trying to fetch ${t.url}: ${o.message}`,"system",o);t[Kn].error=l})}xt.prototype={get body(){return this[Kn].body},get bodyUsed(){return this[Kn].disturbed},arrayBuffer(){return Cr.call(this).then(function(n){return n.buffer.slice(n.byteOffset,n.byteOffset+n.byteLength)})},blob(){let n=this.headers&&this.headers.get("content-type")||"";return Cr.call(this).then(function(t){return Object.assign(new Ta([],{type:n.toLowerCase()}),{[Yn]:t})})},json(){var n=this;return Cr.call(this).then(function(t){try{return JSON.parse(t.toString())}catch(e){return xt.Promise.reject(new Ft(`invalid json response body at ${n.url} reason: ${e.message}`,"invalid-json"))}})},text(){return Cr.call(this).then(function(n){return n.toString()})},buffer(){return Cr.call(this)},textConverted(){var n=this;return Cr.call(this).then(function(t){return c2(t,n.headers)})}};Object.defineProperties(xt.prototype,{body:{enumerable:!0},bodyUsed:{enumerable:!0},arrayBuffer:{enumerable:!0},blob:{enumerable:!0},json:{enumerable:!0},text:{enumerable:!0}});xt.mixIn=function(n){for(const t of Object.getOwnPropertyNames(xt.prototype))if(!(t in n)){const e=Object.getOwnPropertyDescriptor(xt.prototype,t);Object.defineProperty(n,t,e)}};function Cr(){var n=this;if(this[Kn].disturbed)return xt.Promise.reject(new TypeError(`body used already for: ${this.url}`));if(this[Kn].disturbed=!0,this[Kn].error)return xt.Promise.reject(this[Kn].error);let t=this.body;if(t===null)return xt.Promise.resolve(Buffer.alloc(0));if(Na(t)&&(t=t.stream()),Buffer.isBuffer(t))return xt.Promise.resolve(t);if(!(t instanceof ln.default))return xt.Promise.resolve(Buffer.alloc(0));let e=[],i=0,r=!1;return new xt.Promise(function(a,s){let o;n.timeout&&(o=setTimeout(function(){r=!0,s(new Ft(`Response timeout while trying to fetch ${n.url} (over ${n.timeout}ms)`,"body-timeout"))},n.timeout)),t.on("error",function(l){l.name==="AbortError"?(r=!0,s(l)):s(new Ft(`Invalid response body while trying to fetch ${n.url}: ${l.message}`,"system",l))}),t.on("data",function(l){if(r||l===null)return;if(n.size&&i+l.length>n.size){r=!0,s(new Ft(`content size at ${n.url} over limit: ${n.size}`,"max-size"));return}i+=l.length,e.push(l)}),t.on("end",function(){if(r)return;clearTimeout(o);try{a(Buffer.concat(e,i))}catch(l){s(new Ft(`Could not create Buffer from response body for ${n.url}: ${l.message}`,"system",l))}})})}function c2(n,t){if(typeof wl!="function")throw new Error("The package `encoding` must be installed to use the textConverted() function");const e=t.get("content-type");let i="utf-8",r,a;return e&&(r=/charset=([^;]*)/i.exec(e)),a=n.slice(0,1024).toString(),!r&&a&&(r=/0&&arguments[0]!==void 0?arguments[0]:void 0;if(this[ut]=Object.create(null),n instanceof fn){const t=n.raw(),e=Object.keys(t);for(const i of e)for(const r of t[i])this.append(i,r);return}if(!(n==null))if(typeof n=="object"){const t=n[Symbol.iterator];if(t!=null){if(typeof t!="function")throw new TypeError("Header pairs must be iterable");const e=[];for(const i of n){if(typeof i!="object"||typeof i[Symbol.iterator]!="function")throw new TypeError("Each header pair must be iterable");e.push(Array.from(i))}for(const i of e){if(i.length!==2)throw new TypeError("Each header pair must be a name/value tuple");this.append(i[0],i[1])}}else for(const e of Object.keys(n)){const i=n[e];this.append(e,i)}}else throw new TypeError("Provided initializer must be an object")}get(n){n=`${n}`,xa(n);const t=Rr(this[ut],n);return t===void 0?null:this[ut][t].join(", ")}forEach(n){let t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:void 0,e=Ll(this),i=0;for(;i1&&arguments[1]!==void 0?arguments[1]:"key+value";const e=Object.keys(n[ut]).sort();return e.map(t==="key"?function(i){return i.toLowerCase()}:t==="value"?function(i){return n[ut][i].join(", ")}:function(i){return[i.toLowerCase(),n[ut][i].join(", ")]})}const Al=Symbol("internal");function Il(n,t){const e=Object.create(Tl);return e[Al]={target:n,kind:t,index:0},e}const Tl=Object.setPrototypeOf({next(){if(!this||Object.getPrototypeOf(this)!==Tl)throw new TypeError("Value of `this` is not a HeadersIterator");var n=this[Al];const t=n.target,e=n.kind,i=n.index,r=Ll(t,e),a=r.length;return i>=a?{value:void 0,done:!0}:(this[Al].index=i+1,{value:r[i],done:!1})}},Object.getPrototypeOf(Object.getPrototypeOf([][Symbol.iterator]())));Object.defineProperty(Tl,Symbol.toStringTag,{value:"HeadersIterator",writable:!1,enumerable:!1,configurable:!0});function d2(n){const t=Object.assign({__proto__:null},n[ut]),e=Rr(n[ut],"Host");return e!==void 0&&(t[e]=t[e][0]),t}function p2(n){const t=new fn;for(const e of Object.keys(n)){if(lf.test(e))continue;if(Array.isArray(n[e]))for(const i of n[e]){if(Sl.test(i))continue;t[ut][e]===void 0?t[ut][e]=[i]:t[ut][e].push(i)}else Sl.test(n[e])||(t[ut][e]=[n[e]])}return t}const li=Symbol("Response internals"),f2=yl.default.STATUS_CODES;class Rn{constructor(){let n=arguments.length>0&&arguments[0]!==void 0?arguments[0]:null,t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:{};xt.call(this,n,t);const e=t.status||200,i=new fn(t.headers);if(n!=null&&!i.has("Content-Type")){const r=sf(n);r&&i.append("Content-Type",r)}this[li]={url:t.url,status:e,statusText:t.statusText||f2[e],headers:i,counter:t.counter}}get url(){return this[li].url||""}get status(){return this[li].status}get ok(){return this[li].status>=200&&this[li].status<300}get redirected(){return this[li].counter>0}get statusText(){return this[li].statusText}get headers(){return this[li].headers}clone(){return new Rn(af(this),{url:this.url,status:this.status,statusText:this.statusText,headers:this.headers,ok:this.ok,redirected:this.redirected})}}xt.mixIn(Rn.prototype);Object.defineProperties(Rn.prototype,{url:{enumerable:!0},status:{enumerable:!0},ok:{enumerable:!0},redirected:{enumerable:!0},statusText:{enumerable:!0},headers:{enumerable:!0},clone:{enumerable:!0}});Object.defineProperty(Rn.prototype,Symbol.toStringTag,{value:"Response",writable:!1,enumerable:!1,configurable:!0});const jn=Symbol("Request internals"),Nl=Os.default.parse,m2=Os.default.format,g2="destroy"in ln.default.Readable.prototype;function Es(n){return typeof n=="object"&&typeof n[jn]=="object"}function v2(n){const t=n&&typeof n=="object"&&Object.getPrototypeOf(n);return!!(t&&t.constructor.name==="AbortSignal")}class Mi{constructor(n){let t=arguments.length>1&&arguments[1]!==void 0?arguments[1]:{},e;Es(n)?e=Nl(n.url):(n&&n.href?e=Nl(n.href):e=Nl(`${n}`),n={});let i=t.method||n.method||"GET";if(i=i.toUpperCase(),(t.body!=null||Es(n)&&n.body!==null)&&(i==="GET"||i==="HEAD"))throw new TypeError("Request with GET/HEAD method cannot have body");let r=t.body!=null?t.body:Es(n)&&n.body!==null?af(n):null;xt.call(this,r,{timeout:t.timeout||n.timeout||0,size:t.size||n.size||0});const a=new fn(t.headers||n.headers||{});if(r!=null&&!a.has("Content-Type")){const o=sf(r);o&&a.append("Content-Type",o)}let s=Es(n)?n.signal:null;if("signal"in t&&(s=t.signal),s!=null&&!v2(s))throw new TypeError("Expected signal to be an instanceof AbortSignal");this[jn]={method:i,redirect:t.redirect||n.redirect||"follow",headers:a,parsedURL:e,signal:s},this.follow=t.follow!==void 0?t.follow:n.follow!==void 0?n.follow:20,this.compress=t.compress!==void 0?t.compress:n.compress!==void 0?n.compress:!0,this.counter=t.counter||n.counter||0,this.agent=t.agent||n.agent}get method(){return this[jn].method}get url(){return m2(this[jn].parsedURL)}get headers(){return this[jn].headers}get redirect(){return this[jn].redirect}get signal(){return this[jn].signal}clone(){return new Mi(this)}}xt.mixIn(Mi.prototype);Object.defineProperty(Mi.prototype,Symbol.toStringTag,{value:"Request",writable:!1,enumerable:!1,configurable:!0});Object.defineProperties(Mi.prototype,{method:{enumerable:!0},url:{enumerable:!0},headers:{enumerable:!0},redirect:{enumerable:!0},clone:{enumerable:!0},signal:{enumerable:!0}});function y2(n){const t=n[jn].parsedURL,e=new fn(n[jn].headers);if(e.has("Accept")||e.set("Accept","*/*"),!t.protocol||!t.hostname)throw new TypeError("Only absolute URLs are supported");if(!/^https?:$/.test(t.protocol))throw new TypeError("Only HTTP(S) protocols are supported");if(n.signal&&n.body instanceof ln.default.Readable&&!g2)throw new Error("Cancellation of streamed requests with AbortSignal is not supported in node < 8");let i=null;if(n.body==null&&/^(POST|PUT)$/i.test(n.method)&&(i="0"),n.body!=null){const a=of(n);typeof a=="number"&&(i=String(a))}i&&e.set("Content-Length",i),e.has("User-Agent")||e.set("User-Agent","node-fetch/1.0 (+https://github.com/bitinn/node-fetch)"),n.compress&&!e.has("Accept-Encoding")&&e.set("Accept-Encoding","gzip,deflate");let r=n.agent;return typeof r=="function"&&(r=r(t)),!e.has("Connection")&&!r&&e.set("Connection","close"),Object.assign({},t,{method:n.method,headers:d2(e),agent:r})}function Ca(n){Error.call(this,n),this.type="aborted",this.message=n,Error.captureStackTrace(this,this.constructor)}Ca.prototype=Object.create(Error.prototype);Ca.prototype.constructor=Ca;Ca.prototype.name="AbortError";const cf=ln.default.PassThrough,b2=Os.default.resolve;function ui(n,t){if(!ui.Promise)throw new Error("native promise missing, set fetch.Promise to your favorite alternative");return xt.Promise=ui.Promise,new ui.Promise(function(e,i){const r=new Mi(n,t),a=y2(r),s=(a.protocol==="https:"?tf.default:yl.default).request,o=r.signal;let l=null;const u=function(){let m=new Ca("The user aborted a request.");if(i(m),r.body&&r.body instanceof ln.default.Readable&&r.body.destroy(m),!l||!l.body)return;l.body.emit("error",m)};if(o&&o.aborted){u();return}const c=function(){u(),p()},h=s(a);let d;o&&o.addEventListener("abort",c);function p(){h.abort(),o&&o.removeEventListener("abort",c),clearTimeout(d)}r.timeout&&h.once("socket",function(f){d=setTimeout(function(){i(new Ft(`network timeout at: ${r.url}`,"request-timeout")),p()},r.timeout)}),h.on("error",function(f){i(new Ft(`request to ${r.url} failed, reason: ${f.message}`,"system",f)),p()}),h.on("response",function(f){clearTimeout(d);const m=p2(f.headers);if(ui.isRedirect(f.statusCode)){const S=m.get("Location"),L=S===null?null:b2(r.url,S);switch(r.redirect){case"error":i(new Ft(`uri requested responds with a redirect, redirect mode is set to error: ${r.url}`,"no-redirect")),p();return;case"manual":if(L!==null)try{m.set("Location",L)}catch(C){i(C)}break;case"follow":if(L===null)break;if(r.counter>=r.follow){i(new Ft(`maximum redirect reached at: ${r.url}`,"max-redirect")),p();return}const N={headers:new fn(r.headers),follow:r.follow,counter:r.counter+1,agent:r.agent,compress:r.compress,method:r.method,body:r.body,signal:r.signal,timeout:r.timeout,size:r.size};if(f.statusCode!==303&&r.body&&of(r)===null){i(new Ft("Cannot follow redirect with body being a readable stream","unsupported-redirect")),p();return}(f.statusCode===303||(f.statusCode===301||f.statusCode===302)&&r.method==="POST")&&(N.method="GET",N.body=void 0,N.headers.delete("content-length")),e(ui(new Mi(L,N))),p();return}}f.once("end",function(){o&&o.removeEventListener("abort",c)});let g=f.pipe(new cf);const v={url:r.url,status:f.statusCode,statusText:f.statusMessage,headers:m,size:r.size,timeout:r.timeout,counter:r.counter},b=m.get("Content-Encoding");if(!r.compress||r.method==="HEAD"||b===null||f.statusCode===204||f.statusCode===304){l=new Rn(g,v),e(l);return}const w={flush:oi.default.Z_SYNC_FLUSH,finishFlush:oi.default.Z_SYNC_FLUSH};if(b=="gzip"||b=="x-gzip"){g=g.pipe(oi.default.createGunzip(w)),l=new Rn(g,v),e(l);return}if(b=="deflate"||b=="x-deflate"){const S=f.pipe(new cf);S.once("data",function(L){(L[0]&15)===8?g=g.pipe(oi.default.createInflate()):g=g.pipe(oi.default.createInflateRaw()),l=new Rn(g,v),e(l)});return}if(b=="br"&&typeof oi.default.createBrotliDecompress=="function"){g=g.pipe(oi.default.createBrotliDecompress()),l=new Rn(g,v),e(l);return}l=new Rn(g,v),e(l)}),h2(h,r)})}ui.isRedirect=function(n){return n===301||n===302||n===303||n===307||n===308};ui.Promise=global.Promise;var w2=ui});var rr=Ae(I=>{"use strict";Object.defineProperty(I,"__esModule",{value:!0});var xl=function(n,t){return xl=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},xl(n,t)};function $n(n,t){xl(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function de(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function pe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]0;)i=Math.random()*t|0,t--,e=n[t],n[t]=n[i],n[i]=e}function Oa(n,t,e){return Math.max(n,Math.min(t,e))}function F2(n){return n%2===0?n:n+1}function W2(n){for(var t=0,e=0;e=e){r();return}setTimeout(s,o)};s()})}function gm(n,t){for(var e=1,i=-1,r=0;r=0)e*=n[r];else if(n[r]===-1){if(i!==-1)throw Error("Shapes can only have 1 implicit size. "+("Found -1 at dim "+i+" and dim "+r));i=r}else if(n[r]<0)throw Error("Shapes can not be < 0. Found "+n[r]+" at dim "+r);if(i===-1){if(t>0&&t!==e)throw Error("Size("+t+") must match the product of shape "+n);return n}if(e===0)throw Error("Cannot infer the missing size in ["+n+"] when there are 0 elements");if(t%e!==0)throw Error("The implicit shape can't be a fractional number. "+("Got "+t+" / "+e));var a=n.slice();return a[i]=t/e,a}function Ze(n,t){var e=t.length;return n=n==null?t.map(function(i,r){return r}):[].concat(n),E(n.every(function(i){return i>=-e&&io)&&n[o]===1&&(e.push(n[o]),i.push(o)),a[s]<=o&&s++}n[o]!==1&&(e.push(n[o]),i.push(o))}return{newShape:e,keptDims:i}}function Ws(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else throw new Error("Unknown data type "+n);return e}function ym(n,t){var e=null;if(n==null||n==="float32")e=new Float32Array(t);else if(n==="int32")e=new Int32Array(t);else if(n==="bool")e=new Uint8Array(t);else if(n==="string")e=new Array(t);else throw new Error("Unknown data type "+n);return e}function bm(n,t){for(var e=0;e=0;--i)e[i]=e[i+1]*n[i+1];return e}function V2(n,t){return t==="string"?bc(n):zs([n],t)}function zs(n,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(n)&&(n=Vi(n)),Me().getBool("DEBUG")&&bm(n,t),q2(n,t))return n;if(t==null||t==="float32"||t==="complex64")return new Float32Array(n);if(t==="int32")return new Int32Array(n);if(t==="bool"){for(var e=new Uint8Array(n.length),i=0;i=0,function(){return"Tensor must have a shape comprised of positive integers but got "+("shape ["+n+"].")})})}function Y2(n,t){return Me().platform.fetch(n,t)}function bc(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",Me().platform.encode(n,t)}function Ic(n,t){return t===void 0&&(t="utf-8"),t=t||"utf-8",Me().platform.decode(n,t)}function K2(n,t,e){if(t===0)return 0;if(t===1)return n[0];for(var i=n[n.length-1],r=0;r0?m:"")+" "}}console.log("%c"+l+" %c"+o+" %c"+u+"D "+h+" %c"+c+" %c"+d+" %c"+s,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")},n}();function Q2(n,t,e){for(var i={},r={},a=0;a=0;a--)for(var s=n[a],o=s.inputs,h=0;h=0;a--)r(a)}var xm=20,Da=3,Ac=7;function nA(n,t,e,i){var r=Er(t),a=tA(n,t,e,r),s=t.length,o=Ps(n,t,e,r,a),l=["Tensor"];return i&&(l.push(" dtype: "+e),l.push(" rank: "+s),l.push(" shape: ["+t+"]"),l.push(" values:")),l.push(o.map(function(u){return" "+u}).join(` +`)),l.join(` +`)}function tA(n,t,e,i){var r=st(t),a=i[i.length-1],s=new Array(a).fill(0),o=t.length,l=e==="complex64"?Fa(n):n;if(o>1)for(var u=0;uxm){var c=Da*s,h=Array.from(n.slice(0,c)),d=Array.from(n.slice((o-Da)*s,o*s));return e==="complex64"&&(h=Fa(h),d=Fa(d)),["["+h.map(function(C,R){return ka(C,r[R],e)}).join(", ")+", ..., "+d.map(function(C,R){return ka(C,r[o-Da+R],e)}).join(", ")+"]"]}var p=e==="complex64"?Fa(n):Array.from(n);return["["+p.map(function(C,R){return ka(C,r[R],e)}).join(", ")+"]"]}var f=t.slice(1),m=i.slice(1),g=i[0]*s,v=[];if(o>xm){for(var b=0;b=this.shape[i]){var o="Requested out of range element at "+t+". "+(" Buffer shape="+this.shape);throw new Error(o)}i++}for(var l=t[t.length-1],u=0;u0)throw new Error("Backend '"+this.backendName+"' has an internal memory leak "+("("+o+" data ids) after running '"+t+"'"))},n.prototype.runKernelFunc=function(t,e,i,r,a,s,o){var l=this,u,c=[],h=this.isTapeOn();r==null&&(r=this.state.activeScope!=null?this.state.activeScope.name:"");var d=this.state.numBytes,p=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);var f,m=vc(r,this.backendName),g;if(m!=null)f=function(){var w=l.backend.numDataIds();g=m.kernelFunc({inputs:e,attrs:a,backend:l.backend});var S=Array.isArray(g)?g:[g];l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S);var L=S.map(function(R){var D=R.dataId,k=R.shape,W=R.dtype;return l.makeTensorFromDataId(D,k,W)});if(h){var N=l.getTensorsForGradient(r,e,L);if(N==null){o==null&&(o=[]);var C=L.filter(function(R,D){return o[D]});N=(s||[]).slice().concat(C)}c=l.saveTensorsForBackwardMode(N)}return L};else{var v=function(w){if(!h)return;c=w.map(function(S){return l.keep(l.clone(S))})};f=function(){var w=l.backend.numDataIds();g=l.tidy(function(){return t(l.backend,v)});var S=Array.isArray(g)?g:[g];return l.shouldCheckForMemLeaks()&&l.checkKernelForMemLeak(r,w,S),S}}var b;return this.scopedRun(function(){return l.state.kernelDepth++},function(){return l.state.kernelDepth--},function(){!l.ENV.getBool("DEBUG")&&!l.state.profiling?u=f():(b=l.profiler.profileKernel(r,e,function(){return f()}),l.ENV.getBool("DEBUG")&&l.profiler.logKernelProfile(b),u=b.outputs)}),h&&this.addTapeNode(r,e,u,i,c,a),this.state.profiling&&this.state.activeProfile.kernels.push({name:r,bytesAdded:this.state.numBytes-d,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-p,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(e).map(function(w){return e[w]!=null?e[w].shape:null}),outputShapes:u.map(function(w){return w.shape}),kernelTimeMs:b.timeMs,extraInfo:b.extraInfo}),Array.isArray(g)?u:u[0]},n.prototype.saveTensorsForBackwardMode=function(t){var e=this,i=t.map(function(r){return e.keep(e.clone(r))});return i},n.prototype.getTensorsForGradient=function(t,e,i){var r=yc(t);if(r!=null){var a=r.inputsToSave||[],s=r.outputsToSave||[],o=void 0;r.saveAllInputs?(E(Array.isArray(e),function(){return"saveAllInputs is true, expected inputs to be an array."}),o=Object.keys(e).map(function(u){return e[u]})):o=a.map(function(u){return e[u]});var l=i.filter(function(u,c){return s[c]});return o.concat(l)}return null},n.prototype.makeTensor=function(t,e,i,r){if(t==null)throw new Error("Values passed to engine.makeTensor() are null");i=i||"float32",r=r||this.backend;var a=t;i==="string"&&ci(t[0])&&(a=t.map(function(c){return bc(c)}));var s=r.write(a,e,i),o=new Y(e,i,s,this.nextTensorId());if(this.incRef(o,r),i==="string"){var l=this.state.tensorInfo.get(s),u=Im(a);this.state.numBytes+=u-l.bytes,l.bytes=u}return o},n.prototype.makeTensorFromDataId=function(t,e,i,r){i=i||"float32";var a=new Y(e,i,t,this.nextTensorId());return this.incRef(a,r),a},n.prototype.makeVariable=function(t,e,i,r){e===void 0&&(e=!0),i=i||this.nextVariableId().toString(),r!=null&&r!==t.dtype&&(t=t.cast(r));var a=new Wa(t,e,i,this.nextTensorId());if(this.state.registeredVariables[a.name]!=null)throw new Error("Variable with name "+a.name+" was already registered");return this.state.registeredVariables[a.name]=a,this.incRef(a,this.backend),a},n.prototype.incRef=function(t,e){var i=this.state.tensorInfo.has(t.dataId)?this.state.tensorInfo.get(t.dataId).refCount:0;if(this.state.numTensors++,t.dtype==="string"&&this.state.numStringTensors++,i===0){this.state.numDataBuffers++;var r=0;t.dtype!=="complex64"&&t.dtype!=="string"&&(r=t.size*Lm(t.dtype)),this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:r,refCount:0}),this.state.numBytes+=r}this.state.tensorInfo.get(t.dataId).refCount++,t instanceof Wa||this.track(t)},n.prototype.disposeTensor=function(t){if(!this.state.tensorInfo.has(t.dataId))return;this.state.numTensors--,t.dtype==="string"&&this.state.numStringTensors--;var e=this.state.tensorInfo.get(t.dataId),i=e.refCount;i<=1?(t.dtype!=="complex64"&&(this.state.numBytes-=e.bytes),this.state.numDataBuffers--,e.backend.disposeData(t.dataId),this.state.tensorInfo.delete(t.dataId)):this.state.tensorInfo.get(t.dataId).refCount--},n.prototype.disposeVariables=function(){for(var t in this.state.registeredVariables){var e=this.state.registeredVariables[t];this.disposeVariable(e)}},n.prototype.disposeVariable=function(t){this.disposeTensor(t),this.state.registeredVariables[t.name]!=null&&delete this.state.registeredVariables[t.name]},n.prototype.memory=function(){var t=this.backend.memory();return t.numTensors=this.state.numTensors,t.numDataBuffers=this.state.numDataBuffers,t.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(t.unreliable=!0,t.reasons==null&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t},n.prototype.profile=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return pe(this,function(c){switch(c.label){case 0:return this.state.profiling=!0,e=this.state.numBytes,i=this.state.numTensors,this.state.activeProfile.kernels=[],r=this.state.activeProfile,[4,t()];case 1:r.result=c.sent(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max.apply(Math,this.state.activeProfile.kernels.map(function(h){return h.totalBytesSnapshot})),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-i,a=0,s=this.state.activeProfile.kernels,c.label=2;case 2:return a0&&this.state.kernelDepth===0},n.prototype.addTapeNode=function(t,e,i,r,a,s){var o=this,l={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:i,saved:a},u=yc(t);u!=null&&(r=u.gradFunc),r!=null&&(l.gradient=function(c){return c=c.map(function(h,d){if(h==null){var p=i[d],f=kr(p.size,p.dtype);return o.makeTensor(f,p.shape,p.dtype)}return h}),r(c.length>1?c:c[0],a,s)}),this.state.activeTape.push(l)},n.prototype.keep=function(t){return t.kept=!0,t},n.prototype.startTape=function(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++},n.prototype.endTape=function(){this.state.gradientDepth--},n.prototype.startScope=function(t){var e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e},n.prototype.endScope=function(t){for(var e=this,i=Rc(t),r=new Set(i.map(function(l){return l.id})),a=0;a0,function(){return"gradients() received an empty list of xs."}),i!=null&&i.dtype!=="float32")throw new Error("dy must have 'float32' dtype, but has '"+i.dtype+"'");var s=this.scopedRun(function(){return a.startTape()},function(){return a.endTape()},function(){return a.tidy("forward",t)});E(s instanceof Y,function(){return"The result y returned by f() must be a tensor."});var o=Q2(this.state.activeTape,e,s);if(!r&&o.length===0&&e.length>0)throw new Error("Cannot compute gradient of y=f(x) with respect to x. Make sure that the f you passed encloses all operations that lead from x to y.");return this.tidy("backward",function(){var l={};l[s.id]=i??cA(s.shape),eA(l,o,function(c){return a.tidy(c)},hA);var u=e.map(function(c){return l[c.id]});return a.state.gradientDepth===0&&(a.state.activeTape.forEach(function(c){for(var h=0,d=c.saved;h0,function(){return"Element arr["+e.join("][")+"] should be a primitive, "+("but is an array of "+n.length+" elements")}),E(n.length===t[0],function(){return"Element arr["+e.join("][")+"] should have "+t[0]+" "+("elements, but has "+n.length+" elements")});for(var i=t.slice(1),r=0;r=0&&(r=i),Wm(i,r,t,e),n==null||!Et(n)&&!Array.isArray(n)&&typeof n!="number"&&typeof n!="boolean"&&typeof n!="string"){var a=n==null?"null":n.constructor.name;throw new Error("Argument '"+t+"' passed to '"+e+"' must be a "+("Tensor or TensorLike, but got '"+a+"'"))}var s=En(n,r);!Et(n)&&!Array.isArray(n)&&(n=[n]);var o=!0,l=r!=="string"?zs(n,r):Vi(n,[],o);return z.makeTensor(l,s,r)}function Ua(n,t,e,i){if(i===void 0&&(i="numeric"),!Array.isArray(n))throw new Error("Argument "+t+" passed to "+e+" must be a `Tensor[]` or `TensorLike[]`");var r=n;return r.map(function(a,s){return O(a,t+"["+s+"]",e)},i)}var Um="__op";function U(n){var t=Object.keys(n);if(t.length!==1)throw new Error("Please provide an object with a single key (operation name) mapping to a function. Got an object with "+(t.length+" keys."));var e=t[0],i=n[e];e.endsWith("_")&&(e=e.substring(0,e.length-1)),e=e+Um;var r=function(){for(var a=[],s=0;s>10]+(o&1023)]+t[o>>10];a[s]=l}return new Float32Array(r)}}var en=function(){function n(){this.saveRouters=[],this.loadRouters=[]}return n.getInstance=function(){return n.instance==null&&(n.instance=new n),n.instance},n.registerSaveRouter=function(t){n.getInstance().saveRouters.push(t)},n.registerLoadRouter=function(t){n.getInstance().loadRouters.push(t)},n.getSaveHandlers=function(t){return n.getHandlers(t,"save")},n.getLoadHandlers=function(t,e){return n.getHandlers(t,"load",e)},n.getHandlers=function(t,e,i){var r=[],a=e==="load"?n.getInstance().loadRouters:n.getInstance().saveRouters;return a.forEach(function(s){var o=s(t,i);o!==null&&r.push(o)}),r},n}(),TA=function(n){return en.registerSaveRouter(n)},NA=function(n){return en.registerLoadRouter(n)},xA=function(n){return en.getSaveHandlers(n)},CA=function(n,t){return en.getLoadHandlers(n,t)};var kc="tensorflowjs",Fc=1,qi="models_store",mi="model_info_store";function _m(){if(!Me().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");var n=typeof window=="undefined"?self:window,t=n.indexedDB||n.mozIndexedDB||n.webkitIndexedDB||n.msIndexedDB||n.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function Wc(n){var t=n.result;t.createObjectStore(qi,{keyPath:"modelPath"}),t.createObjectStore(mi,{keyPath:"modelPath"})}var Wr=function(){function n(t){if(this.indexedDB=_m(),t==null||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return[2,this.databaseAction(this.modelPath,t)]})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.databaseAction(this.modelPath)]})})},n.prototype.databaseAction=function(t,e){var i=this;return new Promise(function(r,a){var s=i.indexedDB.open(kc,Fc);s.onupgradeneeded=function(){return Wc(s)},s.onsuccess=function(){var o=s.result;if(e==null){var l=o.transaction(qi,"readonly"),u=l.objectStore(qi),c=u.get(i.modelPath);c.onsuccess=function(){if(c.result==null)return o.close(),a(new Error("Cannot find model with path '"+i.modelPath+"' in IndexedDB."));r(c.result.modelArtifacts)},c.onerror=function(g){return o.close(),a(c.error)},l.oncomplete=function(){return o.close()}}else{var h=Ba(e),d=o.transaction(mi,"readwrite"),p=d.objectStore(mi),f=p.put({modelPath:i.modelPath,modelArtifactsInfo:h}),m;f.onsuccess=function(){m=o.transaction(qi,"readwrite");var g=m.objectStore(qi),v=g.put({modelPath:i.modelPath,modelArtifacts:e,modelArtifactsInfo:h});v.onsuccess=function(){return r({modelArtifactsInfo:h})},v.onerror=function(b){p=d.objectStore(mi);var w=p.delete(i.modelPath);w.onsuccess=function(){return o.close(),a(v.error)},w.onerror=function(S){return o.close(),a(v.error)}}},f.onerror=function(g){return o.close(),a(f.error)},d.oncomplete=function(){m==null?o.close():m.oncomplete=function(){return o.close()}}}},s.onerror=function(o){return a(s.error)}})},n.URL_SCHEME="indexeddb://",n}(),Mm=function(n){return Me().getBool("IS_BROWSER")&&(!Array.isArray(n)&&n.startsWith(Wr.URL_SCHEME))?RA(n.slice(Wr.URL_SCHEME.length)):null};en.registerSaveRouter(Mm);en.registerLoadRouter(Mm);function RA(n){return new Wr(n)}function OA(n){return n.startsWith(Wr.URL_SCHEME)?n.slice(Wr.URL_SCHEME.length):n}var EA=function(){function n(){this.indexedDB=_m()}return n.prototype.listModels=function(){return de(this,void 0,void 0,function(){var t=this;return pe(this,function(e){return[2,new Promise(function(i,r){var a=t.indexedDB.open(kc,Fc);a.onupgradeneeded=function(){return Wc(a)},a.onsuccess=function(){var s=a.result,o=s.transaction(mi,"readonly"),l=o.objectStore(mi),u=l.getAll();u.onsuccess=function(){for(var c={},h=0,d=u.result;h0,function(){return"scheme must not be an empty string."});var i=n.getInstance();E(i.managers[t]==null,function(){return"A model store manager is already registered for scheme '"+t+"'."}),i.managers[t]=e},n.getManager=function(t){var e=this.getInstance().managers[t];if(e==null)throw new Error("Cannot find model manager for scheme '"+t+"'");return e},n.getSchemes=function(){return Object.keys(this.getInstance().managers)},n}();function Vs(n){if(n.indexOf(zr)===-1)throw new Error("The url string provided does not contain a scheme. Supported schemes are: "+(""+gi.getSchemes().join(",")));return{scheme:n.split(zr)[0],path:n.split(zr)[1]}}function Gm(n,t,e){return e===void 0&&(e=!1),de(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return E(n!==t,function(){return"Old path and new path are the same: '"+n+"'"}),i=en.getLoadHandlers(n),E(i.length>0,function(){return"Copying failed because no load handler is found for source URL "+n+"."}),E(i.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("load handlers for source URL "+n+".")}),r=i[0],a=en.getSaveHandlers(t),E(a.length>0,function(){return"Copying failed because no save handler is found for destination "+("URL "+t+".")}),E(a.length<2,function(){return"Copying failed because more than one ("+i.length+") "+("save handlers for destination URL "+t+".")}),s=a[0],o=Vs(n).scheme,l=Vs(n).path,u=o===Vs(n).scheme,[4,r.load()];case 1:return c=d.sent(),e&&u?[4,gi.getManager(o).removeModel(l)]:[3,3];case 2:d.sent(),d.label=3;case 3:return[4,s.save(c)];case 4:return h=d.sent(),e&&!u?[4,gi.getManager(o).removeModel(l)]:[3,6];case 5:d.sent(),d.label=6;case 6:return[2,h.modelArtifactsInfo]}})})}function _A(){return de(this,void 0,void 0,function(){var n,t,e,i,r,a,s,o;return pe(this,function(l){switch(l.label){case 0:n=gi.getSchemes(),t={},e=0,i=n,l.label=1;case 1:return e0,function(){return"promises must be a none empty array"})}function o(l,u){E(l>=0&&l<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got startFraction "+l)}),E(u>=0&&u<=1,function(){return"Progress fraction must be in range [0, 1], but "+("got endFraction "+u)}),E(u>=l,function(){return"startFraction must be no more than endFraction, but "+("got startFraction "+l+" and endFraction ")+(""+u)})}return Promise.all(n.map(a))}function $m(n,t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return t==null&&(t={}),e=t.fetchFunc==null?Me().platform.fetch:t.fetchFunc,i=n.map(function(f){return e(f,t.requestInit,{isBinary:!0})}),r=0,a=.5,t.onProgress==null?[4,Promise.all(i)]:[3,2];case 1:return o=p.sent(),[3,4];case 2:return[4,jm(i,t.onProgress,r,a)];case 3:o=p.sent(),p.label=4;case 4:return s=o,l=s.map(function(f){return f.arrayBuffer()}),u=.5,c=1,t.onProgress==null?[4,Promise.all(l)]:[3,6];case 5:return d=p.sent(),[3,8];case 6:return[4,jm(l,t.onProgress,u,c)];case 7:d=p.sent(),p.label=8;case 8:return h=d,[2,h]}})})}function iT(n,t,e,i){return t===void 0&&(t=""),de(this,void 0,void 0,function(){var r,a;return pe(this,function(s){return r=function(o){return $m(o,{requestInit:i})},a=Xm(r),[2,a(n,t,e)]})})}function Xm(n){var t=this;return function(e,i,r){return i===void 0&&(i=""),de(t,void 0,void 0,function(){var a,s,o,l,u,c,h,d,p,f;return pe(this,function(m){switch(m.label){case 0:if(a=e.map(function(){return!1}),s={},o=r!=null?r.map(function(){return!1}):[],l=[],e.forEach(function(g,v){var b=0;g.weights.forEach(function(w){var S="quantization"in w?w.quantization.dtype:w.dtype,L=Oc[S]*st(w.shape),N=function(){a[v]=!0,s[v]==null&&(s[v]=[]),s[v].push({manifestEntry:w,groupOffset:b,sizeBytes:L})};r!=null?r.forEach(function(C,R){C===w.name&&(N(),o[R]=!0)}):N(),l.push(w.name),b+=L})}),!o.every(function(g){return g}))throw u=r.filter(function(g,v){return!o[v]}),new Error("Could not find weights in manifest with names: "+(u.join(", ")+`. +`)+"Manifest JSON has weights with names: "+(l.join(", ")+"."));return c=a.reduce(function(g,v,b){return v&&g.push(b),g},[]),h=[],c.forEach(function(g){e[g].paths.forEach(function(v){var b=i+(i.endsWith("/")?"":"/")+v;h.push(b)})}),[4,n(h)];case 1:return d=m.sent(),p={},f=0,c.forEach(function(g){for(var v=e[g].paths.length,b=0,w=0;w0,function(){return"URL path for http must not be null, undefined or empty."}),Array.isArray(t)&&E(t.length===2,function(){return"URL paths for http must have a length of 2, "+("(actual length is "+t.length+").")}),this.path=t,e.requestInit!=null&&e.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{}}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){var e,i,r,a;return pe(this,function(s){switch(s.label){case 0:if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");return e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit),e.body=new FormData,i=[{paths:["./model.weights.bin"],weights:t.weightSpecs}],r={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,userDefinedMetadata:t.userDefinedMetadata,weightsManifest:i},e.body.append("model.json",new Blob([JSON.stringify(r)],{type:aT}),"model.json"),t.weightData!=null&&e.body.append("model.weights.bin",new Blob([t.weightData],{type:rT}),"model.weights.bin"),[4,this.fetch(this.path,e)];case 1:if(a=s.sent(),a.ok)return[2,{modelArtifactsInfo:Ba(t),responses:[a]}];throw new Error("BrowserHTTPRequest.save() failed due to HTTP response status "+(a.status+"."))}})})},n.prototype.load=function(){return de(this,void 0,void 0,function(){var t,e,i,r,a,s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return[4,this.fetch(this.path,this.requestInit)];case 1:if(t=g.sent(),!t.ok)throw new Error("Request to "+this.path+" failed with status code "+(t.status+". Please verify this URL points to ")+"the model JSON of the model to load.");g.label=2;case 2:return g.trys.push([2,4,,5]),[4,t.json()];case 3:return e=g.sent(),[3,5];case 4:throw i=g.sent(),r="Failed to parse model JSON of response from "+this.path+".",this.path.endsWith(".pb")?r+=" Your path contains a .pb file extension. Support for .pb models have been removed in TensorFlow.js 1.0 in favor of .json models. You can re-convert your Python TensorFlow model using the TensorFlow.js 1.0 conversion scripts or you can convert your.pb models with the 'pb2json'NPM script in the tensorflow/tfjs-converter repository.":r+=" Please make sure the server is serving valid JSON for this request.",new Error(r);case 5:if(a=e.modelTopology,s=e.weightsManifest,o=e.generatedBy,l=e.convertedBy,u=e.format,c=e.userDefinedMetadata,a==null&&s==null)throw new Error("The JSON from HTTP path "+this.path+" contains neither model topology or manifest for weights.");return s!=null?[4,this.loadWeights(s)]:[3,7];case 6:p=g.sent(),h=p[0],d=p[1],g.label=7;case 7:return f={modelTopology:a,weightSpecs:h,weightData:d,userDefinedMetadata:c,generatedBy:o,convertedBy:l,format:u},m=e.modelInitializer,m&&(f.modelInitializer=m),[2,f]}})})},n.prototype.loadWeights=function(t){return de(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N;return pe(this,function(C){switch(C.label){case 0:for(e=Array.isArray(this.path)?this.path[1]:this.path,i=sT(e),r=i[0],a=i[1],s=this.weightPathPrefix||r,o=[],l=0,u=t;lt?n.substring(e):"";return[i+"/",r]}function zc(n){return n.match(Jm.URL_SCHEME_REGEX)!=null}var Zm=function(n,t){if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;var e=!0;return Array.isArray(n)?e=n.every(function(i){return zc(i)}):e=zc(n),e?Pc(n,t):null};en.registerSaveRouter(Zm);en.registerLoadRouter(Zm);function Pc(n,t){return new Jm(n,t)}function oT(n,t){return Pc(n,t)}var _c=function(){function n(t){this.modelArtifacts=t}return n.prototype.load=function(){return de(this,void 0,void 0,function(){return pe(this,function(t){return[2,this.modelArtifacts]})})},n}(),lT=function(){function n(t){this.saveHandler=t}return n.prototype.save=function(t){return de(this,void 0,void 0,function(){return pe(this,function(e){return[2,this.saveHandler(t)]})})},n}();function uT(n,t,e,i){if(arguments.length===1){var r=n.modelTopology!=null||n.weightSpecs!=null;return r?new _c(n):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new _c({modelTopology:n}))}else return console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new _c({modelTopology:n,weightSpecs:t,weightData:e,trainingConfig:i})}function cT(n){return new lT(n)}var hT={__proto__:null,browserFiles:nT,browserHTTPRequest:oT,concatenateArrayBuffers:Dc,decodeWeights:Bm,encodeWeights:yA,fromMemory:uT,getLoadHandlers:CA,getModelArtifactsInfoForJSON:Ba,getSaveHandlers:xA,http:Pc,isHTTPScheme:zc,loadWeights:iT,registerLoadRouter:NA,registerSaveRouter:TA,weightsLoaderFactory:Xm,withSaveHandler:cT,copyModel:HA,listModels:_A,moveModel:VA,removeModel:MA};function dT(n,t){var e=O(n,"x","reshape",null),i={x:e},r={shape:t},a=function(s,o){return t=gm(t,e.size),E(e.size===st(t),function(){return"new shape and old shape must have the same number of elements."}),o([e]),s.reshape(e,t)};return z.runKernelFunc(a,i,null,Uu,r)}var V=U({reshape_:dT});function pT(n,t,e,i){var r;e===void 0&&(e=!1),i===void 0&&(i=!1);var a=O(n,"a","matMul"),s=O(t,"b","matMul");r=rt(a,s),a=r[0],s=r[1],E(a.rank>=2&&s.rank>=2&&a.rank===s.rank,function(){return"Error in matMul: inputs must have the same rank of at least 2, "+("got ranks "+a.rank+" and "+s.rank+".")});var o=e?a.shape[a.rank-2]:a.shape[a.rank-1],l=i?s.shape[s.rank-1]:s.shape[s.rank-2],u=e?a.shape[a.rank-1]:a.shape[a.rank-2],c=i?s.shape[s.rank-2]:s.shape[s.rank-1],h=a.shape.slice(0,-2),d=s.shape.slice(0,-2),p=st(h),f=st(d);E(un(h,d),function(){return"Error in matMul: outer dimensions ("+h+") and ("+(d+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" must match.")}),E(o===l,function(){return"Error in matMul: inner shapes ("+o+") and ("+(l+") of Tensors with shapes "+a.shape+" and ")+(s.shape+" and transposeA="+e)+(" and transposeB="+i+" must match.")});var m=a.shape.slice(0,-2).concat([u,c]),g=e?V(a,[p,o,u]):V(a,[p,u,o]),v=i?V(s,[f,c,l]):V(s,[f,l,c]),b=function(N,C){return C([g,v]),N.batchMatMul(g,v,e,i)},w={a:g,b:v},S={transposeA:e,transposeB:i},L=z.runKernelFunc(b,w,null,Hl,S);return V(L,m)}var Fe=U({matMul_:pT});function fT(n,t,e,i){if(e===void 0&&(e=1),i===void 0&&(i=0),t<2)throw new Error("Error in oneHot: depth must be >=2, but it is "+t);var r=O(n,"indices","oneHot","int32"),a=r.shape.concat([t]),s=function(u,c){return c([r]),V(u.oneHot(V(r,[r.size]),t,e,i),a)},o={indices:r},l={depth:t,onValue:e,offValue:i};return z.runKernelFunc(s,o,null,Ou,l)}var qs=U({oneHot_:fT});function mT(n,t){var e=O(n,"x","transpose");if(t==null&&(t=e.shape.map(function(a,s){return s}).reverse()),E(e.rank===t.length,function(){return"Error in transpose: rank of input "+e.rank+" "+("must match length of perm "+t+".")}),t.forEach(function(a){E(a>=0&&a0&&Number.isInteger(e),function(){return"If provided, numClasses must be a positive integer, "+("but got "+e)}),E(i.rank===1,function(){return"Expected the rank of labels to be 1, but got "+i.rank}),E(r.rank===1,function(){return"Expected the rank of predictions to be 1, "+("but got "+r.rank)}),E(i.shape[0]===r.shape[0],function(){return"Mismatch in the number of examples: "+(i.shape[0]+" vs. "+r.shape[0]+". ")+"Labels and predictions should have the same number of elements."}),E(e>0&&Number.isInteger(e),function(){return"numClasses is required to be a positive integer, but got "+(""+e)});var a=qs(ue(i,"int32"),e),s=qs(ue(r,"int32"),e),o=lt(a);return ue(Fe(o,s),"int32")}var vT=U({confusionMatrix_:gT});var yT={__proto__:null,confusionMatrix:vT};function Qm(n,t,e){if(Hi(n),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");var i=En(n,e);if(i.length!==3&&i.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return pi(n,t,i,e)}var Pr;function bT(n,t){if(t===void 0&&(t=3),t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(n==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");var e=!1,i=!1,r=!1,a=!1,s=!1;if(n.data instanceof Uint8Array)e=!0;else if(typeof ImageData!="undefined"&&n instanceof ImageData)i=!0;else if(typeof HTMLVideoElement!="undefined"&&n instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&n instanceof HTMLImageElement)a=!0;else if(n.getContext!=null)s=!0;else throw new Error("pixels passed to tf.browser.fromPixels() must be either an HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData in browser, or OffscreenCanvas, ImageData in webworker or {data: Uint32Array, width: number, height: number}, "+("but was "+n.constructor.name));if(r){var o=2;if(r&&n.readyState element.")}var l=vc(dc,z.backendName);if(l!=null){var u={pixels:n},c={numChannels:t};return z.runKernel(dc,u,c)}var h=r?[n.videoWidth,n.videoHeight]:[n.width,n.height],d=h[0],p=h[1],f;s?f=n.getContext("2d").getImageData(0,0,d,p).data:i||e?f=n.data:(a||r)&&(Pr==null&&(Pr=document.createElement("canvas").getContext("2d")),Pr.canvas.width=d,Pr.canvas.height=p,Pr.drawImage(n,0,0,d,p),f=Pr.getImageData(0,0,d,p).data);var m;if(t===4)m=new Int32Array(f);else{var g=d*p;m=new Int32Array(g*t);for(var v=0;v4||o===2)throw new Error("toPixels only supports depth of size "+("1, 3 or 4 but got "+o));if(e.dtype!=="float32"&&e.dtype!=="int32")throw new Error("Unsupported type for toPixels: "+e.dtype+". Please use float32 or int32 tensors.");return[4,e.data()];case 1:for(l=b.sent(),u=e.dtype==="float32"?255:1,c=new Uint8ClampedArray(s*a*4),h=0;h1)throw new Error("Tensor values for a float32 Tensor must be in the "+("range [0 - 1] but encountered "+f+"."))}else if(e.dtype==="int32"&&(f<0||f>255))throw new Error("Tensor values for a int32 Tensor must be in the "+("range [0 - 255] but encountered "+f+"."));o===1?(d[0]=f*u,d[1]=f*u,d[2]=f*u):d[p]=f*u}m=h*4,c[m+0]=Math.round(d[0]),c[m+1]=Math.round(d[1]),c[m+2]=Math.round(d[2]),c[m+3]=Math.round(d[3])}return t!=null&&(t.width=s,t.height=a,g=t.getContext("2d"),v=new ImageData(c,s,a),g.putImageData(v,0,0)),e!==n&&e.dispose(),[2,c]}})})}var ST=U({fromPixels_:bT}),LT={__proto__:null,toPixels:wT,fromPixels:ST};function eg(n,t){if(n.rank<1)throw new Error("tf.gatherND() expects the input to be rank 1 or higher,"+(" but the rank was "+n.rank+"."));if(t.rank<1)throw new Error("tf.gatherND() expects the indices to be rank 1 or higher,"+(" but the rank was "+t.rank+"."));if(t.dtype!=="int32")throw new Error("tf.gatherND() expects the indices to be int32 type,"+(" but the dtype was "+t.dtype+"."));if(t.shape[t.rank-1]>n.rank)throw new Error("index innermost dimension length must be <= tensor rank; saw: "+(t.shape[t.rank-1]+" vs. "+n.rank));if(n.size===0)throw new Error("Requested more than 0 entries, but input is empty."+(" Input shape: "+n.shape+"."));for(var e=t.shape,i=e[e.length-1],r=1,a=0;a1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,a="Must have updates.shape = indices.shape[:batchDim] + "+("shape[sliceDim:], got updates.shape: "+e.shape)+(", indices.shape: "+t.shape+", shape: "+n)+(", sliceDim: "+i+", and batchDim: "+r+".");if(e.rank1?t.shape[i-1]:1,a=e.length,s=1,o=r;o0;)n&1&&t.push(e),n/=2,e++;return t}function ig(n,t,e){for(var i=[],r=0;r0){var p=t[0],f=e+1;c=og(s,p,f,i,n),h=lg(o,p,f,r,n),d=rg(a,p,f,n)}else for(var m=0;m-1)a[o]=0;else{var l=ag(t,e,o),u=i[l];n&1<-1)a[o]=Number.MAX_SAFE_INTEGER;else{var l=ag(t,e,o),u=i[l];n&1<0?s=Number.MIN_SAFE_INTEGER:s=Number.MAX_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),s=Oa(0,s,l-1),s}function hg(n,t,e,i,r,a){var s=t[r],o=e[r]||1;(n&1<0?s=Number.MAX_SAFE_INTEGER:s=Number.MIN_SAFE_INTEGER);var l=i[r];return s<0&&(s+=l),o>0?s=Oa(0,s,l):s=Oa(-1,s,l-1),s}function TT(n,t,e){for(var i=e.length,r=0;r1){i=r;break}for(var r=i+1;r0||e[r]!==n[r])return!1;return!0}function NT(n,t){for(var e=n.length>0?n[n.length-1]:1,i=0;i=0?s:(E(s===-1,function(){return"Negative size values should be exactly -1 but got "+(s+" for the slice() size at index "+o+".")}),n.shape[o]-i[o])}),[i,a]}var pg={__proto__:null,assertParamsValid:ng,maskToAxes:Gs,computeOutShape:ig,stridesWithElidedDims:rg,getNormalizedAxes:dg,startIndicesWithElidedDims:og,stopIndicesWithElidedDims:lg,stridesForAxis:ug,startForAxis:cg,stopForAxis:hg,isSliceContinous:TT,computeFlatOffset:NT,parseSliceParams:Vc};var fg=function(){function n(){}return n.prototype.getClassName=function(){return this.constructor.className},n.fromConfig=function(t,e){return new t(e)},n}(),mg=function(){function n(){this.classNameMap={}}return n.getMap=function(){return n.instance==null&&(n.instance=new n),n.instance},n.register=function(t){n.getMap().classNameMap[t.className]=[t,t.fromConfig]},n}();function vi(n){E(n.className!=null,function(){return"Class being registered does not have the static className property defined."}),E(typeof n.className=="string",function(){return"className is required to be a string, but got type "+typeof n.className}),E(n.className.length>0,function(){return"Class being registered has an empty-string as its className, which is disallowed."}),mg.register(n)}var xT={__proto__:null,Serializable:fg,SerializationMap:mg,registerClass:vi};var CT=.001,gg=.1;function RT(n,t,e){return e==null&&(e=qc()),Gc(n,t,function(i,r){return Yc(i,r,e)})}function qc(){return z.backend.floatPrecision()===32?CT:gg}function Gc(n,t,e){var i=!0;if((Et(n)||Et(t))&&(i=!1),Et(n)&&Et(t)&&(i=!0),i){var r=n.constructor.name,a=t.constructor.name;if(r!==a)throw new Error("Arrays are of different type. Actual: "+r+". "+("Expected: "+a))}if(Array.isArray(n)&&Array.isArray(t)){var s=En(n),o=En(t);if(!un(s,o))throw new Error("Arrays have different shapes. "+("Actual: ["+s+"]. Expected: ["+o+"]"))}var l=Et(n)?n:Vi(n),u=Et(t)?t:Vi(t);if(l.length!==u.length)throw new Error("Arrays have different lengths actual: "+l.length+" vs "+("expected: "+u.length+`. +`)+("Actual: "+l+`. +`)+("Expected: "+u+"."));for(var c=0;ce)}function kT(n,t,e){for(var i=0;ie)throw new Error("Value out of range:"+n[i]+" low: "+t+", high: "+e)}function FT(n,t){expect(new Float32Array(n)).toEqual(new Float32Array(t))}var WT={__proto__:null,TEST_EPSILON_FLOAT16:gg,expectArraysClose:RT,testEpsilon:qc,expectPromiseToFail:OT,expectArraysEqual:ET,expectNumbersClose:DT,expectValuesInRange:kT,expectArrayBuffersEqual:FT};var UT="2.6.0";function BT(){Me().set("PROD",!0)}function zT(){Me().set("DEBUG",!0)}function PT(){Me().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function It(n){Me().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(n+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}function _T(){z.disposeVariables()}function MT(){return z}function HT(){return z.memory()}function VT(n){return z.profile(n)}function ft(n,t){return z.tidy(n,t)}function Pt(n){var t=Rc(n);t.forEach(function(e){return e.dispose()})}function vg(n){return z.keep(n)}function qT(n){return z.time(n)}function GT(n){return z.setBackend(n)}function YT(){return z.ready()}function KT(){return z.backendName}function jT(n){z.removeBackend(n)}function $T(n){return z.findBackend(n)}function XT(n){return z.findBackendFactory(n)}function JT(n,t,e){return e===void 0&&(e=1),z.registerBackend(n,t,e)}function ZT(){return z.backend}function QT(n,t){Me().setPlatform(n,t)}function e1(n,t){var e,i=O(n,"a","add"),r=O(t,"b","add");e=rt(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.add(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Ds)}var fe=U({add_:e1});function t1(n,t){var e,i=O(n,"a","floorDiv"),r=O(t,"b","floorDiv");e=rt(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.floorDiv(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,lu)}var Kc=U({floorDiv_:t1});function n1(n,t){var e,i=O(n,"a","div"),r=O(t,"b","div");if(e=rt(i,r),i=e[0],r=e[1],i.dtype==="int32"&&r.dtype==="int32")return Kc(i,r);var a=function(l,u){var c=l.realDivide(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,nu,o)}var Le=U({div_:n1});function i1(n,t){var e,i=O(n,"a","mul"),r=O(t,"b","mul");e=rt(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.multiply(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,xu)}var J=U({mul_:i1});function r1(n){var t=O(n,"x","abs"),e={x:t};return z.runKernelFunc(function(i,r){return r([t]),t.dtype==="complex64"?i.complexAbs(t):i.abs(t)},e,null,Rl)}var Gt=U({abs_:r1});function a1(n){var t=O(n,"x","acos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acos(t);return r([t]),a},e,null,Ol)}var yg=U({acos_:a1});function s1(n){var t=O(n,"x","acosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.acosh(t);return r([t]),a},e,null,El)}var bg=U({acosh_:s1});function o1(n){E(Array.isArray(n),function(){return"The argument passed to tf.addN() must be a list of tensors"}),E(n.length>=1,function(){return"Must pass at least one tensor to tf.addN(), but got "+(""+n.length)});var t=n.map(function(a,s){return O(a,"tensors"+s,"addN")}),e=t[0];t.forEach(function(a){if(a.dtype!==e.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(function(a){if(!un(a.shape,e.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});var i=function(a,s){var o=a.addN(t);return s(t),o},r=t;return z.runKernelFunc(i,r,null,Dl)}var l1=U({addN_:o1});function jc(n,t){for(var e=0;e=0&&t=1,function(){return"Pass at least one tensor to concat"});var e=Ua(n,"tensors","concat");e[0].dtype==="complex64"&&e.forEach(function(s){if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor + with dtype `+s.dtype+". ")});var i=function(s,o){var l=Ze(t,e[0].shape)[0],u=Dg(e.map(function(d){return d.shape}),l);if(st(u)===0)return fi([],u);if(e=e.filter(function(d){return d.size>0}),e.length===1)return e[0];var c=e.map(function(d){return d.shape});Eg(c,l);var h=s.concat(e,l);return o(e),h},r=e,a={axis:t};return z.runKernelFunc(i,r,null,Kl,a)}var Ct=U({concat_:x1});function C1(n){var t=O(n,"x","sigmoid"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sigmoid(t);return r([a]),a},e,null,$u)}var Ki=U({sigmoid_:C1});function R1(n,t,e){var i=O(n,"x","slice");if(i.rank===0)throw new Error("Slicing scalar is not possible");var r=function(o,l){var u=Vc(i,t,e),c=u[0],h=u[1];return ng(i,c,h),l([i]),o.slice(i,c,h)},a={x:i},s={begin:t,size:e};return z.runKernelFunc(r,a,null,Gu,s)}var ze=U({slice_:R1});function O1(n){var t=O(n,"x","tanh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tanh(t);return r([a]),a},e,null,ac)}var js=U({tanh_:O1});function E1(n,t,e,i,r,a){var s=O(n,"forgetBias","basicLSTMCell"),o=O(t,"lstmKernel","basicLSTMCell"),l=O(e,"lstmBias","basicLSTMCell"),u=O(i,"data","basicLSTMCell"),c=O(r,"c","basicLSTMCell"),h=O(a,"h","basicLSTMCell"),d=Ct([u,h],1),p=Fe(d,o),f=fe(p,l),m=f.shape[0],g=f.shape[1]/4,v=[m,g],b=ze(f,[0,0],v),w=ze(f,[0,g],v),S=ze(f,[0,g*2],v),L=ze(f,[0,g*3],v),N=fe(J(Ki(b),js(w)),J(c,Ki(fe(s,S)))),C=J(js(N),Ki(L));return[N,C]}var D1=U({basicLSTMCell_:E1});function k1(n,t,e){var i=O(n,"x","batchToSpaceND"),r=t.reduce(function(l,u){return l*u});E(i.rank>=1+t.length,function(){return"input rank is "+i.rank+" but should be > than blockShape.length "+t.length}),E(e.length===t.length,function(){return"crops.length is "+e.length+" but should be equal to blockShape.length "+t.length}),E(i.shape[0]%r===0,function(){return"input tensor batch is "+i.shape[0]+" but is not divisible by the product of "+("the elements of blockShape "+t.join(" * ")+" === "+r)});var a=function(l){return l.batchToSpaceND(i,t,e)},s={x:i},o={blockShape:t,crops:e};return z.runKernelFunc(a,s,null,Vl,o)}var $s=U({batchToSpaceND_:k1});function F1(n){var t;return n.rank===0||n.rank===1?t=V(n,[1,1,1,n.size]):n.rank===2?t=V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?t=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):t=n,t}function W1(n,t,e,i,r,a){a==null&&(a=.001);var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;i!=null&&(c=O(i,"offset","batchNorm")),E(o.rank===l.rank,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),E(c==null||o.rank===c.rank,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),E(u==null||o.rank===u.rank,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."});var h=F1(s),d=function(g,v){return v([h,o,l,u]),g.batchNorm(h,Xs(o),Xs(l),Xs(c),Xs(u),a)},p={x:h,scale:u,offset:c,mean:o,variance:l},f={varianceEpsilon:a},m=z.runKernelFunc(d,p,null,uu,f);return V(m,s.shape)}function Xs(n){return n==null?null:n.rank===0?V(n,[n.size]):n.rank===1?n:n.rank===2?V(n,[1,1,n.shape[0],n.shape[1]]):n.rank===3?V(n,[1,n.shape[0],n.shape[1],n.shape[2]]):n}var Ma=U({batchNorm_:W1});function U1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===2,function(){return"Error in batchNorm2D: x must be rank 2 but got rank "+(s.rank+".")}),E(o.rank===2||o.rank===1,function(){return"Error in batchNorm2D: mean must be rank 2 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===2||l.rank===1,function(){return"Error in batchNorm2D: variance must be rank 2 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===2||u.rank===1,function(){return"Error in batchNorm2D: scale must be rank 2 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===2||c.rank===1,function(){return"Error in batchNorm2D: offset must be rank 2 or rank 1 "+("but got rank "+c.rank+".")}),Ma(s,o,l,c,u,a)}var B1=U({batchNorm2d_:U1});function z1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===3,function(){return"Error in batchNorm3D: x must be rank 3 but got rank "+(s.rank+".")}),E(o.rank===3||o.rank===1,function(){return"Error in batchNorm3D: mean must be rank 3 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===3||l.rank===1,function(){return"Error in batchNorm3D: variance must be rank 3 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===3||u.rank===1,function(){return"Error in batchNorm3D: scale must be rank 3 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===3||c.rank===1,function(){return"Error in batchNorm3D: offset must be rank 3 or rank 1 "+("but got rank "+c.rank+".")}),Ma(s,o,l,c,u,a)}var P1=U({batchNorm3d_:z1});function _1(n,t,e,i,r,a){var s=O(n,"x","batchNorm"),o=O(t,"mean","batchNorm"),l=O(e,"variance","batchNorm"),u;r!=null&&(u=O(r,"scale","batchNorm"));var c;return i!=null&&(c=O(i,"offset","batchNorm")),E(s.rank===4,function(){return"Error in batchNorm4D: x must be rank 4 but got rank "+(s.rank+".")}),E(o.rank===4||o.rank===1,function(){return"Error in batchNorm4D: mean must be rank 4 or rank 1 but "+("got rank "+o.rank+".")}),E(l.rank===4||l.rank===1,function(){return"Error in batchNorm4D: variance must be rank 4 or rank 1 "+("but got rank "+l.rank+".")}),u!=null&&E(u.rank===4||u.rank===1,function(){return"Error in batchNorm4D: scale must be rank 4 or rank 1 "+("but got rank "+u.rank+".")}),c!=null&&E(c.rank===4||c.rank===1,function(){return"Error in batchNorm4D: offset must be rank 4 or rank 1 "+("but got rank "+c.rank+".")}),Ma(s,o,l,c,u,a)}var M1=U({batchNorm4d_:_1});function H1(n,t){var e=O(n,"broadcastTo","x"),i=e.shape;if(t.some(function(d){return!(d>0)||d%1!==0}))throw new Error("broadcastTo(): Invalid broadcast shape ["+t+"].");if(t.lengthe.rank){for(var r=e.shape.slice();r.length=0;o--)if(a[o]===t[o])s[o]=1;else if(e.shape[o]!==1)throw new Error("broadcastTo(): ["+i+"] cannot be broadcast to ["+t+"].");var l=s.map(function(d,p){return d>1?p:-1}).filter(function(d){return d>=0});if(l.length===0)return Gi(e);var u=function(d){return d.tile(e,s)},c={x:e},h={shape:t,inputShape:a};return z.runKernelFunc(u,c,null,ql,h)}var Js=U({broadcastTo_:H1});function V1(n){var t=O(n,"x","ceil"),e={x:t};return z.runKernelFunc(function(i){return i.ceil(t)},e,null,Gl)}var kg=U({ceil_:V1});function q1(n,t,e){var i=O(n,"x","clipByValue");E(t<=e,function(){return"Error in clip: min ("+t+") must be "+("less than or equal to max ("+e+").")});var r={x:i},a={clipValueMin:t,clipValueMax:e};return z.runKernelFunc(function(s,o){var l=s.clip(i,t,e);return o([i]),l},r,null,Yl,a)}var Fg=U({clipByValue_:q1});function G1(n){return Ct(n,0)}var Y1=U({concat1d_:G1});function K1(n,t){return Ct(n,t)}var j1=U({concat2d_:K1});function $1(n,t){return Ct(n,t)}var X1=U({concat3d_:$1});function J1(n,t){return Ct(n,t)}var Z1=U({concat4d_:J1});function Q1(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","conv2d"),l=O(t,"filter","conv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in conv2d: input must be rank 4, but got rank "+u.rank+"."}),E(l.rank===4,function(){return"Error in conv2d: filter must be rank 4, but got rank "+(l.rank+".")}),s!=null&&E(it(i),function(){return"Error in conv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=r==="NHWC"?u.shape[3]:u.shape[1];E(h===l.shape[2],function(){return"Error in conv2d: depth of input ("+h+") must match "+("input depth for filter "+l.shape[2]+".")}),E(_t(e,a),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")});var d=function(g,v){var b=za(r),w=Fn(u.shape,l.shape,e,a,i,s,!1,b),S=g.conv2d(u,l,w);return v([u,l]),S},p={x:u,filter:l},f={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,jl,f);return c?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Hr=U({conv2d_:Q1});function eN(n,t,e,i,r,a,s){r===void 0&&(r="NWC"),a===void 0&&(a=1);var o=O(n,"x","conv1d"),l=O(t,"filter","conv1d"),u=o,c=!1;o.rank===2&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1]])),E(u.rank===3,function(){return"Error in conv1d: input must be rank 3, but got rank "+u.rank+"."}),E(l.rank===3,function(){return"Error in conv1d: filter must be rank 3, but got rank "+(l.rank+".")}),s!=null&&E(it(i),function(){return"Error in conv1d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")}),E(u.shape[2]===l.shape[1],function(){return"Error in conv1d: depth of input ("+u.shape[2]+") must match "+("input depth for filter "+l.shape[1]+".")}),E(_t(e,a),function(){return"Error in conv1D: Either stride or dilation must be 1. "+("Got stride "+e+" and dilation '"+a+"'")}),E(r==="NWC",function(){return"Error in conv1d: got dataFormat of "+r+" but only NWC is currently supported."});var h=V(l,[1,l.shape[0],l.shape[1],l.shape[2]]),d=V(u,[u.shape[0],1,u.shape[1],u.shape[2]]),p=[1,e],f=[1,a],m="NHWC",g=Hr(d,h,p,i,m,f,s);return c?V(g,[g.shape[2],g.shape[3]]):V(g,[g.shape[0],g.shape[2],g.shape[3]])}var Wg=U({conv1d_:eN});function tN(n,t,e,i,r,a,s){a===void 0&&(a="NHWC"),E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var o=n,l=t,u=!1;t.rank===3&&(u=!0,l=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]),o=[1,n[0],n[1],n[2]]),E(o.length===4,function(){return"Error in conv2dDerInput: inShape must be length 4, but got length "+(o.length+".")}),E(l.rank===4,function(){return"Error in conv2dDerInput: dy must be rank 4, but got "+("rank "+l.rank)}),E(e.rank===4,function(){return"Error in conv2dDerInput: filter must be rank 4, but got "+("rank "+e.rank)});var c=a==="NHWC"?o[3]:o[1],h=a==="NHWC"?l.shape[3]:l.shape[1];E(c===e.shape[2],function(){return"Error in conv2dDerInput: depth of input ("+c+") must "+("match input depth for filter "+e.shape[2]+".")}),E(h===e.shape[3],function(){return"Error in conv2dDerInput: depth of output ("+h+") must "+("match output depth for filter "+e.shape[3]+".")}),s!=null&&E(it(r),function(){return"Error in conv2dDerInput: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+r+".")});var d=function(g,v){var b=1,w=za(a),S=Fn(o,e.shape,i,b,r,s,!1,w),L=g.conv2dDerInput(l,e,S);return v([l,e]),L},p={dy:l,filter:e},f={strides:i,pad:r,dataFormat:a,dimRoundingMode:s,inputShape:o},m=z.runKernelFunc(d,p,null,$l,f);return u?V(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Zc=U({conv2DBackpropInput_:tN});function nN(n,t,e,i,r,a){var s=O(n,"x","conv2dTranspose"),o=O(t,"filter","conv2dTranspose");return Zc(e,s,o,i,r,"NHWC",a)}var Ug=U({conv2dTranspose_:nN});function iN(n,t,e,i,r,a){r===void 0&&(r="NDHWC"),a===void 0&&(a=[1,1,1]);var s=O(n,"x","conv3d"),o=O(t,"filter","conv3d"),l=s,u=!1;s.rank===4&&(u=!0,l=V(s,[1,s.shape[0],s.shape[1],s.shape[2],s.shape[3]])),E(l.rank===5,function(){return"Error in conv3d: input must be rank 5, but got rank "+l.rank+"."}),E(o.rank===5,function(){return"Error in conv3d: filter must be rank 5, but got rank "+(o.rank+".")}),E(l.shape[4]===o.shape[3],function(){return"Error in conv3d: depth of input ("+l.shape[4]+") must match "+("input depth for filter "+o.shape[3]+".")}),E(_t(e,a),function(){return"Error in conv3D: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+a+"'")}),E(r==="NDHWC",function(){return"Error in conv3d: got dataFormat of "+r+" but only NDHWC is currently supported."});var c=function(f,m){var g=Pa(l.shape,o.shape,e,a,i),v=f.conv3d(l,o,g);return m([l,o]),v},h={x:l,filter:o},d={strides:e,pad:i,dataFormat:r,dilations:a},p=z.runKernelFunc(c,h,null,Xl,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var rN=U({conv3d_:iN});function aN(n,t,e,i,r){E(n.length===t.rank,function(){return"Length of inShape "+("("+n.length+") and rank of dy ("+t.rank+") must match")});var a=n,s=t,o=!1;t.rank===4&&(o=!0,s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),a=[1,n[0],n[1],n[2],n[3]]);var l=a[4],u=s.shape[4];E(a.length===5,function(){return"Error in conv3dDerInput: inShape must be length 5, but got length "+(a.length+".")}),E(s.rank===5,function(){return"Error in conv3dDerInput: dy must be rank 5, but got "+("rank "+s.rank)}),E(e.rank===5,function(){return"Error in conv3dDerInput: filter must be rank 5, but got "+("rank "+e.rank)}),E(l===e.shape[3],function(){return"Error in conv3dDerInput: depth of input ("+l+") must "+("match input depth for filter "+e.shape[3]+".")}),E(u===e.shape[4],function(){return"Error in conv3dDerInput: depth of output ("+u+") must "+("match output depth for filter "+e.shape[4]+".")});var c=function(f){var m=1,g=Pa(a,e.shape,i,m,r);return f.conv3dDerInput(s,e,g)},h={dy:s},d={pad:r},p=z.runKernelFunc(c,h,null,Af,d);return o?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var Bg=U({conv3DBackpropInput_:aN});function sN(n,t,e,i,r){var a=O(n,"x","conv3dTranspose"),s=O(t,"filter","conv3dTranspose");return Bg(e,a,s,i,r)}var oN=U({conv3dTranspose_:sN});function lN(n){var t=O(n,"x","cos"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cos(t);return r([t]),a},e,null,Jl)}var Zs=U({cos_:lN});function uN(n){var t=O(n,"x","cosh"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.cosh(t);return r([t]),a},e,null,Zl)}var Qc=U({cosh_:uN});function cN(n,t,e,i){t===void 0&&(t=0),e===void 0&&(e=!1),i===void 0&&(i=!1);var r=O(n,"x","cumsum"),a=function(l,u){var c=nn([t],r.rank),h=r;c!=null&&(h=lt(r,c));var d=kn(1,r.rank)[0],p=l.cumsum(h,d,e,i);if(u([r]),c!=null){var f=Ys(c);p=lt(p,f)}return p},s={x:r},o={axis:t,exclusive:e,reverse:i};return z.runKernelFunc(a,s,null,Ql,o)}var eh=U({cumsum_:cN});function hN(n,t,e){e===void 0&&(e="NHWC");var i=O(n,"x","depthToSpace"),r=e==="NHWC"?i.shape[1]:i.shape[2],a=e==="NHWC"?i.shape[2]:i.shape[3],s=e==="NHWC"?i.shape[3]:i.shape[1];E(r*t>=0,function(){return`Negative dimension size caused by overflow when multiplying + `+r+" and "+t+` for depthToSpace with input shape + `+i.shape}),E(a*t>=0,function(){return`Negative dimension size caused by overflow when multiplying + `+a+" and "+t+` for depthToSpace with input shape + `+i.shape}),E(s%(t*t)===0,function(){return"Dimension size must be evenly divisible by "+t*t+" but is "+s+" for depthToSpace with input shape "+i.shape});var o=function(c){return c.depthToSpace(i,t,e)},l={x:i},u={blockSize:t,dataFormat:e};return z.runKernelFunc(o,l,null,Nf,u)}var zg=U({depthToSpace_:hN});function dN(n,t,e,i,r,a,s){r===void 0&&(r="NHWC"),a===void 0&&(a=[1,1]);var o=O(n,"x","depthwiseConv2d"),l=O(t,"filter","depthwiseConv2d"),u=o,c=!1;o.rank===3&&(c=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2]])),E(u.rank===4,function(){return"Error in depthwiseConv2d: input must be rank 4, but got "+("rank "+u.rank+".")}),E(l.rank===4,function(){return"Error in depthwiseConv2d: filter must be rank 4, but got rank "+(l.rank+".")}),E(u.shape[3]===l.shape[2],function(){return"Error in depthwiseConv2d: number of input channels "+("("+u.shape[3]+") must match the inChannels dimension in ")+("filter "+l.shape[2]+".")}),s!=null&&E(it(i),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+s+" but got pad "+i+".")});var h=function(m,g){a==null&&(a=[1,1]),E(_t(e,a),function(){return"Error in depthwiseConv2d: Either strides or dilations must be "+("1. Got strides "+e+" and dilations '"+a+"'")});var v=Fn(u.shape,l.shape,e,a,i,s,!0),b=m.depthwiseConv2D(u,l,v);return g([u,l]),b},d={x:u,filter:l},p={strides:e,pad:i,dataFormat:r,dilations:a,dimRoundingMode:s},f=z.runKernelFunc(h,d,null,eu,p);return c?V(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var Ha=U({depthwiseConv2d_:dN});function pN(n){var t=O(n,"x","diag"),e=function(r){var a=V(t,[t.size]),s=r.diag(a),o=n.shape.concat(n.shape);return V(s,o)},i={x:t};return z.runKernelFunc(e,i,null,Rf)}var fN=U({diag_:pN});function mN(n,t,e,i,r,a){r===void 0&&(r=[1,1]),a===void 0&&(a="NHWC");var s=O(n,"x","dilation2d"),o=O(t,"filter","dilation2d");E(s.rank===3||s.rank===4,function(){return"Error in dilation2d: input must be rank 3 or 4, but got rank "+(s.rank+".")}),E(o.rank===3,function(){return"Error in dilation2d: filter must be rank 3, but got rank "+(o.rank+".")}),E(a==="NHWC",function(){return"Error in dilation2d: Only NHWC is currently supported, "+("but got dataFormat of "+a)});var l=s,u=!1;s.rank===3&&(l=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),u=!0);var c={x:l,filter:o},h={strides:e,pad:i,dilations:r},d=z.runKernel(tu,c,h);return u?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Pg=U({dilation2d_:mN});function gN(n,t){for(var e=n.length,i=[],r=0;r1&&s===1&&i.unshift(a)}return i}function mt(n,t){for(var e=[],i=0;i1)&&e.unshift(a)}return e}function Qe(n,t){for(var e=[],i=Math.max(n.length,t.length),r=0;rt||i===n?e=!0:i=Bs(n,i+1);return i}function DN(n,t,e){for(var i=[],r=n.length,a=0;a0,function(){return"variableGrads() expects at least one of the input variables to "+("be trainable, but none of the "+a+" variables is ")+"trainable."});var s=!0,o=z.gradients(n,t,null,s),l=o.value,u=o.grads;E(u.some(function(h){return h!=null}),function(){return"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."}),E(l.rank===0,function(){return"The f passed in variableGrads(f) must return a scalar, but it "+("returned a rank-"+l.rank+" tensor")});var c={};return t.forEach(function(h,d){u[d]!=null&&(c[h.name]=u[d])}),r!=null&&r.forEach(function(h){return c[h.name]=null}),{value:l,grads:c}}function Wn(n){return z.customGrad(n)}function io(n){var t=n.filter(function(e){return e==null}).length;if(t>0)throw new Error(`Cannot compute gradient of y=f(x) with respect to x. Make sure that + the f you passed encloses all operations that lead from x to y.`)}function QN(n){var t=O(n,"x","neg"),e={x:t};return z.runKernelFunc(function(i){return i.neg(t)},e,null,Cu)}var gt=U({neg_:QN});function ex(n){var t=O(n,"x","softplus"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.softplus(t);return r([t]),a},e,null,Xu)}var ah=U({softplus_:ex});function tx(n){var t=O(n,"x","logSigmoid"),e=Wn(function(i){var r=gt(ah(gt(i))),a=function(s){var o=J(s,Ki(gt(i)));return o};return{value:r,gradFunc:a}});return e(t)}var Zg=U({logSigmoid_:tx});function nx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","max"),r=function(o,l){var u=Ze(t,i.shape),c=u,h=nn(c,i.rank),d=i;h!=null&&(d=lt(i,h),c=kn(c.length,d.rank));var p=o.max(d,c);h!=null&&d.dispose();var f=p;if(e){var m=tn(f.shape,Ze(t,i.shape));f=V(f,m),p.dispose()}return l([i,f]),f},a={x:i},s={reductionIndices:t,keepDims:e};return z.runKernelFunc(r,a,null,wu,s)}var Ji=U({max_:nx});function ix(n,t){var e,i=O(n,"a","sub"),r=O(t,"b","sub");e=rt(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.subtract(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,ic)}var ge=U({sub_:ix});function rx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","sum");i.dtype==="bool"&&(i=ue(i,"int32"));var r=function(o,l){l([i]);var u=Ze(t,i.shape),c=nn(u,i.rank),h=u,d=i;c!=null&&(d=lt(i,c),h=kn(h.length,i.rank));var p=o.sum(d,h);if(e){var f=tn(p.shape,u);p=V(p,f)}return p},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Zu,s)}var Ie=U({sum_:rx});function ax(n,t){t===void 0&&(t=-1);var e=O(n,"logits","logSoftmax");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error("Log Softmax along a non-last dimension is not yet supported. "+("Logits was rank "+e.rank+" and axis was "+t));var i=function(s,o){var l=!0,u=Ji(n,t,!0),c=ge(n,u),h=ge(ue(c,"float32"),Xi(Ie(gn(c),t,l)));return o([h]),h},r={logits:e},a={axis:t};return z.runKernelFunc(i,r,null,yu,a)}var Qg=U({logSoftmax_:ax});function sx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","logSumExp"),r=Ze(t,i.shape),a=Ji(i,r,!0),s=ge(i,a),o=gn(s),l=Ie(o,r),u=Xi(l),c=fe(V(a,u.shape),u);if(e){var h=tn(c.shape,r);return V(c,h)}return c}var sh=U({logSumExp_:sx});function ox(n,t){var e=O(n,"a","logicalAnd","bool"),i=O(t,"b","logicalAnd","bool");Qe(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalAnd(e,i)},r,null,qf)}var Zi=U({logicalAnd_:ox});function lx(n){var t=O(n,"x","logicalNot","bool"),e={x:t};return z.runKernelFunc(function(i){return i.logicalNot(t)},e,null,Gf)}var ro=U({logicalNot_:lx});function ux(n,t){var e=O(n,"a","logicalOr","bool"),i=O(t,"b","logicalOr","bool");Qe(e.shape,i.shape);var r={a:e,b:i};return z.runKernelFunc(function(a){return a.logicalOr(e,i)},r,null,Yf)}var oh=U({logicalOr_:ux});function cx(n,t){var e=O(n,"a","logicalXor","bool"),i=O(t,"b","logicalXor","bool");return Qe(e.shape,i.shape),Zi(oh(n,t),ro(Zi(n,t)))}var ev=U({logicalXor_:cx});function hx(n,t,e,i,r){var a=O(n,"x","maxPool"),s=1,o=a,l=!1;a.rank===3&&(l=!0,o=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(o.rank===4,function(){return"Error in maxPool: input must be rank 4 but got rank "+o.rank+"."}),E(_t(e,s),function(){return"Error in maxPool: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(it(i),function(){return"Error in maxPool: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var u=function(p,f){var m=_r(o.shape,t,e,1,i,r),g;return m.filterWidth===1&&m.filterHeight===1&&un(m.inShape,m.outShape)?g=o.clone():g=p.maxPool(o,m),f([o,g]),g},c={x:o},h={filterSize:t,strides:e,pad:i,dimRoundingMode:r},d=z.runKernelFunc(u,c,null,Lu,h);return l?V(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var lh=U({maxPool_:hx});function dx(n,t,e,i,r,a,s){t===void 0&&(t=[1,1,1]),a===void 0&&(a="NDHWC"),s==null?s=[1,1,1]:It("dilations is deprecated, this field will be gone in v3.0.0.");var o=O(n,"x","maxPool3d"),l=o,u=!1;o.rank===4&&(u=!0,l=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),E(l.rank===5,function(){return"Error in maxPool3d: x must be rank 5 but got rank "+l.rank+"."}),E(a==="NDHWC",function(){return"Error in maxPool3d: Only NDHWC is currently supported, "+("but got dataFormat of "+a)}),E(_t(e,s),function(){return"Error in maxPool3d: Either strides or dilations must be 1. "+("Got strides "+e+" and dilations '"+s+"'")}),r!=null&&E(it(i),function(){return"Error in maxPool3d: pad must be an integer when using, "+("dimRoundingMode "+r+" but got pad "+i+".")});var c=function(f,m){s==null&&(s=[1,1,1]);var g=_a(l.shape,t,e,s,i,r,a),v=f.maxPool3d(l,g);return m([l,v]),v},h={x:l},d={filterSize:t,strides:e,pad:i,dimRoundingMode:r,dataFormat:a,dilations:s},p=z.runKernelFunc(c,h,null,Iu,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var px=U({maxPool3d_:dx});function fx(n,t,e,i,r){r===void 0&&(r=!1);var a=O(n,"x","maxPoolWithArgmax"),s={x:a},o={filterSize:t,strides:e,pad:i,includeBatchInIndex:r},l=z.runKernel(Xf,s,o);return{result:l[0],indexes:l[1]}}var mx=U({maxPoolWithArgmax_:fx});function Zn(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Zn(n,"float32"),i=Zn(n,"float32");return di(e,i)}var r=kr(st(n),t);return z.makeTensor(r,n,t)}function Gr(n,t){if(t===void 0&&(t="float32"),t==="complex64"){var e=Gr(n,"float32"),i=Zn(n,"float32");return di(e,i)}var r=wc(st(n),t);return z.makeTensor(r,n,t)}function gx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","mean"),r=Ze(t,i.shape),a=Sg(i.shape,r),s=a[1],o=st(s),l=Wn(function(u){var c=be(o),h=c.dtype===u.dtype?u:ue(u,c.dtype),d=Le(h,c),p=Ie(d,t,e),f=function(m){var g=u.shape.slice();r.forEach(function(w){g[w]=1});var v=V(m,g),b=Le(J(v,Gr(u.shape,"float32")),o);return b};return{value:p,gradFunc:f}});return l(i)}var qa=U({mean_:gx});function vx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","min"),r=function(o,l){var u=Ze(t,i.shape),c=u,h=nn(c,i.rank),d=i;h!=null&&(d=lt(i,h),c=kn(c.length,i.rank));var p=o.min(d,c);h!=null&&d.dispose();var f=p;if(e){var m=tn(f.shape,u);f=V(p,m),p.dispose()}return l([i,f]),f},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,Au,s)}var ao=U({min_:vx});function yx(n,t){var e,i=O(n,"a","minimum"),r=O(t,"b","minimum");e=rt(i,r),i=e[0],r=e[1],i.dtype==="bool"&&(i=ue(i,"int32"),r=ue(r,"int32")),Qe(i.shape,r.shape);var a=function(o,l){var u=o.minimum(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Tu)}var so=U({minimum_:yx});function bx(n,t){var e,i=O(n,"a","mod"),r=O(t,"b","mod");e=rt(i,r),i=e[0],r=e[1];var a=function(o,l){var u=o.mod(i,r);return l([i,r]),u},s={a:i,b:r};return z.runKernelFunc(a,s,null,Nu)}var uh=U({mod_:bx});function wx(n){var t=O(n,"x","square"),e={},i=[t],r=[];return z.runKernelFunc(function(a,s){return s([t]),a.square(t)},{x:t},null,"Square",e,i,r)}var Ge=U({square_:wx});function Sx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1),n=O(n,"x","moments");var i=Ze(t,n.shape),r=qa(n,i,e),a=r.shape;e||(a=tn(r.shape,i));var s=Ge(ge(ue(n,"float32"),V(r,a))),o=qa(s,i,e);return{mean:r,variance:o}}var Lx=U({moments_:Sx});function Ix(n,t,e,i){for(var r=O(t,"data","multiRNNCell"),a=Ua(e,"c","multiRNNCell"),s=Ua(i,"h","multiRNNCell"),o=r,l=[],u=0;u2)throw new Error("Rank of probabilities must be 1 or 2, but is "+s);e=e||Math.random();var o=s===1?V(r,[1,-1]):r,l=z.runKernelFunc(function(u){return u.multinomial(o,i,t,e)},{logits2D:o});return s===1?V(l,[l.size]):l}var Nx=U({multinomial_:Tx});function xx(n,t){var e,i=O(n,"a","notEqual"),r=O(t,"b","notEqual");e=rt(i,r),i=e[0],r=e[1],Qe(i.shape,r.shape);var a=function(o){return o.notEqual(i,r)},s={a:i,b:r};return z.runKernelFunc(a,s,null,Jf)}var oo=U({notEqual_:xx});function Cx(n){var t=O(n,"input","real"),e=function(r){return r.real(t)},i={input:t};return z.runKernelFunc(e,i,null,im)}var Ga=U({real_:Cx});function Rx(n){var t=O(n,"x","onesLike"),e=function(r,a){if(t.dtype==="complex64"){var s=ch(Ga(t)),o=Ee(to(t));return di(s,o)}return r.onesLike(t)},i={x:t};return z.runKernelFunc(e,i,null,Ru)}var ch=U({onesLike_:Rx});function Ox(n,t){var e=O(n,"v1","outerProduct"),i=O(t,"v2","outerProduct");E(e.rank===1&&i.rank===1,function(){return"Error in outerProduct: inputs must be rank 1, but got ranks "+(e.rank+" and "+i.rank+".")});var r=V(e,[-1,1]),a=V(i,[1,-1]);return Fe(r,a)}var Ex=U({outerProduct_:Ox});function Dx(n,t,e){e===void 0&&(e=0);var i=O(n,"x","pad");if(i.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");var r=function(o,l){return l([i]),o.pad(i,t,e)},a={paddings:t,constantValue:e},s={x:i};return z.runKernelFunc(r,s,null,Eu,a)}var Qi=U({pad_:Dx});function kx(n,t,e){return e===void 0&&(e=0),E(t.length===2,function(){return"Invalid number of paddings. Must be length of 2."}),Qi(n,[t],e)}var Fx=U({pad1d_:kx});function Wx(n,t,e){return e===void 0&&(e=0),E(t.length===2&&t[0].length===2&&t[1].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Qi(n,t,e)}var Ux=U({pad2d_:Wx});function Bx(n,t,e){return e===void 0&&(e=0),E(t.length===3&&t[0].length===2&&t[1].length===2&&t[2].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Qi(n,t,e)}var zx=U({pad3d_:Bx});function Px(n,t,e){return e===void 0&&(e=0),E(t.length===4&&t[0].length===2&&t[1].length===2&&t[2].length===2&&t[3].length===2,function(){return"Invalid number of paddings. Must be length of 2 each."}),Qi(n,t,e)}var _x=U({pad4d_:Px});function Mx(n,t,e){var i=O(n,"x","spaceToBatchND");E(i.rank>=1+t.length,function(){return"input rank "+i.rank+" should be > than [blockShape] "+t.length}),E(e.length===t.length,function(){return"paddings.shape[0] "+e.length+" must be equal to [blockShape] "+t.length}),E(i.shape.reduce(function(o,l,u){return u>0&&u<=t.length?o&&(l+e[u-1][0]+e[u-1][1])%t[u-1]===0:o},!0),function(){return"input spatial dimensions "+i.shape.slice(1)+" with paddings "+e.toString()+" must be divisible by blockShapes "+t.toString()});var r=function(o){return o.spaceToBatchND(i,t,e)},a={x:i},s={blockShape:t,paddings:e};return z.runKernelFunc(r,a,null,Qu,s)}var lo=U({spaceToBatchND_:Mx});function qx(n,t,e,i,r,a){r==null&&(r=[1,1]),a==null&&(a=1),i===0&&(i="valid");var s=O(n,"x","maxPool"),o=s,l=!1;s.rank===3&&(l=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]])),E(_t(a,r),function(){return"Error in pool: Either strides or dilations must be 1. "+("Got strides "+a+" and dilations '"+r+"'")});var u=_r(o.shape,t,a,r,i),c=[u.dilationHeight,u.dilationWidth],h;i==="same"?h=Vx([u.filterHeight,u.filterWidth],c):h=[[0,0],[0,0]];var d=c[0]===1&&c[1]===1,p=Hx([u.inHeight,u.inWidth],c,h),f=p[0],m=p[1],g=d?i:"valid",v=d?o:lo(o,c,f),b=e==="avg"?function(){return Jc(v,t,a,g)}:function(){return lh(v,t,a,g)},w=b(),S=d?w:$s(w,c,m);return l?V(S,[S.shape[1],S.shape[2],S.shape[3]]):S}function Hx(n,t,e){var i=e.map(function(c){return c[0]}),r=e.map(function(c){return c[1]}),a=n.concat(i,r),s=t.map(function(c,h){return(c-a[h]%c)%c}),o=r.map(function(c,h){return c+s[h]}),l=t.map(function(c,h){return[i[h],o[h]]}),u=t.map(function(c,h){return[0,s[h]]});return[l,u]}function Vx(n,t){var e=n.map(function(s,o){return s+(s-1)*(t[o]-1)}),i=e.map(function(s){return s-1}),r=i.map(function(s){return Math.floor(s/2)}),a=i.map(function(s,o){return s-r[o]});return i.map(function(s,o){return[r[o],a[o]]})}var tv=U({pool_:qx});function Gx(n,t){var e,i=O(n,"base","pow"),r=O(t,"exp","pow");e=rt(i,r),i=e[0],r=e[1];var a={a:i,b:r},s=function(o,l){var u=o.pow(i,r);return l([i,r,u]),u};return z.runKernelFunc(s,a,null,Du)}var Qn=U({pow_:Gx});function Yx(n,t){var e=O(n,"x","prelu"),i=O(t,"alpha","prelu"),r=function(s,o){var l=s.prelu(e,i);return o([e,i]),l},a={x:e,alpha:i};return z.runKernelFunc(r,a,null,ku)}var hh=U({prelu_:Yx});function Kx(n,t,e){t===void 0&&(t=null),e===void 0&&(e=!1);var i=O(n,"x","prod"),r=function(o){i.dtype==="bool"&&(i=ue(i,"int32"));var l=Ze(t,i.shape),u=nn(l,i.rank),c=l,h=i;u!=null&&(h=lt(i,u),c=kn(c.length,i.rank));var d=o.prod(h,c);if(e){var p=tn(d.shape,l);d=V(d,p)}return d},a={x:i},s={axis:t,keepDims:e};return z.runKernelFunc(r,a,null,tm,s)}var nv=U({prod_:Kx});function jx(n,t,e){var i=st(n),r=null;if(e==null||e==="float32")r=new Float32Array(i);else if(e==="int32")r=new Int32Array(i);else if(e==="bool")r=new Uint8Array(i);else throw new Error("Unknown data type "+e);for(var a=0;a>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(Yr,n,!1)}),Jx=er(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(Yr,n,!1)}),Zx=er(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(Yr,n,!1)}),Qx=er(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(Yr,n,!1)}),eC=er(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(Yr,n,!1)}),tC=er(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(Yr,n,!1)}),tr=er(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,N){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||N||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,N=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(N||(S=[N++]);R=1||o===0);var l=Math.sqrt(-2*Math.log(o)/o);e=this.mean+this.stdDev*a*l,i=this.mean+this.stdDev*s*l,(!this.truncated||this.isValidTruncated(e))&&(r=!0)}return(!this.truncated||this.isValidTruncated(i))&&(this.nextVal=this.convertValue(i)),this.convertValue(e)},n.prototype.convertValue=function(t){return this.dtype==null||this.dtype==="float32"?t:Math.round(t)},n.prototype.isValidTruncated=function(t){return t<=this.upper&&t>=this.lower},n}(),iC=function(){function n(t,e,i,r){this.alpha=t,this.beta=1/e,this.dtype=i;var a=r||Math.random();this.randu=dh(a.toString()),this.randn=new ph(0,1,i,!1,this.randu()),t<1?this.d=t+2/3:this.d=t-1/3,this.c=1/Math.sqrt(9*this.d)}return n.prototype.nextValue=function(){for(var t,e,i,r,a,s;;){do r=this.randn.nextValue(),s=1+this.c*r;while(s<=0);if(s*=s*s,t=r*r,e=1-.331*t*t,i=.5*t+this.d*(1-s+Math.log(s)),a=this.randu(),a1;if(s||o||l)return Zn([0],i);var u=Math.abs(Math.ceil((t-n)/e)),c=kr(u,i);t0?o+l:o});t[a]=n.shape[e]-s}E(n.shape[e]===t.reduce(function(o,l){return o+l}),function(){return"The sum of sizes must match the size of the axis dimension."}),i=t}return i}function VC(n,t,e){e===void 0&&(e=0);var i=O(n,"x","split"),r=function(o,l){var u=Ze(e,i.shape)[0],c=cv(i,t,u);return o.split(i,c,u)},a={x:i},s={numOrSizeSplits:t,axis:e};return z.runKernelFunc(r,a,null,ec,s)}var jr=U({split_:VC});function qC(n,t){E(n.dtype==="float32",function(){return"The dtype for rfft() must be real value but got "+n.dtype});var e=n.shape[n.shape.length-1],i=n.size/e,r;if(t!=null&&te){var o=n.shape.map(function(v){return v});o[n.shape.length-1]=t-e,r=Ct([n,Zn(o)],n.shape.length-1),e=t}else r=n;var l=Ee(r),u=V(di(r,l),[i,e]),c=uo(u),h=Math.floor(e/2)+1,d=Ga(c),p=to(c),f=jr(d,[h,e-h],d.shape.length-1),m=jr(p,[h,e-h],p.shape.length-1),g=r.shape.slice();return g[r.shape.length-1]=h,V(di(f[0],m[0]),g)}var co=U({rfft_:qC});function GC(n){var t=O(n,"x","sqrt"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.sqrt(t);return r([t]),a},e,null,Ju)}var Mt=U({sqrt_:GC});function YC(n,t){var e,i=O(n,"a","squaredDifference"),r=O(t,"b","squaredDifference");e=rt(i,r),i=e[0],r=e[1],Qe(i.shape,r.shape);var a=function(l,u){var c=l.squaredDifference(i,r);return u([i,r]),c},s={a:i,b:r},o={};return z.runKernelFunc(a,s,null,nc,o)}var ho=U({squaredDifference_:YC});function KC(n,t){var e=O(n,"x","squeeze");return V(e,vm(e.shape,t).newShape)}var po=U({squeeze_:KC});function jC(n,t){t===void 0&&(t=0);var e=Ua(n,"tensors","stack");if(E(e.length>=1,function(){return"Pass at least one tensor to tf.stack"}),e.length===1)return vn(e[0],t);var i=e[0].rank,r=e[0].shape,a=e[0].dtype;E(t<=i,function(){return"Axis must be <= rank of the tensor"}),e.forEach(function(o){Ue(r,o.shape,"All tensors passed to stack must have matching shapes"),E(a===o.dtype,function(){return"All tensors passed to stack must have matching dtypes"})});var s=e.map(function(o){return vn(o,t)});return Ct(s,t)}var nr=U({stack_:jC});function $C(n,t){t===void 0&&(t=0);var e=O(n,"x","step"),i={x:e},r={alpha:t};return z.runKernelFunc(function(a){return a.step(e,t)},i,null,hc,r)}var $r=U({step_:$C});function XC(n,t,e,i,r,a,s,o,l){r===void 0&&(r=0),a===void 0&&(a=0),s===void 0&&(s=0),o===void 0&&(o=0),l===void 0&&(l=0);var u=O(n,"x","stridedSlice"),c=function(p){i==null&&(i=new Array(t.length));var f=Gs(s);if(f.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(s!==0&&o!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(s!==0&&l!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");var m=u.rank-t.length,g=Gs(o),v=u.shape.slice();g.forEach(function(W){t[W]=0,e[W]=1,v.splice(W,0,1)}),u=V(u,v);var b=dg(u.shape,f,m,t,e,i,r,a,s),w=b.begin,S=b.end,L=b.strides;t=w,e=S,i=L;var N=Gs(l);N.forEach(function(W){e[W]=t[W]+1,i[W]=1});var C=ig(t,e,i),R=C.filter(function(W,F){return N.indexOf(F)===-1}),D=i.every(function(W){return W===1});if(D)return V(ze(u,t,C),R);var k=p.stridedSlice(u,t,e,i);return V(k,R)},h={x:u},d={begin:t,end:e,strides:i,beginMask:r,endMask:a,ellipsisMask:s,newAxisMask:o,shrinkAxisMask:l};return z.runKernelFunc(c,h,null,um,d)}var hv=U({stridedSlice_:XC});function JC(n){var t=O(n,"x","tan"),e={x:t};return z.runKernelFunc(function(i,r){var a=i.tan(t);return r([t]),a},e,null,rc)}var dv=U({tan_:JC});function ja(n,t,e){if(Hi(n),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");var i=En(n,e);if(i.length!==2&&i.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return pi(n,t,i,e)}function ZC(n,t,e){if(Hi(n),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");var i=En(n,e);if(i.length!==4&&i.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return pi(n,t,i,e)}function QC(n,t,e){if(Hi(n),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");var i=En(n,e);if(i.length!==5&&i.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return pi(n,t,i,e)}function eR(n,t,e){if(Hi(n),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");var i=En(n,e);if(i.length!==6&&i.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(i.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||i,pi(n,t,i,e)}function tR(n,t,e){t===void 0&&(t=1),e===void 0&&(e=!0);var i=O(n,"x","topk");if(i.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");var r=i.shape[i.shape.length-1];if(t>r)throw new Error("'k' passed to topk() must be <= the last dimension ("+r+") "+("but got "+t));var a={x:i},s={k:t,sorted:e},o=z.runKernelFunc(function(c){return c.topk(i,t,e)},a,null,cm,s),l=o[0],u=o[1];return{values:l,indices:u}}var pv=U({topk_:tR});function nR(n,t,e,i,r){if(t===void 0&&(t=0),e===void 0&&(e=1),i!=null&&i==="bool")throw new Error("Unsupported data type $ { dtype }");for(var a=new ph(t,e,i,!0,r),s=Dn(n,i),o=0;o0,function(){return"The input tensor must be at least 1D"});var i={x:e},r={axis:t},a=z.runKernel(hm,i,r),s=a[0],o=a[1];return{values:s,indices:o}}var fv=U({unique_:rR});function aR(n,t,e){var i=O(n,"x","unsortedSegmentSum"),r=O(t,"segmentIds","unsortedSegmentSum","int32");E(it(e),function(){return"numSegments must be of dtype int"});var a={x:i,segmentIds:r},s={numSegments:e},o=function(l,u){var c=l.unsortedSegmentSum(i,r,e);return u([r]),c};return z.runKernelFunc(o,a,null,uc,s)}var wh=U({unsortedSegmentSum_:aR});function sR(n,t){t===void 0&&(t=0);var e=O(n,"x","unstack");E(t>=-e.shape.length&&t0,function(){return"mask cannot be scalar"}),Ue(o.slice(a,a+s),r.shape,"mask's shape must match the first K dimensions of tensor's shape,"),l=1,u=a;u2)throw new Error("sparseIndices should be a scalar, vector, or matrix,"+(" but got shape "+n.shape+"."));var r=n.rank>0?n.shape[0]:1,a=n.rank>1?n.shape[1]:1;if(e.length!==a)throw new Error("outputShape has incorrect number of elements:,"+(" "+e.length+", should be: "+a+"."));var s=t.size;if(!(t.rank===0||t.rank===1&&s===r))throw new Error("sparseValues has incorrect shape "+(t.shape+", should be [] or ["+r+"]"));if(t.dtype!==i.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function DR(n,t,e,i){i===void 0&&(i=0);var r=O(n,"sparseIndices","sparseToDense","int32"),a=O(t,"sparseValues","sparseToDense"),s=O(i,"defaultValue","sparseToDense",a.dtype);ER(r,a,e,s);var o={sparseIndices:r,sparseValues:a,defaultValue:s},l={outputShape:e};return z.runKernelFunc(function(u){return u.sparseToDense(r,a,e,s)},o,null,lm,l)}var kR=U({sparseToDense_:DR});function FR(n,t){var e=O(t,"indices","gatherND","int32"),i=O(n,"x","gatherND"),r=function(s){return s.gatherND(i,e)},a={params:i,indices:e};return z.runKernelFunc(r,a,null,Bf)}var WR=U({gatherND_:FR});function UR(n,t){if(t==null)return n.shape.slice();if(un(n.shape,t))return t;if(n.shape.length===t.length){for(var e=[],i=0;i=0&&t<1,function(){return"rate must be a float in the range [0, 1), but got "+t+"."}),t===0)return n instanceof Y?r.clone():r;var a=UR(r,e),s=1-t,o=Le(Qs(fe(iv(a,0,1,"float32",i),s)),s);return J(r,o)}var zR=U({dropout_:BR});function kv(n){return Math.floor(Math.pow(2,Math.ceil(Math.log(n)/Math.log(2))))}function Sh(n,t,e){for(var i=1-n%2,r=new Float32Array(n),a=0;a1,function(){return"inTopK() expects the predictions to be of rank 2 or higher, "+("but got "+i.rank)}),E(i.rank-1===r.rank,function(){return"predictions rank should be 1 larger than targets rank, but got predictions rank "+(i.rank+" and targets rank "+r.rank)}),Ue(i.shape.slice(0,i.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension."),a=i.shape[i.shape.length-1],E(e>0&&e<=a,function(){return"'k' passed to inTopK() must be > 0 && <= the predictions last "+("dimension ("+a+"), but got "+e)}),[4,i.data()];case 1:return s=v.sent(),[4,r.data()];case 2:for(o=v.sent(),l=[s.length/a,a],u=l[0],c=l[1],h=Ws("bool",u),d=0;d0&&(e=Ie(e,i)),V(e,n.shape)}function yo(n,t,e){if(t==="linear")return n;if(t==="relu")return Ya(n);if(t==="elu")return th(n);if(t==="relu6")return mh(n);if(t==="prelu")return hh(n,e);throw new Error("Unknown fused activation "+t+".")}var bo=function(n,t){var e=n>0;return!e||t==="linear"};function HR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(d=d||"linear",bo(z.state.gradientDepth,d)===!1){var f=Hr(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),yo(f,d,p)}var m=O(t,"x","conv2d"),g=O(e,"filter","conv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused conv2d: input must be rank 4, but got rank "+(v.rank+".")}),E(g.rank===4,function(){return"Error in fused conv2d: filter must be rank 4, but got rank "+(g.rank+".")}),u!=null&&E(it(r),function(){return"Error in fused conv2d: pad must be an integer when using, "+("dimRoundingMode "+u+" but got pad "+r+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in conv2d: depth of input ("+v.shape[3]+") must match "+("input depth for filter "+g.shape[2]+".")}),E(_t(i,l),function(){return"Error in conv2D: Either strides or dilations must be 1. "+("Got strides "+i+" and dilations '"+l+"'")}),E(s==="NHWC",function(){return"Error in conv2d: got dataFormat of "+s+" but only NHWC is currently supported."});var w=Fn(v.shape,g.shape,i,l,r,u),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=rt(S,m)[0],Qe(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused conv2d"));var N=function(F,P){var H=P,_=H[0],K=H[1],j=H[2],q=H[3],G=go(F,j,d);E(yi(l),function(){return"Error in gradient of fused conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+l+"'")});var Z=Zc(K.shape,G,_,i,r),X=Lh(K,G,_.shape,i,r),ee=[Z,X];if(q!=null){var ne=vo(q,G);ee.push(ne)}return ee},C=function(F){var P=F.fusedConv2d({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=Wn(function(F,P,H){var _=z.runKernelFunc(C,R,null,fc,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:N}});return k(v,g)}else{var W=Wn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,fc,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:N}});return W(v,g,S)}}var VR=U({fusedConv2d_:HR});function qR(n,t,e,i){var r=n;n.rank===3&&(r=V(n,[1,n.shape[0],n.shape[1],n.shape[2]]));var a=t;a.rank===3&&(a=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(l){return l.depthwiseConv2DDerFilter(r,a,i)},o={x:r,dy:a};return z.runKernelFunc(s,o,null,xf)}var Fv=U({depthwiseConv2dNativeBackpropFilter_:qR});function GR(n,t,e,i){var r=t,a=!1;t.rank===3&&(a=!0,r=V(t,[1,t.shape[0],t.shape[1],t.shape[2]]));var s=function(u){return u.depthwiseConv2DDerInput(r,e,i)},o={dy:r},l=z.runKernelFunc(s,o,null,Cf);return a?V(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var Wv=U({depthwiseConv2dNativeBackpropInput_:GR});function YR(n){var t=n.x,e=n.filter,i=n.strides,r=n.pad,a=n.dataFormat,s=a===void 0?"NHWC":a,o=n.dilations,l=o===void 0?[1,1]:o,u=n.dimRoundingMode,c=n.bias,h=n.activation,d=h===void 0?"linear":h,p=n.preluActivationWeights;if(bo(z.state.gradientDepth,d)===!1){var f=Ha(t,e,i,r,s,l,u);return c!=null&&(f=fe(f,c)),yo(f,d,p)}var m=O(t,"x","depthwiseConv2d"),g=O(e,"filter","depthwiseConv2d"),v=m,b=!1;m.rank===3&&(b=!0,v=V(m,[1,m.shape[0],m.shape[1],m.shape[2]])),E(v.rank===4,function(){return"Error in fused depthwiseConv2d: input must be rank 4, but got "+("rank "+v.rank+".")}),E(g.rank===4,function(){return"Error in fused depthwiseConv2d: filter must be rank 4, "+("but got rank "+g.rank+".")}),E(v.shape[3]===g.shape[2],function(){return"Error in fused depthwiseConv2d: number of input channels "+("("+v.shape[3]+") must match the inChannels dimension in ")+("filter "+g.shape[2]+".")}),l==null&&(l=[1,1]),E(_t(i,l),function(){return"Error in fused depthwiseConv2d: Either strides or dilations must "+("be 1. Got strides "+i+" and dilations '"+l+"'")}),u!=null&&E(it(r),function(){return"Error in fused depthwiseConv2d: pad must be an integer when "+("using dimRoundingMode "+u+" but got pad "+r+".")});var w=Fn(v.shape,g.shape,i,l,r,u,!0),S;c!=null&&(S=O(c,"bias","fused conv2d"),S=rt(S,m)[0],Qe(w.outShape,S.shape));var L;p!=null&&(L=O(p,"prelu weights","fused depthwiseConv2d"));var N=function(F,P){E(yi(l),function(){return"Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var H=P[0],_=P[1],K=P[2],j=P[3],q=go(F,K,d),G=Wv(_.shape,q,H,w),Z=Fv(_,q,H.shape,w);if(j!=null){var X=vo(S,q);return[G,Z,X]}return[G,Z]},C=function(F){var P=F.fusedDepthwiseConv2D({input:v,filter:g,convInfo:w,bias:S,activation:d,preluActivationWeights:L});return P},R={x:v,filter:g,bias:S,preluActivationWeights:L},D={strides:i,pad:r,dataFormat:s,dilations:l,dimRoundingMode:u,activation:d};if(c==null){var k=Wn(function(F,P,H){var _=z.runKernelFunc(C,R,null,mc,D);return H([P,F,_]),b&&(_=V(_,[_.shape[1],_.shape[2],_.shape[3]])),{value:_,gradFunc:N}});return k(v,g)}else{var W=Wn(function(F,P,H,_){var K=z.runKernelFunc(C,R,null,mc,D);return _([P,F,K,H]),b&&(K=V(K,[K.shape[1],K.shape[2],K.shape[3]])),{value:K,gradFunc:N}});return W(v,g,S)}}var KR=U({fusedDepthwiseConv2d_:YR});function jR(n){var t,e=n.a,i=n.b,r=n.transposeA,a=r===void 0?!1:r,s=n.transposeB,o=s===void 0?!1:s,l=n.bias,u=n.activation,c=u===void 0?"linear":u,h=n.preluActivationWeights;if(bo(z.state.gradientDepth,c)===!1){var d=Fe(e,i,a,o);return l!=null&&(d=fe(d,l)),yo(d,c,h)}var p=O(e,"a","fused matMul"),f=O(i,"b","fused matMul");t=rt(p,f),p=t[0],f=t[1];var m=a?p.shape[p.rank-2]:p.shape[p.rank-1],g=o?f.shape[f.rank-1]:f.shape[f.rank-2],v=a?p.shape[p.rank-1]:p.shape[p.rank-2],b=o?f.shape[f.rank-2]:f.shape[f.rank-1],w=p.shape.slice(0,-2),S=f.shape.slice(0,-2),L=st(w),N=st(S);E(p.rank>=2&&f.rank>=2&&p.rank===f.rank,function(){return"Error in fused matMul: inputs must have the same rank of at least "+("2, got ranks "+p.rank+" and "+f.rank+".")}),E(un(w,S),function(){return"Error in fused matMul: outer dimensions ("+w+") and ("+(S+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" must match.")}),E(m===g,function(){return"Error in fused matMul: inner shapes ("+m+") and ("+(g+") of Tensors with shapes "+p.shape+" and ")+(f.shape+" and transposeA="+a)+(" and transposeB="+o+" must match.")});var C=p.shape.slice(0,-2).concat([v,b]),R=a?V(p,[L,m,v]):V(p,[L,v,m]),D=o?V(f,[N,b,g]):V(f,[N,g,b]),k;l!=null&&(k=O(l,"bias","fused matMul"),k=rt(k,p)[0],Qe(C,k.shape));var W;h!=null&&(W=O(h,"prelu weights","fused matMul"));var F=function(q,G){var Z=G[0],X=G[1],ee=G[2],ne=G[3],ie=go(V(q,ee.shape),ee,c),te,re;if(!a&&!o?(te=Fe(ie,X,!1,!0),re=Fe(Z,ie,!0,!1)):!a&&o?(te=Fe(ie,X,!1,!1),re=Fe(ie,Z,!0,!1)):a&&!o?(te=Fe(X,ie,!1,!0),re=Fe(Z,ie,!1,!1)):(te=Fe(X,ie,!0,!0),re=Fe(ie,Z,!0,!0)),l!=null){var le=vo(ne,ie);return[te,re,le]}else return[te,re]},P=function(q){var G=q.fusedBatchMatMul({a:R,b:D,transposeA:a,transposeB:o,bias:k,activation:c,preluActivationWeights:W});return G},H={a:R,b:D,bias:k,preluActivationWeights:W},_={transposeA:a,transposeB:o,activation:c};if(l==null){var K=Wn(function(q,G,Z){var X=z.runKernelFunc(P,H,null,pc,_);return Z([q,G,X]),{value:V(X,C),gradFunc:F}});return K(R,D)}else{var j=Wn(function(q,G,Z,X){var ee=z.runKernelFunc(P,H,null,pc,_);return X([q,G,ee,Z]),{value:V(ee,C),gradFunc:F}});return j(R,D,k)}}var $R=U({fusedMatMul_:jR});var XR={__proto__:null,conv2d:VR,depthwiseConv2d:KR,matMul:$R};function JR(n){return Sh(n,.54,.46)}var ZR=U({hammingWindow_:JR});function QR(n){return Sh(n,.5,.5)}var Uv=U({hannWindow_:QR});function eO(n,t,e,i,r){i===void 0&&(i=!1),r===void 0&&(r=0);for(var a=0,s=[];a+t<=n.size;)s.push(ze(n,a,t)),a+=e;if(i)for(;a=1&&i[1]>=1,function(){return"cropSize must be atleast [1,1], but was "+i}),E(r==="bilinear"||r==="nearest",function(){return"method must be bilinear or nearest, but was "+r});var c=function(f){return f.cropAndResize(s,o,l,i,r,a)},h={image:s,boxes:o,boxInd:l},d={method:r,extrapolationValue:a,cropSize:i},p=z.runKernelFunc(c,h,null,Tf,d);return p}var rO=U({cropAndResize_:iO});function aO(n){var t=O(n,"image","flipLeftRight","float32");E(t.rank===4,function(){return"Error in flipLeftRight: image must be rank 4,"+("but got rank "+t.rank+".")});var e={image:t},i=z.runKernel(Uf,e,{});return i}var sO=U({flipLeftRight_:aO});function oO(n,t,e,i){e===void 0&&(e=0),i===void 0&&(i=.5);var r=O(n,"image","rotateWithOffset","float32");E(r.rank===4,function(){return"Error in rotateWithOffset: image must be rank 4,"+("but got rank "+r.rank+".")});var a={image:r},s={radians:t,fillValue:e,center:i},o=z.runKernel(dm,a,s);return o}var lO=U({rotateWithOffset_:oO});function Xr(n,t,e,i,r,a){i==null&&(i=.5),r==null&&(r=Number.NEGATIVE_INFINITY),a==null&&(a=0);var s=n.shape[0];return e=Math.min(e,s),E(0<=i&&i<=1,function(){return"iouThreshold must be in [0, 1], but was '"+i+"'"}),E(n.rank===2,function(){return"boxes must be a 2D tensor, but was of rank '"+n.rank+"'"}),E(n.shape[1]===4,function(){return"boxes must have 4 columns, but 2nd dimension was "+n.shape[1]}),E(t.rank===1,function(){return"scores must be a 1D tensor"}),E(t.shape[0]===s,function(){return"scores has incompatible shape with boxes. Expected "+s+", "+("but was "+t.shape[0])}),E(0<=a&&a<=1,function(){return"softNmsSigma must be in [0, 1], but was '"+a+"'"}),{maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a}}function uO(n,t,e,i,r){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY);var a=O(n,"boxes","nonMaxSuppression"),s=O(t,"scores","nonMaxSuppression"),o=Xr(a,s,e,i,r);e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold;var l={maxOutputSize:e,iouThreshold:i,scoreThreshold:r};return z.runKernelFunc(function(u){return u.nonMaxSuppression(a,s,e,i,r)},{boxes:a,scores:s},null,Zf,l)}var cO=U({nonMaxSuppression_:uO});function dO(n,t,e){var i=hO(n,t,e),r=i<0?-(i+1):i;n.splice(r,0,t)}function hO(n,t,e){return fO(n,t,e||pO)}function pO(n,t){return n>t?1:n>>1);var o=e(t,n[a]);o>0?i=a+1:(r=a,s=!o)}return s?i:-i-1}function zv(n,t,e,i,r){return Ih(n,t,e,i,r,0).selectedIndices}function Pv(n,t,e,i,r,a){return Ih(n,t,e,i,r,0,!1,a,!0)}function _v(n,t,e,i,r,a){return Ih(n,t,e,i,r,a,!0)}function Ih(n,t,e,i,r,a,s,o,l){s===void 0&&(s=!1),o===void 0&&(o=!1),l===void 0&&(l=!1);for(var u=[],c=0;cr&&u.push({score:t[c],boxIndex:c,suppressBeginIndex:0});u.sort(Mv);for(var h=a>0?-.5/a:0,d=[],p=[];d.length0;){var f=u.pop(),m=f.score,g=f.boxIndex,v=f.suppressBeginIndex;if(m=v;--w){var S=mO(n,g,d[w]);if(S>=i){b=!0;break}if(f.score=f.score*gO(i,h,S),f.score<=r)break}f.suppressBeginIndex=d.length,b||(f.score===m?(d.push(g),p.push(f.score)):f.score>r&&dO(u,f,Mv))}var L=d.length,N=e-L;o&&N>0&&(d.push.apply(d,new Array(N).fill(0)),p.push.apply(p,new Array(N).fill(0)));var C={selectedIndices:Kr(d,"int32")};return s&&(C.selectedScores=Kr(p,"float32")),l&&(C.validOutputs=be(L,"int32")),C}function mO(n,t,e){var i=n.subarray(t*4,t*4+4),r=n.subarray(e*4,e*4+4),a=Math.min(i[0],i[2]),s=Math.min(i[1],i[3]),o=Math.max(i[0],i[2]),l=Math.max(i[1],i[3]),u=Math.min(r[0],r[2]),c=Math.min(r[1],r[3]),h=Math.max(r[0],r[2]),d=Math.max(r[1],r[3]),p=(o-a)*(l-s),f=(h-u)*(d-c);if(p<=0||f<=0)return 0;var m=Math.max(a,u),g=Math.max(s,c),v=Math.min(o,h),b=Math.min(l,d),w=Math.max(v-m,0)*Math.max(b-g,0);return w/(p+f-w)}function gO(n,t,e){var i=Math.exp(t*e*e);return e<=n?i:0}function Mv(n,t){return n.score-t.score||n.score===t.score&&t.boxIndex-n.boxIndex}function vO(n,t,e,i,r){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),de(this,void 0,void 0,function(){var a,s,o,l,u,c,h;return pe(this,function(d){switch(d.label){case 0:return a=O(n,"boxes","nonMaxSuppressionAsync"),s=O(t,"scores","nonMaxSuppressionAsync"),o=Xr(a,s,e,i,r),e=o.maxOutputSize,i=o.iouThreshold,r=o.scoreThreshold,[4,Promise.all([a.data(),s.data()])];case 1:return l=d.sent(),u=l[0],c=l[1],h=zv(u,c,e,i,r),a!==n&&a.dispose(),s!==t&&s.dispose(),[2,h]}})})}var yO=vO;function bO(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Xr(s,o,e,i,r,a);e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;var u={boxes:s,scores:o},c={maxOutputSize:e,iouThreshold:i,scoreThreshold:r,softNmsSigma:a},h=z.runKernel(em,u,c);return{selectedIndices:h[0],selectedScores:h[1]}}var wO=U({nonMaxSuppressionWithScore_:bO});function SO(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=0),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d;return pe(this,function(p){switch(p.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Xr(s,o,e,i,r,a),e=l.maxOutputSize,i=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma,[4,Promise.all([s.data(),o.data()])];case 1:return u=p.sent(),c=u[0],h=u[1],d=_v(c,h,e,i,r,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,d]}})})}var LO=SO;function IO(n,t,e,i,r,a){i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1);var s=O(n,"boxes","nonMaxSuppression"),o=O(t,"scores","nonMaxSuppression"),l=Xr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,d={boxes:s,scores:o},p={maxOutputSize:u,iouThreshold:c,scoreThreshold:h,padToMaxOutputSize:a},f=z.runKernel(Qf,d,p);return{selectedIndices:f[0],validOutputs:f[1]}}var AO=U({nonMaxSuppressionPadded_:IO});function TO(n,t,e,i,r,a){return i===void 0&&(i=.5),r===void 0&&(r=Number.NEGATIVE_INFINITY),a===void 0&&(a=!1),de(this,void 0,void 0,function(){var s,o,l,u,c,h,d,p,f,m;return pe(this,function(g){switch(g.label){case 0:return s=O(n,"boxes","nonMaxSuppressionAsync"),o=O(t,"scores","nonMaxSuppressionAsync"),l=Xr(s,o,e,i,r,null),u=l.maxOutputSize,c=l.iouThreshold,h=l.scoreThreshold,[4,Promise.all([s.data(),o.data()])];case 1:return d=g.sent(),p=d[0],f=d[1],m=Pv(p,f,u,c,h,a),s!==n&&s.dispose(),o!==t&&o.dispose(),[2,m]}})})}var NO=TO;function xO(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeBilinear");E(i.rank===3||i.rank===4,function(){return"Error in resizeBilinear: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeBilinear: new shape must 2D, but got shape "+(t+".")});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l=function(d,p){return p([r]),d.resizeBilinear(r,s,o,e)},u={images:r},c={alignCorners:e,size:t},h=z.runKernelFunc(l,u,null,zu,c);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Hv=U({resizeBilinear_:xO});function CO(n,t,e){e===void 0&&(e=!1);var i=O(n,"images","resizeNearestNeighbor");E(i.rank===3||i.rank===4,function(){return"Error in resizeNearestNeighbor: x must be rank 3 or 4, but got "+("rank "+i.rank+".")}),E(t.length===2,function(){return"Error in resizeNearestNeighbor: new shape must 2D, but got shape "+(t+".")}),E(i.dtype==="float32"||i.dtype==="int32",function(){return"`images` must have `int32` or `float32` as dtype"});var r=i,a=!1;i.rank===3&&(a=!0,r=V(i,[1,i.shape[0],i.shape[1],i.shape[2]]));var s=t[0],o=t[1],l={images:r},u={alignCorners:e,size:t},c=function(d,p){return p([r]),d.resizeNearestNeighbor(r,s,o,e)},h=z.runKernelFunc(c,l,null,Bu,u);return a?V(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Vv=U({resizeNearestNeighbor_:CO});function RO(n,t,e){E(t%1===0,function(){return"bandPart(): numLower must be an integer, got "+t+"."}),E(e%1===0,function(){return"bandPart(): numUpper must be an integer, got "+e+"."});var i=O(n,"a","bandPart");E(i.rank>=2,function(){return"bandPart(): Rank must be at least 2, got "+i.rank+"."});var r=i.shape,a=i.shape.slice(-2),s=a[0],o=a[1];if(!(t<=s))throw new Error("bandPart(): numLower ("+t+")"+(" must not be greater than the number of rows ("+s+")."));if(!(e<=o))throw new Error("bandPart(): numUpper ("+e+")"+(" must not be greater than the number of columns ("+o+")."));t<0&&(t=s),e<0&&(e=o);var l=V(fh(0,s,1,"int32"),[-1,1]),u=fh(0,o,1,"int32"),c=ge(l,u),h=Zi($i(c,be(+t,"int32")),ji(c,be(-e,"int32"))),d=Zn([s,o],i.dtype);return V(nr(fo(V(i,[-1,s,o])).map(function(p){return mn(h,p,d)})),r)}var OO=U({bandPart_:RO});function EO(n){var t;if(Array.isArray(n)){t=!1,E(n!=null&&n.length>0,function(){return"Gram-Schmidt process: input must not be null, undefined, or empty"});for(var e=n[0].shape[0],i=function(l){E(n[l].shape[0]===e,function(){return"Gram-Schmidt: Non-unique lengths found in the input vectors: "+("("+n[l].shape[0]+" vs. "+e+")")})},r=1;r0)for(var c=0;c=2,function(){return"qr() requires input tensor to have a rank >= 2, but got rank "+n.rank}),n.rank===2)return qv(n,t);var e=n.shape.slice(0,n.shape.length-2).reduce(function(l,u){return l*u}),i=fo(V(n,[e,n.shape[n.shape.length-2],n.shape[n.shape.length-1]]),0),r=[],a=[];i.forEach(function(l){var u=qv(l,t),c=u[0],h=u[1];r.push(c),a.push(h)});var s=V(nr(r,0),n.shape),o=V(nr(a,0),n.shape);return[s,o]}function qv(n,t){return t===void 0&&(t=!1),z.tidy(function(){E(n.shape.length===2,function(){return"qr2d() requires a 2D Tensor, but got a "+n.shape.length+"D Tensor."});for(var e=n.shape[0],i=n.shape[1],r=qg(e),a=Gi(n),s=ja([[1]],[1,1]),o=Gi(s),l=e>=i?i:e,u=function(h){var d,p=a,f=o,m=r;d=z.tidy(function(){var g=ze(a,[h,h],[e-h,1]),v=mo(g),b=ze(a,[h,h],[1,1]),w=mn(bi(b,0),ja([[-1]]),ja([[1]])),S=ge(b,J(w,v)),L=Le(g,S);L.shape[0]===1?o=Gi(s):o=Ct([s,ze(L,[1,0],[L.shape[0]-1,L.shape[1]])],0);var N=gt(Le(Fe(w,S),v)),C=ze(a,[h,0],[e-h,i]),R=J(N,o),D=lt(o);if(h===0)a=ge(C,Fe(R,Fe(D,C)));else{var k=ge(C,Fe(R,Fe(D,C)));a=Ct([ze(a,[0,0],[h,i]),k],0)}var W=lt(R),F=ze(r,[0,h],[e,r.shape[1]-h]);if(h===0)r=ge(F,Fe(Fe(F,o),W));else{var P=ge(F,Fe(Fe(F,o),W));r=Ct([ze(r,[0,0],[e,h]),P],1)}return[o,a,r]}),o=d[0],a=d[1],r=d[2],Pt([p,f,m])},c=0;ci&&(r=ze(r,[0,0],[e,i]),a=ze(a,[0,0],[i,i])),[r,a]})}var FO=U({qr_:kO});(function(n){n[n.NONE=0]="NONE",n[n.MEAN=1]="MEAN",n[n.SUM=2]="SUM",n[n.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(I.Reduction||(I.Reduction={}));function WO(n,t,e){e===void 0&&(e=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var i=O(n,"losses","computeWeightedLoss"),r=null;t!=null&&(r=O(t,"weights","computeWeightedLoss"));var a=r==null?i:J(i,r);if(e===I.Reduction.NONE)return a;if(e===I.Reduction.SUM)return Ie(a);if(e===I.Reduction.MEAN){if(r==null)return qa(a);var s=i.size/r.size,o=Le(Ie(a),Ie(r));return s>1?Le(o,be(s)):o}if(e===I.Reduction.SUM_BY_NONZERO_WEIGHTS){if(r==null)return Le(Ie(a),be(i.size));var l=J(r,Gr(i.shape)),u=ue(Ie(oo(l,be(0))),"float32");return Le(Ie(a),u)}throw Error("Unknown reduction: "+e)}var ti=U({computeWeightedLoss_:WO});function UO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","absoluteDifference"),a=O(t,"predictions","absoluteDifference"),s=null;e!=null&&(s=O(e,"weights","absoluteDifference")),Ue(r.shape,a.shape,"Error in absoluteDifference: ");var o=Gt(ge(r,a));return ti(o,s,i)}var BO=U({absoluteDifference_:UO});function zO(n,t,e,i,r){r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","cosineDistance"),s=O(t,"predictions","cosineDistance"),o=null;i!=null&&(o=O(i,"weights","cosineDistance")),Ue(a.shape,s.shape,"Error in cosineDistance: ");var l=be(1),u=ge(l,Ie(J(a,s),e,!0));return ti(u,o,r)}var PO=U({cosineDistance_:zO});function _O(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","hingeLoss"),a=O(t,"predictions","hingeLoss"),s=null;e!=null&&(s=O(e,"weights","hingeLoss")),Ue(r.shape,a.shape,"Error in hingeLoss: ");var o=be(1);r=ge(J(be(2),r),o);var l=Ya(ge(o,J(r,a)));return ti(l,s,i)}var MO=U({hingeLoss_:_O});function HO(n,t,e,i,r){i===void 0&&(i=1),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","huberLoss"),s=O(t,"predictions","huberLoss"),o=null;e!=null&&(o=O(e,"weights","huberLoss")),Ue(a.shape,s.shape,"Error in huberLoss: ");var l=be(i),u=Gt(ge(s,a)),c=so(u,l),h=ge(u,c),d=fe(J(be(.5),Ge(c)),J(l,h));return ti(d,o,r)}var VO=U({huberLoss_:HO});function qO(n,t,e,i,r){i===void 0&&(i=1e-7),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"labels","logLoss"),s=O(t,"predictions","logLoss"),o=null;e!=null&&(o=O(e,"weights","logLoss")),Ue(a.shape,s.shape,"Error in logLoss: ");var l=be(1),u=be(i),c=gt(J(a,Xi(fe(s,u)))),h=J(ge(l,a),Xi(fe(ge(l,s),u))),d=ge(c,h);return ti(d,o,r)}var GO=U({logLoss_:qO});function YO(n,t,e,i){i===void 0&&(i=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var r=O(n,"labels","meanSquaredError"),a=O(t,"predictions","meanSquaredError"),s=null;e!=null&&(s=O(e,"weights","meanSquaredError")),Ue(r.shape,a.shape,"Error in meanSquaredError: ");var o=ho(r,a);return ti(o,s,i)}var KO=U({meanSquaredError_:YO});function jO(n,t){var e=O(n,"labels","sigmoidCrossEntropyWithLogits"),i=O(t,"logits","sigmoidCrossEntropyWithLogits");Ue(e.shape,i.shape,"Error in sigmoidCrossEntropyWithLogits: ");var r=Ya(i),a=J(i,e),s=rh(gn(gt(Gt(i))));return fe(ge(r,a),s)}function $O(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"multiClassLabels","sigmoidCrossEntropy"),s=O(t,"logits","sigmoidCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","sigmoidCrossEntropy")),Ue(a.shape,s.shape,"Error in sigmoidCrossEntropy: "),i>0){var l=be(i),u=be(1),c=be(.5);a=fe(J(a,ge(u,l)),J(c,l))}var h=jO(a,s);return ti(h,o,r)}var XO=U({sigmoidCrossEntropy_:$O});function JO(n,t,e){if(e===void 0&&(e=-1),e===-1&&(e=t.rank-1),e!==t.rank-1)throw Error("Softmax cross entropy along a non-last dimension is not yet "+("supported. Labels / logits was rank "+t.rank+" ")+("and dim was "+e));var i=Wn(function(r,a,s){var o=!0,l=sh(a,[e],o),u=ge(ue(a,"float32"),l);s([r,u]);var c=gt(J(u,r)),h=Ie(c,[e]),d=function(p,f){var m=f[0],g=f[1],v=tn(p.shape,[e]);return[J(V(p,v),ge(ue(m,"float32"),gn(g))),J(V(p,v),ge(gn(g),ue(m,"float32")))]};return{value:h,gradFunc:d}});return i(n,t)}function ZO(n,t,e,i,r){i===void 0&&(i=0),r===void 0&&(r=I.Reduction.SUM_BY_NONZERO_WEIGHTS);var a=O(n,"onehotLabels","softmaxCrossEntropy"),s=O(t,"logits","softmaxCrossEntropy"),o=null;if(e!=null&&(o=O(e,"weights","softmaxCrossEntropy")),Ue(a.shape,s.shape,"Error in softmaxCrossEntropy: "),i>0){var l=be(i),u=be(1),c=be(a.shape[1]);a=fe(J(a,ge(u,l)),Le(l,c))}var h=JO(a,s);return ti(h,o,r)}var QO=U({softmaxCrossEntropy_:ZO});var eE={fft:uo,ifft:Ka,rfft:co,irfft:bh},tE={hammingWindow:ZR,hannWindow:Uv,frame:Bv,stft:nO},nE={flipLeftRight:sO,resizeNearestNeighbor:Vv,resizeBilinear:Hv,rotateWithOffset:lO,cropAndResize:rO,nonMaxSuppression:cO,nonMaxSuppressionAsync:yO,nonMaxSuppressionWithScore:wO,nonMaxSuppressionWithScoreAsync:LO,nonMaxSuppressionPadded:AO,nonMaxSuppressionPaddedAsync:NO},iE={bandPart:OO,gramSchmidt:DO,qr:FO},rE={absoluteDifference:BO,computeWeightedLoss:ti,cosineDistance:PO,hingeLoss:MO,huberLoss:VO,logLoss:GO,meanSquaredError:KO,sigmoidCrossEntropy:XO,softmaxCrossEntropy:QO};var wi=function(n){$n(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.minimize=function(e,i,r){i===void 0&&(i=!1);var a=this.computeGradients(e,r),s=a.value,o=a.grads;if(r!=null){var l=r.map(function(u){return{name:u.name,tensor:o[u.name]}});this.applyGradients(l)}else this.applyGradients(o);return Pt(o),i?s:(s.dispose(),null)},Object.defineProperty(t.prototype,"iterations",{get:function(){return this.iterations_==null&&(this.iterations_=0),this.iterations_},enumerable:!0,configurable:!0}),t.prototype.incrementIterations=function(){this.iterations_=this.iterations+1},t.prototype.computeGradients=function(e,i){return Jg(e,i)},t.prototype.dispose=function(){this.iterations_!=null&&Pt(this.iterations_)},t.prototype.saveIterations=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){return this.iterations_==null&&(this.iterations_=0),[2,{name:"iter",tensor:be(this.iterations_,"int32")}]})})},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for this optimizer yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for this optimizer class "+(""+this.getClassName()))})})},t.prototype.extractIterations=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return i=this,[4,e[0].tensor.data()];case 1:return i.iterations_=r.sent()[0],[2,e.slice(1)]}})})},t}(fg);Object.defineProperty(wi,Symbol.hasInstance,{value:function(n){return n.minimize!=null&&n.computeGradients!=null&&n.applyGradients!=null}});var Ah=function(n){$n(t,n);function t(e,i,r){r===void 0&&(r=null);var a=n.call(this)||this;return a.learningRate=e,a.rho=i,a.epsilon=r,a.accumulatedGrads=[],a.accumulatedUpdates=[],r==null&&(a.epsilon=z.backend.epsilon()),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedGrads[s]==null&&(i.accumulatedGrads[s]={originalName:a+"/accum_grad",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedUpdates[s]==null&&(i.accumulatedUpdates[s]={originalName:a+"/accum_var",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable,h=i.accumulatedUpdates[s].variable;ft(function(){var d=fe(J(c,i.rho),J(Ge(u),1-i.rho)),p=J(Le(Mt(fe(h,i.epsilon)),Mt(fe(c,i.epsilon))),u),f=fe(J(h,i.rho),J(Ge(p),1-i.rho));c.assign(d),h.assign(f);var m=fe(J(p,-i.learningRate),o);o.assign(m)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedUpdates!=null&&(Pt(this.accumulatedGrads.map(function(e){return e.variable})),Pt(this.accumulatedUpdates.map(function(e){return e.variable})))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedGrads.concat(this.accumulatedUpdates),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=e.length/2,r=!1,this.accumulatedGrads=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedUpdates=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.rho,i.epsilon)},t.className="Adadelta",t}(wi);vi(Ah);var Th=function(n){$n(t,n);function t(e,i){i===void 0&&(i=.1);var r=n.call(this)||this;return r.learningRate=e,r.initialAccumulatorValue=i,r.accumulatedGrads=[],r}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulatedGrads[s]==null){var l=!1;i.accumulatedGrads[s]={originalName:a+"/accumulator",variable:ft(function(){return nh(o.shape,i.initialAccumulatorValue).variable(l)})}}var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedGrads[s].variable;ft(function(){var h=fe(c,Ge(u));c.assign(h);var d=fe(J(Le(u,Mt(fe(h,z.backend.epsilon()))),-i.learningRate),o);o.assign(d)})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedGrads!=null&&Pt(this.accumulatedGrads.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulatedGrads.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulatedGrads=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}},t.fromConfig=function(e,i){return new e(i.learningRate,i.initialAccumulatorValue)},t.className="Adagrad",t}(wi);vi(Th);var Nh=function(n){$n(t,n);function t(e,i,r,a){a===void 0&&(a=null);var s=n.call(this)||this;return s.learningRate=e,s.beta1=i,s.beta2=r,s.epsilon=a,s.accumulatedFirstMoment=[],s.accumulatedSecondMoment=[],ft(function(){s.accBeta1=be(i).variable(),s.accBeta2=be(r).variable()}),a==null&&(s.epsilon=z.backend.epsilon()),s}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=ge(1,i.accBeta2);r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:ft(function(){return Ee(u).variable(c)})}),i.accumulatedSecondMoment[l]==null&&(i.accumulatedSecondMoment[l]={originalName:o+"/v",variable:ft(function(){return Ee(u).variable(c)})});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedSecondMoment[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=fe(J(p,i.beta2),J(Ge(h),1-i.beta2)),g=Le(f,a),v=Le(m,s);d.assign(f),p.assign(m);var b=fe(J(Le(g,fe(Mt(v),i.epsilon)),-i.learningRate),u);u.assign(b)}),i.accBeta1.assign(J(i.accBeta1,i.beta1)),i.accBeta2.assign(J(i.accBeta2,i.beta2))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&Pt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedSecondMoment!=null&&Pt(this.accumulatedSecondMoment.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedFirstMoment.concat(this.accumulatedSecondMoment),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r,a=this;return pe(this,function(s){switch(s.label){case 0:return[4,this.extractIterations(e)];case 1:return e=s.sent(),ft(function(){a.accBeta1.assign(Qn(a.beta1,a.iterations_+1)),a.accBeta2.assign(Qn(a.beta2,a.iterations_+1))}),i=e.length/2,r=!1,this.accumulatedFirstMoment=e.slice(0,i).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),this.accumulatedSecondMoment=e.slice(i,i*2).map(function(o){return{originalName:o.name,variable:o.tensor.variable(r)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon)},t.className="Adam",t}(wi);vi(Nh);var xh=function(n){$n(t,n);function t(e,i,r,a,s){a===void 0&&(a=null),s===void 0&&(s=0);var o=n.call(this)||this;return o.learningRate=e,o.beta1=i,o.beta2=r,o.epsilon=a,o.decay=s,o.accumulatedFirstMoment=[],o.accumulatedWeightedInfNorm=[],ft(function(){o.iteration=be(0).variable(),o.accBeta1=be(i).variable()}),a==null&&(o.epsilon=z.backend.epsilon()),o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);ft(function(){var a=ge(1,i.accBeta1),s=Le(-i.learningRate,fe(J(i.iteration,i.decay),1));r.forEach(function(o,l){var u=z.registeredVariables[o],c=!1;i.accumulatedFirstMoment[l]==null&&(i.accumulatedFirstMoment[l]={originalName:o+"/m",variable:Ee(u).variable(c)}),i.accumulatedWeightedInfNorm[l]==null&&(i.accumulatedWeightedInfNorm[l]={originalName:o+"/v",variable:Ee(u).variable(c)});var h=Array.isArray(e)?e[l].tensor:e[o];if(h==null)return;var d=i.accumulatedFirstMoment[l].variable,p=i.accumulatedWeightedInfNorm[l].variable,f=fe(J(d,i.beta1),J(h,1-i.beta1)),m=J(p,i.beta2),g=Gt(h),v=qr(m,g);d.assign(f),p.assign(v);var b=fe(J(Le(s,a),Le(f,fe(v,i.epsilon))),u);u.assign(b)}),i.iteration.assign(fe(i.iteration,1)),i.accBeta1.assign(J(i.accBeta1,i.beta1))}),this.incrementIterations()},t.prototype.dispose=function(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&Pt(this.accumulatedFirstMoment.map(function(e){return e.variable})),this.accumulatedWeightedInfNorm!=null&&Pt(this.accumulatedWeightedInfNorm.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){throw new Error("getWeights() is not implemented for Adamax yet.")})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){throw new Error("setWeights() is not implemented for Adamax yet.")})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon,decay:this.decay}},t.fromConfig=function(e,i){return new e(i.learningRate,i.beta1,i.beta2,i.epsilon,i.decay)},t.className="Adamax",t}(wi);vi(xh);var wo=function(n){$n(t,n);function t(e){var i=n.call(this)||this;return i.learningRate=e,i.setLearningRate(e),i}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=Array.isArray(e)?e[s].tensor:e[a];if(o==null)return;var l=z.registeredVariables[a];ft(function(){var u=fe(J(i.c,o),l);l.assign(u)})}),this.incrementIterations()},t.prototype.setLearningRate=function(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=vg(be(-e))},t.prototype.dispose=function(){this.c.dispose()},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()]]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){return pe(this,function(i){switch(i.label){case 0:return[4,this.extractIterations(e)];case 1:if(e=i.sent(),e.length!==0)throw new Error("SGD optimizer does not have settable weights.");return[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate}},t.fromConfig=function(e,i){return new e(i.learningRate)},t.className="SGD",t}(wi);vi(wo);var Ch=function(n){$n(t,n);function t(e,i,r){r===void 0&&(r=!1);var a=n.call(this,e)||this;return a.learningRate=e,a.momentum=i,a.useNesterov=r,a.accumulations=[],a.m=be(a.momentum),a}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a];if(i.accumulations[s]==null){var l=!1;i.accumulations[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}}var u=i.accumulations[s].variable,c=Array.isArray(e)?e[s].tensor:e[a];if(c==null)return;ft(function(){var h,d=fe(J(i.m,u),c);i.useNesterov?h=fe(J(i.c,fe(c,J(d,i.m))),o):h=fe(J(i.c,d),o),u.assign(d),o.assign(h)})}),this.incrementIterations()},t.prototype.dispose=function(){this.m.dispose(),this.accumulations!=null&&Pt(this.accumulations.map(function(e){return e.variable}))},t.prototype.setMomentum=function(e){this.momentum=e},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){return pe(this,function(e){switch(e.label){case 0:return[4,this.saveIterations()];case 1:return[2,[e.sent()].concat(this.accumulations.map(function(i){return{name:i.originalName,tensor:i.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i;return pe(this,function(r){switch(r.label){case 0:return[4,this.extractIterations(e)];case 1:return e=r.sent(),i=!1,this.accumulations=e.map(function(a){return{originalName:a.name,variable:a.tensor.variable(i)}}),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}},t.fromConfig=function(e,i){return new e(i.learningRate,i.momentum,i.useNesterov)},t.className="Momentum",t}(wo);vi(Ch);var Rh=function(n){$n(t,n);function t(e,i,r,a,s){i===void 0&&(i=.9),r===void 0&&(r=0),a===void 0&&(a=null),s===void 0&&(s=!1);var o=n.call(this)||this;if(o.learningRate=e,o.decay=i,o.momentum=r,o.epsilon=a,o.accumulatedMeanSquares=[],o.accumulatedMoments=[],o.accumulatedMeanGrads=[],o.centered=s,a==null&&(o.epsilon=z.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.");return o}return t.prototype.applyGradients=function(e){var i=this,r=Array.isArray(e)?e.map(function(a){return a.name}):Object.keys(e);r.forEach(function(a,s){var o=z.registeredVariables[a],l=!1;i.accumulatedMeanSquares[s]==null&&(i.accumulatedMeanSquares[s]={originalName:a+"/rms",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMoments[s]==null&&(i.accumulatedMoments[s]={originalName:a+"/momentum",variable:ft(function(){return Ee(o).variable(l)})}),i.accumulatedMeanGrads[s]==null&&i.centered&&(i.accumulatedMeanGrads[s]={originalName:a+"/mg",variable:ft(function(){return Ee(o).variable(l)})});var u=Array.isArray(e)?e[s].tensor:e[a];if(u==null)return;var c=i.accumulatedMeanSquares[s].variable,h=i.accumulatedMoments[s].variable;ft(function(){var d=fe(J(c,i.decay),J(Ge(u),1-i.decay));if(i.centered){var p=i.accumulatedMeanGrads[s].variable,f=fe(J(p,i.decay),J(u,1-i.decay)),m=Le(J(u,i.learningRate),Mt(ge(d,fe(Ge(f),i.epsilon)))),g=fe(J(h,i.momentum),m);c.assign(d),p.assign(f),h.assign(g);var v=ge(o,g);o.assign(v)}else{var b=fe(J(c,i.decay),J(Ge(u),1-i.decay)),g=fe(J(h,i.momentum),Le(J(u,i.learningRate),Mt(fe(b,i.epsilon))));c.assign(b),h.assign(g);var v=ge(o,g);o.assign(v)}})}),this.incrementIterations()},t.prototype.dispose=function(){this.accumulatedMeanSquares!=null&&Pt(this.accumulatedMeanSquares.map(function(e){return e.variable})),this.accumulatedMeanGrads!=null&&this.centered&&Pt(this.accumulatedMeanGrads.map(function(e){return e.variable})),this.accumulatedMoments!=null&&Pt(this.accumulatedMoments.map(function(e){return e.variable}))},t.prototype.getWeights=function(){return de(this,void 0,void 0,function(){var e;return pe(this,function(i){switch(i.label){case 0:return e=this.accumulatedMeanSquares.concat(this.accumulatedMoments),this.centered&&e.push.apply(e,this.accumulatedMeanGrads),[4,this.saveIterations()];case 1:return[2,[i.sent()].concat(e.map(function(r){return{name:r.originalName,tensor:r.variable}}))]}})})},t.prototype.setWeights=function(e){return de(this,void 0,void 0,function(){var i,r;return pe(this,function(a){switch(a.label){case 0:return[4,this.extractIterations(e)];case 1:return e=a.sent(),i=this.centered?e.length/3:e.length/2,r=!1,this.accumulatedMeanSquares=e.slice(0,i).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.accumulatedMoments=e.slice(i,i*2).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}}),this.centered&&(this.accumulatedMeanGrads=e.slice(i*2,i*3).map(function(s){return{originalName:s.name,variable:s.tensor.variable(r)}})),[2]}})})},t.prototype.getConfig=function(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}},t.fromConfig=function(e,i){return new e(i.learningRate,i.decay,i.momentum,i.epsilon,i.centered)},t.className="RMSProp",t}(wi);vi(Rh);var ir=function(){function n(){}return n.sgd=function(t){return new wo(t)},n.momentum=function(t,e,i){return i===void 0&&(i=!1),new Ch(t,e,i)},n.rmsprop=function(t,e,i,r,a){return e===void 0&&(e=.9),i===void 0&&(i=0),r===void 0&&(r=null),a===void 0&&(a=!1),new Rh(t,e,i,r,a)},n.adam=function(t,e,i,r){return t===void 0&&(t=.001),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),new Nh(t,e,i,r)},n.adadelta=function(t,e,i){return t===void 0&&(t=.001),e===void 0&&(e=.95),i===void 0&&(i=null),new Ah(t,e,i)},n.adamax=function(t,e,i,r,a){return t===void 0&&(t=.002),e===void 0&&(e=.9),i===void 0&&(i=.999),r===void 0&&(r=null),a===void 0&&(a=0),new xh(t,e,i,r,a)},n.adagrad=function(t,e){return e===void 0&&(e=.1),new Th(t,e)},n}();var aE={sgd:ir.sgd,momentum:ir.momentum,adadelta:ir.adadelta,adagrad:ir.adagrad,rmsprop:ir.rmsprop,adamax:ir.adamax,adam:ir.adam};var sE=function(){return typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:function(n){return n()}}();function oE(){return new Promise(function(n){return sE(function(){return n()})})}function lE(n,t,e){var i=e*(typeof n=="number"?n:n[0]),r=t*(typeof n=="number"?n:n[1]);return[i,r]}function uE(n,t,e,i){i===void 0&&(i=!0);var r=[];if(i)r=r.concat(t.slice(0)),r.push(n[0]/e),r=r.concat(n.slice(1));else{r=r.concat(n[0]);for(var a=t.length,s=0;s=t*2+1||r%2===1?s.push(r):a.push(r);i.push.apply(i,a),i.push(0),i.push.apply(i,s)}return i}function hE(n,t,e,i){i===void 0&&(i=!0);var r=[];i?r.push(n[0]/e):r.push(n[0]*e);for(var a=1;a0&&(o=Ie(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ie(o,l)),V(o,i.shape)};return{a,b:s}}};var HE={kernelName:Dl,saveAllInputs:!0,gradFunc:function(n,t){var e={};return t.forEach(function(i,r){e[r]=function(){return n.clone()}}),e}};var VE={kernelName:kl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var qE={kernelName:Fl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Ee(e)}}}};var GE={kernelName:Wl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Le(n,Mt(ge(be(1),Ge(ue(e,"float32")))))}}}};var YE={kernelName:Ul,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=Mt(fe(be(1),Ge(ue(e,"float32"))));return Le(n,i)}}}};var KE={kernelName:Pl,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=Qe(e.shape,i.shape),a=function(){var o=fe(Ge(e),Ge(i)),l=J(n,Le(i,o)),u=mt(e.shape,r);return u.length>0&&(l=Ie(l,u)),V(l,e.shape)},s=function(){var o=fe(Ge(e),Ge(i)),l=gt(J(n,Le(e,o))),u=mt(i.shape,r);return u.length>0&&(l=Ie(l,u)),V(l,i.shape)};return{a,b:s}}};var jE={kernelName:Bl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Le(n,fe(Ge(ue(e,"float32")),1))}}}};var $E={kernelName:zl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Le(n,ge(be(1),Ge(ue(e,"float32"))))}}}};function XE(n,t,e,i,r,a,s){r===void 0&&(r=[1,1,1]);var o=O(n,"dy","avgPool3dBackprop"),l=O(t,"input","avgPool3dBackprop"),u=o,c=l,h=!1;l.rank===4&&(h=!0,u=V(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=V(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]])),E(u.rank===5,function(){return"Error in avgPool3dBackprop: dy must be rank 5 but got rank "+(u.rank+".")}),E(c.rank===5,function(){return"Error in avgPool3dBackprop: input must be rank 5 but got rank "+(c.rank+".")}),E(_t(i,r),function(){return"Error in avgPool3dBackprop: Either strides or dilations "+("must be 1. Got strides "+i+" and dilations '"+r+"'")}),s!=null&&E(it(a),function(){return"Error in maxPool3dBackprop: pad must be an integer when "+("using, dimRoundingMode "+s+" but got pad "+a+".")});var d=function(g){var v=_a(c.shape,e,i,r,a,s);return g.avgPool3dBackprop(u,c,v)},p={dy:u,input:c},f={filterSize:e,strides:i,dilations:r,pad:a,dimRoundingMode:s},m=z.runKernelFunc(d,p,null,wf,f);return h?V(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var JE=U({avgPool3dBackprop_:XE});var ZE={kernelName:Ml,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.dilations,l=r.pad,u=r.dimRoundingMode,c=o??[1,1,1];return{x:function(){return JE(n,i,a,s,c,l,u)}}}};function QE(n,t,e,i,r){var a=O(n,"dy","avgPoolBackprop"),s=O(t,"input","avgPoolBackprop");E(s.rank===a.rank,function(){return"Rank of input ("+s.rank+") does not match rank of dy ("+a.rank+")"});var o=s,l=a,u=!1;s.rank===3&&(u=!0,o=V(s,[1,s.shape[0],s.shape[1],s.shape[2]]),l=V(a,[1,a.shape[0],a.shape[1],a.shape[2]])),E(l.rank===4,function(){return"Error in avgPoolBackprop: dy must be rank 4 but got rank "+(l.rank+".")}),E(o.rank===4,function(){return"Error in avgPoolBackprop: input must be rank 4 but got rank "+(o.rank+".")});var c=function(f){var m=_r(o.shape,e,i,1,r);return f.avgPoolBackprop(l,o,m)},h={dy:l,input:o},d={filterSize:e,strides:i,pad:r},p=z.runKernelFunc(c,h,null,bf,d);return u?V(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var eD=U({avgPoolBackprop_:QE});var tD={kernelName:_l,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.filterSize,s=r.strides,o=r.pad;return{x:function(){return eD(n,i,a,s,o)}}}};var nD={kernelName:Hl,inputsToSave:["a","b"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.transposeA,l=s.transposeB;return!o&&!l?{a:function(){return Fe(n,a,!1,!0)},b:function(){return Fe(r,n,!0,!1)}}:!o&&l?{a:function(){return Fe(n,a,!1,!1)},b:function(){return Fe(n,r,!0,!1)}}:o&&!l?{a:function(){return Fe(a,n,!1,!0)},b:function(){return Fe(r,n,!1,!1)}}:{a:function(){return Fe(a,n,!0,!0)},b:function(){return Fe(n,r,!0,!0)}}}};var iD={kernelName:Vl,gradFunc:function(n,t,e){var i=e,r=i.blockShape,a=i.crops;return{x:function(){return lo(n,r,a)}}}};var rD={kernelName:ql,gradFunc:function(n,t,e){for(var i=e,r=i.inputShape,a=i.shape,s=Array.from(a),o=r.length-1;o>=0;o--)if(r[o]===a[o])s[o]=1;else if(r[o]!==1)throw new Error("broadcastTo(): ["+r+"] cannot be broadcast to ["+a+"].");for(var l=[],o=0;o1&&l.push(o);return{x:function(){return Ie(n,l,!0)}}}};var aD={kernelName:ks,gradFunc:function(n){return{x:function(){return n.clone()}}}};var sD={kernelName:Gl,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var oD={kernelName:Yl,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.clipValueMin,s=r.clipValueMax;return{x:function(){return mn(Zi(ji(i,a),$i(i,s)),n,Ee(n))}}}};var lD={kernelName:Kl,saveAllInputs:!0,gradFunc:function(n,t,e){var i=t.map(function(l){return l.shape}),r=e.axis,a=Ze(r,t[0].shape)[0],s=i.map(function(l){return l[a]}),o=jr(n,s,a);return o.map(function(l){return function(){return l}})}};var uD={kernelName:jl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.dilations,l=s.strides,u=s.pad,c=s.dataFormat;return E(yi(o),function(){return"Error in gradient of conv2D: dilation rates greater than 1 "+("are not yet supported in gradients. Got dilations '"+o+"'")}),{x:function(){return Zc(r.shape,n,a,l,u,c)},filter:function(){return Lh(r,n,a.shape,l,u,c)}}}};var cD={kernelName:$l,inputsToSave:["dy","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s=e,o=s.strides,l=s.pad,u=s.dataFormat,c=s.dimRoundingMode;return{dy:function(){return Hr(n,a,o,l,u,1,c)},filter:function(){return Lh(n,r,a.shape,o,l,u,c)}}}};function hD(n,t,e,i,r){var a=n;n.rank===4&&(a=V(n,[1,n.shape[0],n.shape[1],n.shape[2],n.shape[3]]));var s=t;s.rank===4&&(s=V(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),E(a.rank===5,function(){return"Error in conv3dDerFilter: input must be rank 5, but got shape "+(a.shape+".")}),E(s.rank===5,function(){return"Error in conv3dDerFilter: dy must be rank 5, but got shape "+(s.shape+".")}),E(e.length===5,function(){return"Error in conv3dDerFilter: filterShape must be length 5, but got "+(e+".")}),E(a.shape[4]===e[3],function(){return"Error in conv3dDerFilter: depth of input "+a.shape[4]+") must "+("match input depth in filter ("+e[3]+".")}),E(s.shape[4]===e[4],function(){return"Error in conv3dDerFilter: depth of dy ("+s.shape[4]+") must "+("match output depth for filter ("+e[4]+").")});var o=function(c){var h=1,d=Pa(a.shape,e,i,h,r);return c.conv3dDerFilter(a,s,d)},l={x:a,y:s},u={strides:i,pad:r};return z.runKernelFunc(o,l,null,If,u)}var dD=U({conv3DBackpropFilter_:hD});var pD={kernelName:Xl,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad;E(yi(r),function(){return"Error in gradient of conv3D: dilation rates greater than 1 are "+("not yet supported in gradients. Got dilations '"+r+"'")});var o=t[0],l=t[1];return{x:function(){return Bg(o.shape,n,l,a,s)},filter:function(){return dD(o,n,l.shape,a,s)}}}};var fD={kernelName:Jl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(gt(vh(ue(e,"float32"))),n)}}}};var mD={kernelName:Zl,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(yh(ue(e,"float32")),n)}}}};var gD={kernelName:Ql,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e,a=r.axis,s=r.exclusive,o=r.reverse;return{x:function(){var l=nn([a],i.rank),u=eh(n,a,s,!o);return l!=null&&(u=lt(u,l)),u}}}};var vD={kernelName:eu,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=e,r=i.dilations,a=i.strides,s=i.pad,o=i.dimRoundingMode,l=r??[1,1];E(yi(l),function(){return"Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations "+("'"+l+"'")});var u=t,c=u[0],h=u[1];E(c.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: input must be "+("rank 4, but got rank "+c.rank+".")}),E(h.rank===4,function(){return"Error in gradient of depthwiseConv2dNative: filter must be "+("rank 4, but got rank "+h.rank+".")}),E(c.shape[3]===h.shape[2],function(){return"Error in gradient of depthwiseConv2d: number of input "+("channels ("+c.shape[3]+") must match the inChannels dimension ")+("in filter "+h.shape[2]+".")}),E(_t(a,l),function(){return"Error in gradient of depthwiseConv2d: Either strides or "+("dilations must be 1. Got strides "+a+" and dilations ")+("'"+l+"'.")}),o!=null&&E(it(s),function(){return"Error in depthwiseConv2d: pad must be an integer when using, "+("dimRoundingMode "+o+" but got pad "+s+".")});var d=Fn(c.shape,h.shape,a,l,s,o,!0);return{x:function(){return Wv(c.shape,n,h,d)},filter:function(){return Fv(c,n,h.shape,d)}}}};var yD={kernelName:tu,inputsToSave:["x","filter"],gradFunc:function(n,t,e){var i=t,r=i[0],a=i[1],s={x:r,filter:a,dy:n},o={x:r,filter:a,dy:n};return{x:function(){return z.runKernel(Of,s,e)},filter:function(){return z.runKernel(Ef,o,e)}}}};var bD={kernelName:nu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=Qe(e.shape,i.shape),a=function(){var o=Le(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ie(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ie(o,l),i.shape));var u=Ge(i);return gt(Le(o,ue(u,"float32")))};return{a,b:s}}};var wD={kernelName:iu,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=function(a){return a.eluDer(n,e)},r={dy:n,y:e};return{x:function(){return z.runKernelFunc(i,r,null,Df)}}}};var SD={kernelName:ru,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J(gn(gt(Ge(e))),2/Math.sqrt(Math.PI));return{x:function(){return J(n,i)}}}};var LD={kernelName:au,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,e)}}}};var ID={kernelName:su,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,gn(e))}}}};var AD={kernelName:ou,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var TD={kernelName:lu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=Qe(e.shape,i.shape),a=function(){var o=Le(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ie(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);l.length>0&&(o=V(Ie(o,l),i.shape));var u=Ge(i);return gt(Le(o,ue(u,"float32")))};return{a,b:s}}};var ND={kernelName:uu,inputsToSave:["x","mean","variance","scale"],gradFunc:function(n,t,e){var i=e.varianceEpsilon,r=t[0],a=t[1],s=t[2],o=t[3],l=o??be(1),u=mt(a.shape,r.shape),c=[];if(a.rank===1){for(var h=0;h0?V(Ie(n,o),e.shape):n},s=function(){var o=J(n,gt(Qs(Le(e,i)))),l=mt(i.shape,r);return l.length>0?V(Ie(o,l),i.shape):o};return{a,b:s}}};var $D={kernelName:xu,inputsToSave:["a","b"],gradFunc:function(n,t){var e=t[0],i=t[1],r=Qe(e.shape,i.shape),a=function(){var o=J(n,ue(i,"float32")),l=mt(e.shape,r);return l.length>0?V(Ie(o,l),e.shape):o},s=function(){var o=J(n,ue(e,"float32")),l=mt(i.shape,r);return l.length>0?V(Ie(o,l),i.shape):o};return{a,b:s}}};var XD={kernelName:Cu,gradFunc:function(n){return{x:function(){return gt(n)}}}};var JD={kernelName:Ou,inputsToSave:["indices"],gradFunc:function(n,t){var e=t[0];return{indices:function(){return Zn(e.shape,"float32")}}}};var ZD={kernelName:Ru,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var Jv={kernelName:Eu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.paddings,a=r.map(function(s){return s[0]});return{x:function(){return ze(n,a,i.shape)}}}};var QD={kernelName:Du,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:function(n,t){var e=t[0],i=t[1],r=t[2],a=e,s=i,o=Qe(a.shape,s.shape),l=function(){var c=ue(s,"float32"),h=J(n,J(c,Qn(a,ge(c,be(1))))),d=mt(a.shape,o);return d.length>0&&(h=Ie(h,d)),V(h,a.shape)},u=function(){var c=bi(a,0),h=mn(c,Xi(a),Ee(a)),d=J(n,J(r,h)),p=mt(s.shape,o);return p.length>0&&(d=Ie(d,p)),V(d,s.shape)};return{a:l,b:u}}};var ek={kernelName:ku,inputsToSave:["x","alpha"],gradFunc:function(n,t){var e=t[0],i=t[1],r=bi(e,0);return{x:function(){return mn(r,n,J(n,i))},alpha:function(){var a=mn(r,Ee(n),J(n,e)),s=mt(i.shape,n.shape);return s.length>0&&(a=Ie(a,s)),V(a,i.shape)}}}};var tk={kernelName:Fu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Le(n,gt(Ge(e)))}}}};var nk={kernelName:Pu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0],i=J($i(e,6),$r(e));return{x:function(){return J(n,ue(i,"float32"))}}}};var ik={kernelName:Wu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,ue($r(e),"float32"))}}}};var rk={kernelName:Uu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return V(n,e.shape)}}}};var ak={kernelName:zu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeBilinearBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,am,e)};return{images:s}}};var sk={kernelName:Bu,inputsToSave:["images"],gradFunc:function(n,t,e){var i=t[0],r=function(o){var l=e.alignCorners;return o.resizeNearestNeighborBackprop(n,i,l)},a={images:i},s=function(){return z.runKernelFunc(r,a,null,rm,e)};return{images:s}}};var ok={kernelName:_u,gradFunc:function(n,t,e){var i=e.dims,r=Ze(i,n.shape);return{x:function(){return ei(n,r)}}}};var lk={kernelName:Mu,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var uk={kernelName:Hu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return gt(Le(n,J(Qn(e,1.5),2)))}}}};var ck={kernelName:Vu,inputsToSave:["condition"],gradFunc:function(n,t){var e=t[0];return{condition:function(){return ue(Ee(e),"float32")},t:function(){return J(n,ue(e,n.dtype))},e:function(){return J(n,ue(ro(e),n.dtype))}}}};var hk={kernelName:qu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){var i=bi(e,be(0)),r=be(Gv),a=be(Yv),s=J(n,a),o=J(J(n,r),gn(ue(e,"float32")));return mn(i,s,o)}}}};var dk={kernelName:$u,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(n,J(e,ge(be(1),e)))}}}};var pk={kernelName:ju,gradFunc:function(n){return{x:function(){return Ee(n)}}}};var fk={kernelName:Yu,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Zs(ue(e,"float32")),n)}}}};var mk={kernelName:Ku,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(Qc(ue(e,"float32")),n)}}}};var gk={kernelName:Gu,inputsToSave:["x"],gradFunc:function(n,t,e){for(var i=t[0],r=e,a=r.begin,s=r.size,o=i.shape,l=Vc(i,a,s),u=l[0],c=l[1],h=[],d=0;d0&&(o=Ie(o,l)),V(o,e.shape)},s=function(){var o=n,l=mt(i.shape,r);return l.length>0&&(o=Ie(o,l)),V(gt(o),i.shape)};return{a,b:s}}};var Ak={kernelName:Zu,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=i.shape.slice(),a=e.axis,s=Ze(a,i.shape);s.forEach(function(u){r[u]=1});var o=V(n,r),l=J(o,Gr(i.shape,"float32"));return{x:function(){return l}}}};var Tk={kernelName:rc,inputsToSave:["x"],gradFunc:function(n,t){var e=t[0];return{x:function(){return Le(n,Ge(Zs(e)))}}}};var Nk={kernelName:ac,outputsToSave:[!0],gradFunc:function(n,t){var e=t[0];return{x:function(){return J(ge(be(1),Ge(e)),n)}}}};var xk={kernelName:sc,inputsToSave:["x"],gradFunc:function(n,t,e){var i=t[0],r=e.reps,a=function(){var s=Ee(i);if(i.rank===1)for(var o=0;o{"use strict";Object.defineProperty($e,"__esModule",{value:!0});var y=rr();var Eh=function(n,t){return Eh=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Eh(n,t)};function Q(n,t){Eh(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}var Yt=function(){return Yt=Object.assign||function(t){for(var e,i=1,r=arguments.length;i0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]t?1:0}function So(n,t){return-1*Bk(n,t)}function Li(n){if(n==null)return n;for(var t=[],e=0,i=n;e=0),Bn(i>=e),Array.isArray(n)&&n.length>=e&&n.length<=i&&n.every(function(r){return typeof r===t})}function At(n,t){Array.isArray(n)?(y.util.assert(n.length>0,function(){return t+" is unexpectedly an empty array."}),n.forEach(function(e,i){return At(e,"element "+(i+1)+" of "+t)})):y.util.assert(Number.isInteger(n)&&n>0,function(){return"Expected "+t+" to be a positive integer, but got "+(ny(n)+".")})}function ny(n){return n===null?"null":Array.isArray(n)?"["+n.map(function(t){return ny(t)}).join(",")+"]":typeof n=="string"?'"'+n+'"':""+n}function Pk(n,t){var e=y.util.now(),i,r=function(){for(var a=[],s=0;s0){var e=n+"_"+t;return Zr.set(e,1),e}else return n}var Xk=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function dy(n){return!!n.match(Xk)}function Jk(n){return n===parseInt(n.toString(),10)}function Ii(n,t,e){t==null&&(t=0),e==null&&(e=n.length);for(var i=1,r=t;r= 2"+(" but got x shape = "+n.shape+" and y shape = "+t.shape));if(t.rank>=3){var r=n.shape.slice(-1)[0],a=t.shape.slice(-2)[0];if(r!==a)throw new Te("If rank y >= 3, then the second last dim"+(" of y must equal the last dim of x but got x shape = "+n.shape+" and ")+(" y shape = "+t.shape))}if(n.rank===2&&t.rank===2){var s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?_h(n.rank,i,yn()):null,activation:e})}else{var l=n.shape.slice(),u=l.pop();n=n.reshape([-1,u]);var c=t.shape.slice(),h=c.pop(),a=c.pop(),d=c.concat([h]),p=Array.from({length:t.rank},function(b,w){return w===0?t.rank-2:w<=t.rank-2?w-1:w});t=t.transpose(p).reshape([a,-1]);var f=l.concat(d),s=!1,o=!1;return y.fused.matMul({a:n,b:t,transposeA:s,transposeB:o,bias:i?_h(n.rank,i,yn()):null,activation:e}).reshape(f)}}function vy(n,t,e){return y.tidy(function(){return Array.isArray(t)?t=y.tensor1d(t,"int32"):t=t.toInt(),y.gather(n,t,e)})}function es(n){return y.mul(n,n)}function _h(n,t,e){var i=t.shape;if(t.rank!==1&&t.rank!==n)throw new M("Unexpected bias dimensions: "+t.rank+("; expected it to be 1 or "+n));if(n===5){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1,1]):t.reshape([1,i[3],i[0],i[1],i[2]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===4){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1,1]):t.reshape([1,i[2],i[0],i[1]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,1,i[0]]):t.reshape([1].concat(i))}else if(n===3){if(e==="channelsFirst")return i.length===1?t.reshape([1,i[0],1]):t.reshape([1,i[1],i[0]]);if(e==="channelsLast")return i.length===1?t.reshape([1,1,i[0]]):t.reshape([1].concat(i))}else if(n<3)return t;throw new M("Unsupported input rank by biasAdd: "+t.rank)}function Pn(n,t,e){return y.tidy(function(){return e==null&&(e=yn()),ct(e),n.add(_h(n.rank,t,e))})}function t3(n,t){if(t===void 0&&(t=1),t!==1)throw new Te("Support for alpha values other than 1 ("+t+") is not implemented yet.");return y.elu(n)}function n3(n){return y.tidy(function(){return y.div(n,y.abs(n).add(1))})}function yy(n,t,e,i){return y.tidy(function(){return y.dropout(n,t,e,i)})}function i3(n){return y.tidy(function(){var t=y.add(.5,y.mul(.2,n));return y.clipByValue(t,0,1)})}function ts(n,t,e){return e===void 0&&(e=!1),e?n():t()}var r3=["fanIn","fanOut","fanAvg"],a3=["normal","uniform","truncatedNormal"];function s3(n){Jr(r3,"FanMode",n)}function o3(n){Jr(a3,"Distribution",n)}var hn=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.fromConfigUsesCustomObjects=function(){return!1},t.prototype.getConfig=function(){return{}},t}(y.serialization.Serializable),by=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.zeros(e,i)},t.className="Zeros",t}(hn);y.serialization.registerClass(by);var Mh=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.apply=function(e,i){return y.ones(e,i)},t.className="Ones",t}(hn);y.serialization.registerClass(Mh);var wy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(typeof e!="object")throw new M("Expected argument of type ConstantConfig but got "+e);if(e.value===void 0)throw new M("config must have value set but got "+e);return i.value=e.value,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){return y.mul(y.scalar(r.value),y.ones(e,i))})},t.prototype.getConfig=function(){return{value:this.value}},t.className="Constant",t}(hn);y.serialization.registerClass(wy);var Sy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MINVAL=-.05,i.DEFAULT_MAXVAL=.05,i.minval=e.minval||i.DEFAULT_MINVAL,i.maxval=e.maxval||i.DEFAULT_MAXVAL,i.seed=e.seed,i}return t.prototype.apply=function(e,i){return y.randomUniform(e,this.minval,this.maxval,i)},t.prototype.getConfig=function(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}},t.className="RandomUniform",t}(hn);y.serialization.registerClass(Sy);var Ly=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("randomNormal does not support dType "+i+".");return Io(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="RandomNormal",t}(hn);y.serialization.registerClass(Ly);var Iy=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.DEFAULT_MEAN=0,i.DEFAULT_STDDEV=.05,i.mean=e.mean||i.DEFAULT_MEAN,i.stddev=e.stddev||i.DEFAULT_STDDEV,i.seed=e.seed,i}return t.prototype.apply=function(e,i){if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te("truncatedNormal does not support dType "+i+".");return y.truncatedNormal(e,this.mean,this.stddev,i,this.seed)},t.prototype.getConfig=function(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}},t.className="TruncatedNormal",t}(hn);y.serialization.registerClass(Iy);var Ay=function(n){Q(t,n);function t(e){var i=n.call(this)||this;return i.gain=e.gain!=null?e.gain:1,i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length!==2||e[0]!==e[1])throw new M("Identity matrix initializer can only be used for 2D square matrices.");return y.mul(r.gain,y.eye(e[0]))})},t.prototype.getConfig=function(){return{gain:this.gain}},t.className="Identity",t}(hn);y.serialization.registerClass(Ay);function l3(n,t){t===void 0&&(t="channelsLast");var e,i;if(ct(t),n.length===2)e=n[0],i=n[1];else if([3,4,5].indexOf(n.length)!==-1){if(t==="channelsFirst"){var r=Ii(n,2);e=n[1]*r,i=n[0]*r}else if(t==="channelsLast"){var r=Ii(n,0,n.length-2);e=n[n.length-2]*r,i=n[n.length-1]*r}}else{var a=Ii(n);e=Math.sqrt(a),i=Math.sqrt(a)}return[e,i]}var Kt=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e.scale<0)throw new M("scale must be a positive float. Got: "+e.scale);return i.scale=e.scale==null?1:e.scale,i.mode=e.mode==null?"fanIn":e.mode,s3(i.mode),i.distribution=e.distribution==null?"normal":e.distribution,o3(i.distribution),i.seed=e.seed,i}return t.prototype.apply=function(e,i){var r=l3(e),a=r[0],s=r[1],o=this.scale;if(this.mode==="fanIn"?o/=Math.max(1,a):this.mode==="fanOut"?o/=Math.max(1,s):o/=Math.max(1,(a+s)/2),this.distribution==="normal"){var l=Math.sqrt(o);if(i=i||"float32",i!=="float32"&&i!=="int32")throw new Te(this.getClassName()+" does not support dType "+i+".");return y.truncatedNormal(e,0,l,i,this.seed)}else{var u=Math.sqrt(3*o);return y.randomUniform(e,-u,u,i)}},t.prototype.getConfig=function(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}},t.className="VarianceScaling",t}(hn);y.serialization.registerClass(Kt);var Hh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="GlorotUniform",t}(Kt);y.serialization.registerClass(Hh);var Vh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="GlorotNormal",t}(Kt);y.serialization.registerClass(Vh);var qh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="HeNormal",t}(Kt);y.serialization.registerClass(qh);var Gh=function(n){Q(t,n);function t(e){return n.call(this,{scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="HeUniform",t}(Kt);y.serialization.registerClass(Gh);var Yh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="LeCunNormal",t}(Kt);y.serialization.registerClass(Yh);var Kh=function(n){Q(t,n);function t(e){return n.call(this,{scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})||this}return t.prototype.getClassName=function(){return Kt.className},t.className="LeCunNormal",t}(Kt);y.serialization.registerClass(Kh);var Ty=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(i.DEFAULT_GAIN=1,i.gain=e.gain==null?i.DEFAULT_GAIN:e.gain,i.seed=e.seed,i.seed!=null)throw new Te("Random seed is not implemented for Orthogonal Initializer yet.");return i}return t.prototype.apply=function(e,i){var r=this;return y.tidy(function(){if(e.length<2)throw new Te("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn("Orthogonal initializer is being called on a matrix with more "+("than 2000 ("+e[0]*e[1]+") elements: ")+"Slowness may result.");var a=e[0]>e[1]?[e[1],e[0]]:e,s=Io(a,0,1,"float32"),o=y.linalg.gramSchmidt(s);return e[0]>e[1]&&(o=o.transpose()),y.mul(r.gain,o)})},t.prototype.getConfig=function(){return{gain:this.gain,seed:this.seed}},t.className="Orthogonal",t}(hn);y.serialization.registerClass(Ty);var Ny={constant:"Constant",glorotNormal:"GlorotNormal",glorotUniform:"GlorotUniform",heNormal:"HeNormal",heUniform:"HeUniform",identity:"Identity",leCunNormal:"LeCunNormal",leCunUniform:"LeCunUniform",ones:"Ones",orthogonal:"Orthogonal",randomNormal:"RandomNormal",randomUniform:"RandomUniform",truncatedNormal:"TruncatedNormal",varianceScaling:"VarianceScaling",zeros:"Zeros"};function xy(n,t){return t===void 0&&(t={}),$a(n,y.serialization.SerializationMap.getMap().classNameMap,t,"initializer")}function at(n){return kh(n)}function et(n){if(typeof n=="string"){var t=n in Ny?Ny[n]:n;if(t==="GlorotNormal")return new Vh;if(t==="GlorotUniform")return new Hh;if(t==="HeNormal")return new qh;if(t==="HeUniform")return new Gh;if(t==="LeCunNormal")return new Yh;if(t==="LeCunUniform")return new Kh;var e={};return e.className=t,e.config={},xy(e)}else return n instanceof hn?n:xy(n)}function u3(){return new by}function c3(){return new Mh}function h3(n){return new wy(n)}function d3(n){return new Sy(n)}function p3(n){return new Ly(n)}function f3(n){return new Iy(n)}function m3(n){return new Ay(n)}function g3(n){return new Kt(n)}function v3(n){return new Hh(n)}function y3(n){return new Vh(n)}function b3(n){return new qh(n)}function w3(n){return new Gh(n)}function S3(n){return new Yh(n)}function L3(n){return new Kh(n)}function I3(n){return new Ty(n)}var A3={__proto__:null,zeros:u3,ones:c3,constant:h3,randomUniform:d3,randomNormal:p3,truncatedNormal:f3,identity:m3,varianceScaling:g3,glorotUniform:v3,glorotNormal:y3,heNormal:b3,heUniform:w3,leCunNormal:S3,leCunUniform:L3,orthogonal:I3};var T3=0;function Cy(){return T3++}var Ao={};function To(n){return n===void 0&&(n=""),n in Ao||(Ao[n]=0),Ao[n]+=1,n+Ao[n].toString()}function jh(n){return Array.isArray(n)&&Array.isArray(n[0])}function No(n){return n.length===0?[]:Array.isArray(n[0])?n:[n]}function xe(n){var t;if(Array.isArray(n)){if(n.length!==1)throw new M("Expected Tensor length to be 1; got "+n.length);t=n[0]}else t=n;return t}function Ye(n){if(Array.isArray(n)&&Array.isArray(n[0])){if(n.length===1)return n=n,n[0];throw new M("Expected exactly 1 Shape; got "+n.length)}else return n}function xo(n){for(var t=0,e=0,i=n;e1)throw new Si("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use `getInputAt(nodeIndex)` instead.');if(this.inboundNodes.length===0)throw new Si("Layer "+this.name+" is not connected, no input to return.");return Ht(this.getNodeAtIndex(0,"input").inputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"output",{get:function(){if(this.inboundNodes.length===0)throw new Si("Layer "+this.name+" has no inbound nodes.");if(this.inboundNodes.length>1)throw new Si("Layer "+this.name+' has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `getOutputAt(nodeIndex)` instead.');return Ht(this.getNodeAtIndex(0,"output").outputTensors)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this._losses},enumerable:!0,configurable:!0}),t.prototype.calculateLosses=function(){return this.losses.map(function(e){return e()})},Object.defineProperty(t.prototype,"updates",{get:function(){return this._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"built",{get:function(){return this._built},set:function(e){this._built=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainable",{get:function(){return this.trainable_},set:function(e){this._trainableWeights.forEach(function(i){return i.trainable=e}),this.trainable_=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable_?this._trainableWeights.filter(function(e){return e.trainable}):[]},set:function(e){this._trainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this._trainableWeights.filter(function(e){return!e.trainable}).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)},set:function(e){this._nonTrainableWeights=e},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"weights",{get:function(){return this.trainableWeights.concat(this.nonTrainableWeights)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"stateful",{get:function(){return this._stateful},enumerable:!0,configurable:!0}),t.prototype.resetStates=function(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")},t.prototype.assertInputCompatibility=function(e){if(e=Xe(e),this.inputSpec==null||this.inputSpec.length===0)return;var i=Xe(this.inputSpec);if(e.length!==i.length)throw new M("Layer "+this.name+" expects "+i.length+" inputs, "+("but it received "+e.length+" input tensors. ")+("Input received: "+e));for(var r=0;r=0?l[c]:l[l.length+c];if(h!=null&&[h,null].indexOf(d)===-1)throw new M("Input "+r+" is incompatible with layer "+(this.name+": expected axis "+c+" of input shape to ")+("have value "+h+" but got shape "+l+"."))}}if(s.shape!=null)for(var p=0;p0&&Array.isArray(R[0])?v=R.map(function(W,F){return new wn(D,W,r,Xe(e),i,r.name,F)}):v=new wn(D,R,r,Xe(e),i,r.name),r.addInboundNode(e,v,null,null,C,R,i),r._refCount++,r.activityRegularizer!=null)throw new Te("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return v}})},t.prototype.warnOnIncompatibleInputShape=function(e){if(this.batchInputShape==null)return;if(e.length!==this.batchInputShape.length)console.warn("The rank of the input tensor provided (shape: "+(JSON.stringify(e)+") does not match that of the ")+("batchInputShape ("+JSON.stringify(this.batchInputShape)+") ")+("of the layer "+this.name));else{var i=!1;this.batchInputShape.forEach(function(r,a){r!=null&&e[a]!=null&&e[a]!==r&&(i=!0)}),i&&console.warn("The shape of the input tensor "+("("+JSON.stringify(e)+") does not ")+("match the expectation of layer "+this.name+": ")+(""+JSON.stringify(this.batchInputShape)))}},Object.defineProperty(t.prototype,"outputShape",{get:function(){if(this.inboundNodes==null||this.inboundNodes.length===0)throw new Si("The layer "+this.name+" has never been called and thus has no defined output shape.");for(var e=[],i=0,r=this.inboundNodes;i0)&&(t=n.sourceLayer,e=n.nodeIndex),t.inboundNodes.length===0)return[n];var i=t.inboundNodes[e];if(i.inboundLayers.length===0)return i.inputTensors;for(var r=[],a=0;a0?[4,Promise.all(t)]:[3,2];case 1:for(o=u.sent(),l=0;l=0&&Number.isInteger(t),function(){return"Verbosity level is expected to be an integer >= 0, "+("but got "+t)}),n.checkForDuplicate(e),n.constructors[t]==null&&(n.constructors[t]=[]),n.constructors[t].push(e)},n.checkForDuplicate=function(t){for(var e in n.constructors){var i=n.constructors[+e];i.forEach(function(r){if(r===t)throw new M("Duplicate callback constructor.")})}},n.clear=function(){n.constructors={}},n.createCallbacks=function(t){var e=[];for(var i in n.constructors){var r=+i;t>=r&&e.push.apply(e,n.constructors[r])}return e.map(function(a){return new a})},n.constructors={},n}();function _y(n,t,e,i,r,a,s,o,l){var u=new Uy,c=[new D3].concat(Py.createCallbacks(t));n!=null&&c.push.apply(c,n),c.push(u);var h=new Wy(c);return h.setParams({epochs:e,initialEpoch:i,samples:r,steps:a,batchSize:s,verbose:t,doValidation:o,metrics:l}),{callbackList:h,history:u}}function Sn(n,t,e){return t===void 0&&(t={}),e===void 0&&(e=!1),$a(n,y.serialization.SerializationMap.getMap().classNameMap,t,"layer",e)}function Ro(n,t){return y.tidy(function(){n.dtype!=="float32"&&(n=n.asType("float32"));var e=y.sum(es(n),t,!0),i=y.fill(e.shape,vt()),r=y.sqrt(y.maximum(e,i));return y.div(n,r)})}function ur(n,t){return y.tidy(function(){return y.mean(es(y.sub(t,n)),-1)})}function Oo(n,t){return y.tidy(function(){return y.mean(y.abs(y.sub(t,n)),-1)})}function ta(n,t){return y.tidy(function(){var e=y.sub(n,t),i=y.clipByValue(y.abs(n),vt(),Number.MAX_VALUE),r=y.abs(y.div(e,i));return y.mul(100,y.mean(r,-1))})}function k3(n,t){return y.tidy(function(){var e=y.clipByValue(t,vt(),Number.MAX_VALUE),i=y.log(y.add(1,e)),r=y.clipByValue(n,vt(),Number.MAX_VALUE),a=y.log(y.add(1,r));return y.mean(es(y.sub(i,a)),-1)})}function F3(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(es(e),-1)})}function W3(n,t){return y.tidy(function(){var e=y.maximum(0,y.sub(1,y.mul(n,t)));return y.mean(e,-1)})}function U3(n,t){return y.tidy(function(){var e=y.sum(y.mul(n,t),-1),i=y.max(y.mul(y.sub(1,n),t),-1);return y.maximum(0,y.add(1,y.sub(i,e)))})}function B3(n,t){return y.tidy(function(){var e=Math.log(2),i=y.sub(t,n),r=y.sub(y.add(i,y.softplus(y.mul(-2,i))),e);return y.mean(r,-1)})}function is(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){if(e)t=y.softmax(t);else{var i=y.sum(t,t.shape.length-1,!0);t=y.div(t,i)}return t=y.clipByValue(t,vt(),1-vt()),y.neg(y.sum(y.mul(n.toFloat(),y.log(t)),t.shape.length-1))})}function Eo(n,t,e){return e===void 0&&(e=!1),y.tidy(function(){var i=y.floor(Qk(n)).toInt();t=y.clipByValue(t,vt(),1-vt());var r=t.shape,a=y.oneHot(i,r[r.length-1]).reshape(r);return is(a,t,e)})}function z3(n,t){if(!y.util.arraysEqual(n.shape,t.shape))throw new M("logits and labels must have the same shape, but got shapes "+(JSON.stringify(n.shape)+" and "+JSON.stringify(t.shape)));return y.tidy(function(){var e=t.relu(),i=t.abs().neg();return e.sub(t.mul(n)).add(i.exp().log1p())})}function Do(n,t){return y.tidy(function(){var e;return e=y.clipByValue(t,vt(),1-vt()),e=y.log(y.div(e,y.sub(1,e))),y.mean(z3(n,e),-1)})}function P3(n,t){return y.tidy(function(){var e=y.clipByValue(n,vt(),1),i=y.clipByValue(t,vt(),1);return y.sum(y.mul(n,y.log(y.div(e,i))),-1)})}function _3(n,t){return y.tidy(function(){var e=y.log(y.add(vt(),t));return y.mean(y.sub(t,y.mul(n,e)),-1)})}function Jh(n,t){return y.tidy(function(){var e=Ro(n,-1),i=Ro(t,-1),r=y.mul(e,i);return y.neg(y.sum(r,-1))})}var ko={meanSquaredError:ur,meanAbsoluteError:Oo,meanAbsolutePercentageError:ta,meanSquaredLogarithmicError:k3,squaredHinge:F3,hinge:W3,categoricalHinge:U3,logcosh:B3,categoricalCrossentropy:is,sparseCategoricalCrossentropy:Eo,binaryCrossentropy:Do,kullbackLeiblerDivergence:P3,poisson:_3,cosineProximity:Jh};function Zh(n){if(typeof n=="string"){if(n in ko)return ko[n];var t="Unknown loss "+n;throw n.toLowerCase().includes("softmaxcrossentropy")&&(t="Unknown loss "+n+'. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy'),new M(t)}else return n}function Qh(n,t){return y.tidy(function(){var e=y.mul(.5,y.onesLike(t)),i=Za(y.greater(t,e),n.dtype);return y.mean(y.equal(n,i),-1)})}function ed(n,t){return y.tidy(function(){return Za(y.equal(y.argMax(n,-1),y.argMax(t,-1)),"float32")})}function My(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(1)).sum().cast("float32")})}function M3(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(1),t.equal(0)).sum().cast("float32")})}function H3(n,t){return y.tidy(function(){return y.logicalAnd(n.equal(0),t.equal(1)).sum().cast("float32")})}function Hy(n,t){return y.tidy(function(){var e=My(n,t),i=H3(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function V3(n,t){return y.tidy(function(){var e=My(n,t),i=M3(n,t),r=e.add(i);return y.where(y.greater(r,0),e.div(r),0).cast("float32")})}function Vy(n,t){return Do(n,t)}function qy(n,t){return n.rank===t.rank&&(n=n.squeeze([n.rank-1])),t=t.argMax(-1),t.dtype!==n.dtype&&(t=t.asType(n.dtype)),y.equal(n,t).asType("float32")}var q3=ur,G3=ur,Y3=Oo,K3=Oo,j3=ta,$3=ta,td=is,X3=Jh,Gy=Eo,Fo={binaryAccuracy:Qh,categoricalAccuracy:ed,precision:Hy,categoricalCrossentropy:td,sparseCategoricalCrossentropy:Gy,mse:q3,MSE:G3,mae:Y3,MAE:K3,mape:j3,MAPE:$3,cosine:X3};function J3(n){if(typeof n=="string"&&n in Fo)return Fo[n];if(typeof n!="string"&&n!=null)return n;throw new M("Unknown metric "+n)}function Wo(n){if(Bn(n!==null,"Unknown LossOrMetricFn "+n),typeof n=="string")return n;for(var t=void 0,e=0,i=Object.keys(ko);eYy&&console.warn('User-defined metadata of model "'+t+'" is too large in '+("size (length="+i.length+" when serialized). It is not ")+"recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= "+(Yy+"."))}}function nd(n){if(n===null)return!0;if(typeof n=="object")if(Object.getPrototypeOf(n)===Object.prototype){for(var t=Object.keys(n),e=0,i=t;e1||o.length===1&&o[0].inboundLayers.length>1){t=!1;break}i.push.apply(i,o)}if(t)for(var l=0,u=n.layers;l0&&(i=i.slice(0,i.length-1)+" "),i+=n[r],i=i.slice(0,t[r]),i+=" ".repeat(t[r]-i.length);e(i)}function t4(n,t,e){var i;try{i=JSON.stringify(n.outputShape)}catch(o){i="multiple"}var r=n.name,a=n.getClassName(),s=[r+" ("+a+")",i,n.countParams().toString()];Uo(s,t,e)}function n4(n,t,e,i){var r;try{r=JSON.stringify(n.outputShape)}catch(v){r="multiple"}for(var a=[],s=0,o=n.inboundNodes;s0&&e.indexOf(l)===-1)continue;for(var u=0;ui.maxNumTensors&&(i.maxNumTensors=w),w0,function(){return"Expected at least one fetch, got none"});var e=[],i={};if(n.length===1){var r=Xy(n[0],t);e=r.sorted,i=r.recipientMap}else for(var a=new Set,s=0,o=n;s0;){var c=l[l.length-1];if(e.has(c.name)){l.pop();continue}var h=u[u.length-1]===l.length-1;if(c.inputs.length===0||h)l.pop(),i.push(c),e.add(c.name),h&&u.pop();else{u.push(l.length-1);for(var d=0,p=c.inputs;d1 nodes"),Bn(c===0,"input layer has >1 tensors"),i.inputLayers.push(l),i.inputLayersNodeIndices.push(u),i.inputLayersTensorIndices.push(c)}i.inputNames=[],i.outputNames=[],i.feedInputShapes=[],i.feedInputNames=[],i.feedOutputNames=[];for(var p=0;p=0;)pt.splice(pt.indexOf(zt),1);L.push(zt)},C=[],R=[],D=0,k=i.outputs;Don?1:0});for(var he=0,ye=le;he0)throw new M("Container instance unexpectedly contains _trainableWeights.The trainable weights of a Container are a union of the trainable weights of its consituent Layers. Its own _trainableWeights must remain an empty Array.");if(!this.trainable)return[];for(var e=[],i=0,r=this.layers;i0)throw new M(v.length+" of "+a+" weights are not set: "+(""+v))}Xh(d)},t.prototype.updatedConfig=function(){var e=this.getConfig(),i={};return i.className=this.getClassName(),i.config=e,i.kerasVersion="tfjs-layers "+rd,i.backend="TensorFlow.js",i},t.prototype.toJSON=function(e,i){i===void 0&&(i=!0);var r=id(this.updatedConfig());return i?JSON.stringify(r):r},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=Xe(e);for(var a=new na,s=0;s1)for(var c=0,h=u;c0){for(var m=[],g=0;g0&&G.apply(Ht(X),ee)}function c(G){var Z=G.name,X=Sn(G,i.customObjects!=null?i.customObjects:{});X.setFastWeightInitDuringBuild(a),s[Z]=X;var ee=G.inboundNodes;ee.forEach(function(ne){if(!(ne instanceof Array))throw new M("Corrupted configuration, expected array for nodeData: "+ne);l(X,ne)})}for(var h=i.name,d=i.layers,p=0,f=d;p0&&typeof n[Object.keys(n)[0]]=="object"){var r=[];return t.forEach(function(a){a in n?r.push(n[a]):r.push(null)}),r}else throw new Error("The model has multiple ("+i+") outputs, "+("so "+e+" must be either an array with ")+(i+" elements or an object with "+t+" keys. ")+("Provided "+e+" not understood: "+JSON.stringify(n)))}function Jy(n,t){return u4(n,t,"classWeight")}function Zy(n,t,e,i){return Se(this,void 0,void 0,function(){var r,a,s,o,l;return ve(this,function(u){switch(u.label){case 0:if(t!=null||i!=null)throw new Error("Support sampleWeight is not implemented yet");return e!=null?(r=y.tidy(function(){if(n.shape.length===1)return n.clone();if(n.shape.length===2)if(n.shape[1]>1){var c=1;return n.argMax(c)}else{if(n.shape[1]===1)return n.reshape([n.shape[0]]);throw new Error("Encountered unexpected last-dimension size ("+n.shape[1]+") during handling of class weights. The size is expected to be >= 1.")}else throw new Error("Unexpected rank of target (y) tensor ("+n.rank+") during handling of class weights. The rank is expected to be 1 or 2.")}),o=(s=Array).from,[4,r.data()]):[3,2];case 1:return a=o.apply(s,[u.sent()]),y.dispose(r),l=[],a.forEach(function(c){if(e[c]==null)throw new Error("classWeight must contain all classes in the training data. "+("The class "+c+" exists in the data but not in ")+"classWeight");l.push(e[c])}),[2,y.tensor1d(l,"float32")];case 2:return[2,null]}})})}function c4(n,t){return y.mul(n,t)}var h4=32;function eb(n,t){var e,i,r=t;e=r.xs,i=r.ys,y.util.assert(e!=null&&i!=null,function(){return"A Dataset iterator for fitDataset() is expected to generate objects of the form `{xs: xVal, ys: yVal}`, where the two values may be `tf.Tensor`, an array of Tensors, or a map of string to Tensor. The provided Dataset instead generates "+(""+t)});var a=Qy("input",n.inputNames,e),s=Qy("output",n.outputNames,i),o=a[0].shape[0];y.util.assert(a.length===n.inputs.length,function(){return"LayersModel has "+n.inputs.length+" inputs, but the dataset "+("provides "+a.length+" inputs. (Expected input keys: ")+(JSON.stringify(n.inputNames)+")")}),y.util.assert(s.length===n.outputs.length,function(){return"LayersModel has "+n.outputs.length+" outputs, but the dataset "+("provides "+s.length+" outputs. (Expected output keys: ")+(JSON.stringify(n.outputNames)+")")});for(var l=function(d){y.util.assert(a[d].shape[0]===o,function(){return"Batch size mismatch: input "+(n.inputNames[d]+" has "+a[d].shape[0]+"; ")+("expected "+o+" based on input "+n.inputNames[0]+".")})},u=0;u0&&Number.isInteger(e.epochs),function(){return"For fitDataset(), config.epochs is expected to be a positive "+("integer, but got "+e.epochs)}),y.util.assert(!i||e.batchesPerEpoch>0&&Number.isInteger(e.batchesPerEpoch),function(){return"For fitDataset(), config.batchesPerEpoch is expected to be a "+("positive integer if specified, but got "+e.batchesPerEpoch)}),y.util.assert(e.validationSplit==null,function(){return"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."}),n.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");n.isTraining=!0,X.label=1;case 1:return X.trys.push([1,,26,27]),r=e.validationData!=null,a=void 0,s=void 0,r&&(tb(e.validationData)?y.util.assert(e.validationBatches==null||e.validationBatches>0&&Number.isInteger(e.validationBatches),function(){return"For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, "+("but got "+e.validationBatches)}):(o=d4(e.validationData),a=o.xs,s=o.ys)),l=n.makeTrainFunction(),u=n.getDedupedMetricsNames(),c=void 0,r?c=u.slice().concat(u.map(function(ee){return"val_"+ee})):c=u.slice(),h=zy(e.callbacks,e.yieldEvery),d=e.verbose==null?1:e.verbose,p=_y(h,d,e.epochs,null,null,p4(t,e),null,r,c),f=p.callbackList,m=p.history,f.setModel(n),n.history=m,[4,f.onTrainBegin()];case 2:return X.sent(),n.stopTraining_=!1,g=e.initialEpoch==null?0:e.initialEpoch,[4,t.iterator()];case 3:v=X.sent(),X.label=4;case 4:return g=e.batchesPerEpoch:L.done)?r?(G=void 0,tb(e.validationData)?(Z=Xe,[4,n.evaluateDataset(e.validationData,{batches:e.validationBatches})]):[3,17]):[3,19]:[3,20];case 16:return G=Z.apply(void 0,[X.sent()]),[3,18];case 17:G=Xe(n.evaluate(a,s,{batchSize:e.validationBatchSize==null?h4:e.validationBatchSize,verbose:0})),X.label=18;case 18:for(F=0;F0)throw new Te("Verbose mode is not implemented yet.");return y.util.assert(!i||e.batches>0&&Number.isInteger(e.batches),function(){return"Test loop expects `batches` to be a positive integer, but "+("received "+JSON.stringify(e.batches))}),m4(t)?(o=t,[3,3]):[3,1];case 1:return[4,t.iterator()];case 2:o=f.sent(),f.label=3;case 3:s=o,l=0,u=0,c=function(){var m;return ve(this,function(g){switch(g.label){case 0:return[4,s.next()];case 1:return m=g.sent(),a=y.tidy(function(){if(m.value){var v=eb(n,m.value),b=v.xs,w=v.ys,S=b.concat(w),L=y.tidy(function(){return r(S)});if(y.dispose(S),u===0)for(var N=0;N0&&y.dispose(W)},N=0;N0&&Number.isInteger(n),function(){return"batchSize is required to be a positive integer, but got "+n})}function ss(n,t,e){return n==null?[null]:Array.isArray(n)?n.map(function(i){return lr(i,t,e-t)}):lr(n,t,e-t)}function od(n,t){return y.tidy(function(){return n==null?null:Array.isArray(n)?n.map(function(e){return od(e,t)}):vy(n,t.dtype==="int32"?t:t.toInt())})}function ld(n,t){for(var e=[],i=0,r=null;i=n&&(r=n),e.push([i,r]),i=r;return e}function v4(n,t,e,i,r,a,s,o,l,u,c,h,d,p,f){return Se(this,void 0,void 0,function(){var m,g,v,b,w,S,L,N,C;return ve(this,function(R){switch(R.label){case 0:if(r==null&&(r=32),a==null&&(a=1),c==null&&(c=!0),d==null&&(d=0),m=!1,l!=null&&u!=null&&(m=!0),f!=null&&(m=!0,p==null))throw new M("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");return g=n.checkNumSamples(e,r,p,"steps_per_epoch"),g!=null&&(v=bn(0,g)),s==null&&(s=1),b=_y(o,s,a,d,g,p,r,m,h),w=b.callbackList,S=b.history,w.setModel(n),n.history=S,[4,w.onTrainBegin()];case 1:R.sent(),n.stopTraining_=!1,L=function(D){var k,W,F,P,H,_;return ve(this,function(K){switch(K.label){case 0:return[4,w.onEpochBegin(D)];case 1:if(K.sent(),k={},!(p!=null))return[3,2];throw new Te("stepsPerEpoch mode is not implemented yet.");case 2:if(c==="batch")throw new Te("batch shuffling is not implemneted yet");c&&y.util.shuffle(v),W=y.tensor1d(v),F=ld(g,r),P=function(j){var q;return ve(this,function(G){switch(G.label){case 0:return q={},[4,w.onBatchBegin(j,q)];case 1:return G.sent(),y.tidy(function(){var Z=F[j][0],X=F[j][1],ee=lr(W,Z,X-Z);q.batch=j,q.size=X-Z;for(var ne=od(e,ee),ie=t(ne),te=0;te0))return[3,4];if(f=!0,i.validationData.length===2)s=i.validationData[0],o=i.validationData[1];else throw i.validationData.length===3?new Te("validationData including sample weights is not supported yet."):new M("When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; "+(i.validationData+" is invalid."));return g=!0,[4,n.standardizeUserData(s,o,null,null,g,h)];case 3:return v=W.sent(),l=v[0],u=v[1],m=l.concat(u),[3,5];case 4:i.validationSplit!=null&&i.validationSplit>0&&i.validationSplit<1?(f=!0,b=Math.floor(r[0].shape[0]*(1-i.validationSplit)),w=r[0].shape[0],l=ss(r,b,w),r=ss(r,0,b),u=ss(a,b,w),a=ss(a,0,b),m=l.concat(u)):i.validationSteps!=null&&(f=!0),W.label=5;case 5:return S=r.concat(a).concat(c),n.checkTrainableWeightsConsistency(),L=n.makeTrainFunction(),N=n.getDedupedMetricsNames(),C=void 0,R=void 0,f?(n.makeTestFunction(),C=n.testFunction,R=N.slice().concat(N.map(function(F){return"val_"+F}))):(C=null,m=[],R=N.slice()),D=zy(i.callbacks,i.yieldEvery),[4,v4(n,L,S,N,h,i.epochs,i.verbose,D,C,m,i.shuffle,R,i.initialEpoch,null,null)];case 6:return k=W.sent(),[2,k];case 7:return n.isTraining=!1,cr(r,t),cr(a,e),cr(l,s),cr(u,o),c!=null&&y.dispose(c),[7];case 8:return[2]}})})}function nb(n){var t=[];n instanceof y.Tensor&&(n=[n]);for(var e=0;e0)a=!0;else if(ib(n)){for(var s in n)if(n.hasOwnProperty(s)){a=!0;break}}else a=!0;if(a)throw new M("Error when checking model "+r+" expected no data, "+("but got "+n))}return[]}if(n==null)return t.map(function(g){return null});var o;if(ib(n)){n=n,o=[];for(var l=0,u=t;l1)throw new M("The model "+r+" expects "+t.length+" Tensor(s), "+("but only received one Tensor. Found: Tensor with shape "+n.shape));o=[n]}if(o=nb(o),e!=null)for(var h=0;h=0&&f!==m)throw new M("Error when checking "+r+": expected "+t[h]+" "+("to have shape ["+e[h]+"], but got array with shape ")+("["+d.shape+"]."))}}return o}function w4(n,t,e){var i=Li(n.map(function(a){return a.shape[0]}));i.sort();var r=Li(t.map(function(a){return a.shape[0]}));if(r.sort(),i.length>1)throw new M("All input Tensors (x) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(n.map(function(a){return a.shape}))));if(r.length>1)throw new M("All target Tensors (y) should have the same number of samples. Got array shapes: "+(""+JSON.stringify(t.map(function(a){return a.shape}))));if(i.length>0&&r.length>0&&!y.util.arraysEqual(i,r))throw new M("Input Tensors should have the same number of samples as target "+("Tensors. Found "+i[0]+" input sample(s) and "+r[0]+" target ")+"sample(s).")}function S4(n,t,e){for(var i=[ur,Do,is],r=0;r1)throw new M("The model expects "+t.length+" "+r+" Tensors, but only received one Tensor. Found: array with shape "+(JSON.stringify(n.shape)+"."));a=[n]}if(e!=null)for(var s=0;s1&&(i.metricsTensors.push([b,v]),i.metricsNames.push(i.outputNames[v]+"_loss"))}});var m=L4(e.metrics,this.outputNames),g=function(v,b,w){i.outputNames.length>1&&(b=i.outputNames[v]+"_"+b),i.metricsNames.push(b),i.metricsTensors.push([w,v])};or("metric",function(){for(var v=function(w){if(f.indexOf(w)!==-1)return"continue";var S=m[w],L=function(N){for(var C="",R,D,k,W=function(_){if(typeof _=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(_)!==-1){var K=i.internalOutputShapes[w];K[K.length-1]===1||i.lossFunctions[w]===Do?["accuracy","acc"].indexOf(_)!==-1?D=Qh:["crossentropy","ce"].indexOf(_)!==-1&&(D=Vy):i.lossFunctions[w]===Eo?["accuracy","acc"].indexOf(_)!==-1?D=qy:["crossentropy","ce"].indexOf(_)!==-1&&(D=Gy):["accuracy","acc"].indexOf(_)!==-1?D=ed:["crossentropy","ce"].indexOf(_)!==-1&&(D=td);var j=void 0;["accuracy","acc"].indexOf(_)!==-1?j="acc":["crossentropy","ce"].indexOf(_)!==-1&&(j="ce"),k=D,R=C+j}else{var q=J3(_);k=q,R=C+Wo(_)}var G;or(R,function(){G=k}),g(w,R,G)},F=0,P=N;F0){var d=[];throw i.forEach(function(p,f){p==null&&d.push(e[f])}),new M("Cannot find SymbolicTensors for output name(s): "+(""+JSON.stringify(d)))}return i},t.prototype.predictLoop=function(e,i,r){var a=this;return i===void 0&&(i=32),r===void 0&&(r=!1),y.tidy(function(){var s=a.checkNumSamples(e);if(r)throw new Te("Verbose predictLoop() is not implemented yet.");for(var o=ld(s,i),l=a.outputs.map(function(h){return[]}),u=function(h){var d=y.tidy(function(){var p=o[h][0],f=o[h][1],m=ss(e,p,f),g=[];if(Array.isArray(m))for(var v=0;v0&&e[0].shape[0]%a!==0)throw new M("In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size "+(a+". Found: "+e[0].shape[0]+" sample(s)."));return[e,i]},t.prototype.standardizeUserData=function(e,i,r,a,s,o){return s===void 0&&(s=!0),Se(this,void 0,void 0,function(){var l,u,c,h,d,p,f,m;return ve(this,function(g){switch(g.label){case 0:if(l=this.standardizeUserDataXY(e,i,s,o),u=l[0],c=l[1],r!=null)throw new Error("sample weight is not supported yet.");if(h=null,!(a!=null))return[3,4];d=Jy(a,this.outputNames),h=[],p=0,g.label=1;case 1:return p0)throw new Te("Verbose mode is not implemented yet.");if(s!=null)throw new Te("steps mode in testLoop() is not implemented yet");for(var c=ld(l,r),h=y.tensor1d(bn(0,l)),d=0;d1){var o=ty(e.slice(0,r),a);s+="_"+o}i.push(s)}return i},t.prototype.makeTrainFunction=function(){var e=this;return function(i){var r=[],a=i.slice(0,e.inputs.length),s=i.slice(e.inputs.length,e.inputs.length+e.outputs.length),o=i.slice(e.inputs.length+e.outputs.length,e.inputs.length+e.outputs.length*2),l=[],u=function(){for(var p=[],f=0;f1&&f1)throw new M("Found more than one ("+r.length+") save handlers for "+("URL '"+e+"'"));e=r[0]}if(e.save==null)throw new M("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[4,y.io.encodeWeights(this.getNamedWeights(i))];case 1:return a=w.sent(),s=!1,o=null,l=this.toJSON(o,s),u={modelTopology:l,format:I4,generatedBy:"TensorFlow.js tfjs-layers v"+rd,convertedBy:null},c=i==null?!1:i.includeOptimizer,c&&this.optimizer!=null?(u.trainingConfig=this.getTrainingConfig(),h="optimizer",g=(m=y.io).encodeWeights,[4,this.optimizer.getWeights()]):[3,4];case 2:return[4,g.apply(m,[w.sent(),h])];case 3:d=w.sent(),p=d.data,f=d.specs,(b=a.specs).push.apply(b,f),a.data=y.io.concatenateArrayBuffers([a.data,p]),w.label=4;case 4:return this.userDefinedMetadata!=null&&(v=!0,Ky(this.userDefinedMetadata,this.name,v),u.userDefinedMetadata=this.userDefinedMetadata),u.weightData=a.data,u.weightSpecs=a.specs,[2,e.save(u)]}})})},t.prototype.setUserDefinedMetadata=function(e){Ky(e,this.name),this.userDefinedMetadata=e},t.prototype.getUserDefinedMetadata=function(){return this.userDefinedMetadata},t.className="Model",t}(l4);y.serialization.registerClass(Ni);var A4=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.className="Functional",t}(Ni);y.serialization.registerClass(A4);function T4(n,t){return Se(this,void 0,void 0,function(){var e,i,r,a,s,o,l,u;return ve(this,function(c){switch(c.label){case 0:return"modelTopology"in n||(n={modelTopology:n}),n=n,e=n.modelTopology,e.model_config!=null&&(e=e.model_config),i=rs(e),r=Sn(i,t),n.weightsManifest!=null?[4,y.io.loadWeights(n.weightsManifest,n.pathPrefix,r.weights.map(function(h){return h.originalName}))]:[3,2];case 1:for(a=c.sent(),s={},o=0,l=r.weights;o1)throw new M("Found more than one ("+e.length+") load handlers for "+("URL '"+n+"'"));n=e[0]}return[2,N4(n,void 0,t)]})})}function N4(n,t,e){return Se(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h;return ve(this,function(d){switch(d.label){case 0:if(e==null&&(e={}),n.load==null)throw new M("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,n.load()];case 1:if(i=d.sent(),r=i.modelTopology,r.model_config!=null&&(r=r.model_config),a=e.strict==null?!0:e.strict,s=i.weightData!=null&&i.weightSpecs!=null&&a,o=Sn(rs(r),t,s),l=i.trainingConfig,l!=null&&o.loadTrainingConfig(l),i.userDefinedMetadata!=null&&o.setUserDefinedMetadata(i.userDefinedMetadata),!(i.weightData!=null))return[3,4];if(i.weightSpecs==null)throw new M("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");return u=C4(i.weightData,i.weightSpecs),c=u.modelWeights,h=u.optimizerWeights,o.loadWeights(c,a),o.optimizer!=null&&h.length>0?[4,o.optimizer.setWeights(h)]:[3,3];case 2:d.sent(),d.label=3;case 3:y.dispose(c),y.dispose(h.map(function(p){return p.tensor})),d.label=4;case 4:return[2,o]}})})}function C4(n,t){var e=y.io.decodeWeights(n,t),i={},r=[];return t.forEach(function(a){a.group==="optimizer"?r.push({name:a.name,tensor:e[a.name]}):i[a.name]=e[a.name]}),{modelWeights:i,optimizerWeights:r}}var cd=function(n){Q(t,n);function t(e){var i=n.call(this,{inputs:[],outputs:[]})||this;if(e=e||{},i.trainable=!0,i.built=!1,i.name=e.name!=null?e.name:To("sequential_"),e.layers!=null)for(var r=0,a=e.layers;r 0 "+("but got "+JSON.stringify(e.filters)))},t}(bb),fd=function(n){Q(t,n);function t(e){var i=n.call(this,2,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!Wh(e.kernelSize,"number",1,2))throw new M("Conv2D expects config.kernelSize to be number or number[] with "+("length 1 or 2, but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv2D",t}(Po);y.serialization.registerClass(fd);var wb=function(n){Q(t,n);function t(e){var i=n.call(this,3,e)||this;return t.verifyArgs(e),i}return t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.rank,e},t.verifyArgs=function(e){if(typeof e.kernelSize!="number"&&!(Array.isArray(e.kernelSize)&&(e.kernelSize.length===1||e.kernelSize.length===3)))throw new M("Conv3D expects config.kernelSize to be number or"+(" [number, number, number], but received "+JSON.stringify(e.kernelSize)+"."))},t.className="Conv3D",t}(Po);y.serialization.registerClass(wb);var Sb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;if(i.inputSpec=[new Tt({ndim:4})],i.padding!=="same"&&i.padding!=="valid")throw new M("Conv2DTranspose currently supports only padding modes 'same' "+("and 'valid', but received padding mode "+i.padding));return i}return t.prototype.build=function(e){var i;if(e=Ye(e),e.length!==4)throw new M("Input should have rank 4; Received input shape: "+JSON.stringify(e));var r=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[r]==null)throw new M("The channel dimension of the inputs should be defined. Found `None`.");var a=e[r],s=this.kernelSize.concat([this.filters,a]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Tt({ndim:4,axes:(i={},i[r]=a,i)})],this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);if(a.shape.length!==4)throw new M("Conv2DTranspose.call() expects input tensor to be rank-4, but "+("received a tensor of rank-"+a.shape.length));var s=a.shape,o=s[0],l,u;r.dataFormat==="channelsFirst"?(l=2,u=3):(l=1,u=2);var c=s[l],h=s[u],d=r.kernelSize[0],p=r.kernelSize[1],f=r.strides[0],m=r.strides[1],g=zo(c,f,d,r.padding),v=zo(h,m,p,r.padding),b=[o,g,v,r.filters];r.dataFormat!=="channelsLast"&&(a=y.transpose(a,[0,2,3,1]));var w=y.conv2dTranspose(a,r.kernel.read(),b,r.strides,r.padding);return r.dataFormat!=="channelsLast"&&(w=y.transpose(w,[0,3,1,2])),r.bias!=null&&(w=Pn(w,r.bias.read(),r.dataFormat)),r.activation!=null&&(w=r.activation.apply(w)),w})},t.prototype.computeOutputShape=function(e){e=Ye(e);var i=e.slice(),r,a,s;this.dataFormat==="channelsFirst"?(r=1,a=2,s=3):(r=3,a=1,s=2);var o=this.kernelSize[0],l=this.kernelSize[1],u=this.strides[0],c=this.strides[1];return i[r]=this.filters,i[a]=zo(i[a],u,o,this.padding),i[s]=zo(i[s],c,l,this.padding),i},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this);return delete e.dilationRate,e},t.className="Conv2DTranspose",t}(fd);y.serialization.registerClass(Sb);var $4=function(n){Q(t,n);function t(e,i){var r=n.call(this,e,i)||this;if(r.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",r.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",r.depthwiseKernel=null,r.pointwiseKernel=null,i.filters==null)throw new M("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(i.kernelInitializer!=null||i.kernelRegularizer!=null||i.kernelConstraint!=null)throw new M("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(i.padding!=null&&i.padding!=="same"&&i.padding!=="valid")throw new M("SeparableConv"+r.rank+"D supports only padding modes: "+("'same' and 'valid', but received "+JSON.stringify(i.padding)));return r.depthMultiplier=i.depthMultiplier==null?1:i.depthMultiplier,r.depthwiseInitializer=et(i.depthwiseInitializer||r.DEFAULT_DEPTHWISE_INITIALIZER),r.depthwiseRegularizer=tt(i.depthwiseRegularizer),r.depthwiseConstraint=bt(i.depthwiseConstraint),r.pointwiseInitializer=et(i.depthwiseInitializer||r.DEFAULT_POINTWISE_INITIALIZER),r.pointwiseRegularizer=tt(i.pointwiseRegularizer),r.pointwiseConstraint=bt(i.pointwiseConstraint),r}return t.prototype.build=function(e){var i;if(e=Ye(e),e.length1&&(t=n.slice(1,n.length)),n=n[0]}function r(a){return a==null||Array.isArray(a)?a:[a]}return t=r(t),e=r(e),{inputs:n,initialState:t,constants:e}}function Cb(n,t,e,i,r,a,s,o){return i===void 0&&(i=!1),s===void 0&&(s=!1),o===void 0&&(o=!1),y.tidy(function(){var l=t.shape.length;if(l<3)throw new M("Input should be at least 3D, but is "+l+"D.");var u=[1,0].concat(bn(2,l));if(t=y.transpose(t,u),a!=null)throw new Te("The rnn() functoin of the deeplearn.js backend does not support constants yet.");s&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=r.asType("bool").asType("float32"),r.rank===l-1&&(r=y.expandDims(r,-1)),r=y.transpose(r,u)),i&&(t=y.reverse(t,0),r!=null&&(r=y.reverse(r,0)));var c=[],h,d=e,p=t.shape[0],f=y.unstack(t),m;r!=null&&(m=y.unstack(r));for(var g=function(S){var L=f[S],N=y.tidy(function(){return n(L,d)});if(r==null)h=N[0],d=N[1];else{var C=y.tidy(function(){var R=m[S],D=y.onesLike(R).sub(R),k=N[0].mul(R).add(d[0].mul(D)),W=d.map(function(F,P){return N[1][P].mul(R).add(F.mul(D))});return{output:k,newStates:W}});h=C.output,d=C.newStates}o&&c.push(h)},v=0;v1?Bh(r,[1,a]):r}):i.cell.stateSize>1?[Bh(r,[1,i.cell.stateSize])]:[r]})},Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.trainable?this.cell.trainableWeights:[]},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights},enumerable:!0,configurable:!0}),t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(e)},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(i.numConstants=this.numConstants);var r=this.cell.getConfig();return this.getClassName()===t.className&&(i.cell={className:this.cell.getClassName(),config:r}),Yt({},r,e,i)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.cell,s=Sn(a,r);return new e(Object.assign(i,{cell:s}))},t.className="RNN",t}(De);y.serialization.registerClass(Ri);var ra=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t}(De),gd=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.DEFAULT_ACTIVATION="tanh",i.DEFAULT_KERNEL_INITIALIZER="glorotNormal",i.DEFAULT_RECURRENT_INITIALIZER="orthogonal",i.DEFAULT_BIAS_INITIALIZER="zeros",i.units=e.units,At(i.units,"units"),i.activation=Ci(e.activation==null?i.DEFAULT_ACTIVATION:e.activation),i.useBias=e.useBias==null?!0:e.useBias,i.kernelInitializer=et(e.kernelInitializer||i.DEFAULT_KERNEL_INITIALIZER),i.recurrentInitializer=et(e.recurrentInitializer||i.DEFAULT_RECURRENT_INITIALIZER),i.biasInitializer=et(e.biasInitializer||i.DEFAULT_BIAS_INITIALIZER),i.kernelRegularizer=tt(e.kernelRegularizer),i.recurrentRegularizer=tt(e.recurrentRegularizer),i.biasRegularizer=tt(e.biasRegularizer),i.kernelConstraint=bt(e.kernelConstraint),i.recurrentConstraint=bt(e.recurrentConstraint),i.biasConstraint=bt(e.biasConstraint),i.dropout=Qr([1,Ai([0,e.dropout==null?0:e.dropout])]),i.recurrentDropout=Qr([1,Ai([0,e.recurrentDropout==null?0:e.recurrentDropout])]),i.stateSize=i.units,i.dropoutMask=null,i.recurrentDropoutMask=null,i}return t.prototype.build=function(e){e=Ye(e),this.kernel=this.addWeight("kernel",[e[e.length-1],this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(e=e,e.length!==2)throw new M("SimpleRNNCell expects 2 input Tensors, got "+e.length+".");var a=e[1];e=e[0];var s=i.training==null?!1:i.training;01){for(var s=[0],o=2;o1)throw new M("Can not merge tensors with different batch sizes. "+("Got tensors with shapes: "+JSON.stringify(e)+"."));for(var o=e[0]==null?null:e[0].slice(1),l=1;l1){var S=bn(1,h).concat([0]);a.push(y.transpose(c,S)),p=!0}else a.push(c)}var L=r.mergeFunction(a),N=L.rank;if(p){if(N==null){var C=L.shape,R=C.length,v=C[R-1],b=[v].concat(C.slice(0,C.length-1));L=y.transpose(L.reshape([-1,v]),[1,0]).reshape(b)}else if(N>1){var S=[N-1].concat(bn(0,N-1));L=y.transpose(L,S)}}return L}}else return r.mergeFunction(e)})},t.prototype.computeOutputShape=function(e){e=e;var i;e[0]==null?i=null:i=e[0].slice(1);for(var r=1;r1)throw new M("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))},t.prototype.mergeFunction=function(e){var i=this;return y.tidy(function(){return Ph(e,i.axis)})},t.prototype.computeOutputShape=function(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new M("A `Concatenate` layer should be called on a list of inputs.");for(var i=e,r=i[0].slice(),a=this.axis<0?r.length+this.axis:this.axis,s=0,o=i.slice(1);s3||t.shape.length>3)throw new Te("batchDot is not implemented for tensors of 4D or higher rank yet");if(y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of x to be >= 2, "+("but got "+n.shape.length)}),y.util.assert(n.shape.length>=2,function(){return"batchDot requires the rank of y to be >= 2, "+("but got "+t.shape.length)}),typeof e=="number"&&(e=[e,e]),n.dtype==="complex64"||t.dtype==="complex64")throw new Te("batchDot is not implemented for complex64-type Tensors yet.");var i=n.shape.length,r=t.shape.length;e==null&&(e=[i-1,r-2]);var a=e;return y.tidy(function(){var s;if(i>r){s=i-r;for(var o=[],l=0;li){s=r-i;for(var o=[],l=0;l0){var d=void 0;i>r?d=i+r-3:d=i-1;for(var p=[],l=d;l3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);if(i[a[0]]!==r[a[1]])throw new M("Dimension incompatibility: "+(i[a[0]]+" !== "+r[a[1]]))},t.prototype.mergeFunction=function(e){if(e.length!==2)throw new M("A `Dot` layer must be called on exactly 2 inputs, "+("but received "+e.length+" input(s)."));var i=e[0],r=e[1],a;return Array.isArray(this.axes)?a=this.axes.map(function(s,o){return os(s,e[o].shape.length)}):a=[os(this.axes,i.shape.length),os(this.axes,r.shape.length)],this.normalize&&(i=Ro(i,a[0]),r=Ro(r,a[1])),Z4(i,r,a)},t.prototype.interpretAxes=function(e,i){var r;return Array.isArray(this.axes)?r=this.axes:r=[os(this.axes,e.length),os(this.axes,i.length)],r},t.prototype.computeOutputShape=function(e){y.util.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),function(){return"A `Dot` layer should be called on a list of exactly 2 inputs."});var i=e[0].slice(),r=e[1].slice();if(i.length>3||r.length>3)throw new Te("Dot layer does not support tensors of 4D or higher rank yet.");var a=this.interpretAxes(i,r);i.splice(a[0],1),r.splice(a[1],1),r.splice(0,1);var s=i.concat(r);return s.length===1&&s.push(1),s},t.prototype.computeMask=function(e,i){return null},t.prototype.getConfig=function(){var e={axes:this.axes,normalize:this.normalize},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="Dot",t}(hr);y.serialization.registerClass(jb);var $b=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.stddev=e.stddev,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={stddev:this.stddev};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e),s=function(){return Io(a.shape,0,r.stddev).add(a)},o=ts(s,function(){return a},i.training||!1);return o})},t.className="GaussianNoise",t}(De);y.serialization.registerClass($b);var Xb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i}return t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i);var a=xe(e);if(r.rate>0&&r.rate<1){var s=function(){var o=Math.sqrt(r.rate/(1-r.rate));return a.mul(Io(a.shape,1,o))};return ts(s,function(){return a},i.training||!1)}return a})},t.className="GaussianDropout",t}(De);y.serialization.registerClass(Xb);var Jb=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i.rate=e.rate,i.noiseShape=e.noiseShape,i}return t.prototype._getNoiseShape=function(e){return this.noiseShape||xe(e).shape},t.prototype.computeOutputShape=function(e){return e},t.prototype.getConfig=function(){var e=n.prototype.getConfig.call(this),i={rate:this.rate};return Object.assign(i,e),i},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){if(r.rate<1&&r.rate>0){var a=r._getNoiseShape(e),s=function(){var o=xe(e),l=1.6732632423543772,u=1.0507009873554805,c=-l*u,h=y.greaterEqual(y.randomUniform(a),r.rate);h=Za(h,"float32");var d=Math.pow((1-r.rate)*(1+r.rate*Math.pow(c,2)),-.5),p=-d*c*r.rate,f=o.mul(h).add(h.add(-1).mul(c));return f.mul(d).add(p)};return ts(s,function(){return xe(e)},i.training||!1)}return e})},t.className="AlphaDropout",t}(De);y.serialization.registerClass(Jb);function ls(n,t,e,i,r,a){a===void 0&&(a=.001);var s;if(n.rank===2)s=y.batchNorm2d(n,t,e,i,r,a);else if(n.rank===3)s=y.batchNorm3d(n,t,e,i,r,a);else if(n.rank===4)s=y.batchNorm4d(n,t,e,i,r,a);else throw new Te("batchNormalization is not implemented for array of rank "+n.rank+" yet");return s}function Q4(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){var a=y.moments(n,i),s=a.mean,o=a.variance,l=ls(n,s,o,e,t,r);return[l,s,o]})}function eF(n,t,e,i,r){return r===void 0&&(r=.001),y.tidy(function(){for(var a=y.moments(n,i),s=a.mean,o=a.variance,l=[],u=0,c=bn(0,n.rank);u=0?this.axis:this.axis+e.length,a=e[r];if(a==null)throw new M("Axis "+r+" of input tensor should have a defined dimension but the layer received an input with shape "+(JSON.stringify(e)+"."));this.inputSpec=[new Tt({ndim:e.length,axes:(i={},i[r]=a,i)})];var s=[a];this.scale&&(this.gamma=this.addWeight("gamma",s,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",s,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",s,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",s,null,this.movingVarianceInitializer,null,!1),this.built=!0},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=i.training==null?!1:i.training,s=xe(e),o=s.shape,l=o.length,u=bn(0,l),c=r.axis>=0?r.axis:r.axis+l;u.splice(c,1);var h=ar(1,l);h[c]=o[c];var d=u.slice();d.sort();var p=!y.util.arraysEqual(d,bn(0,l).slice(0,l-1)),f=function(){if(p){var L=r.movingMean.read().reshape(h),N=r.movingVariance.read().reshape(h),C=r.center?r.beta.read().reshape(h):null,R=r.scale?r.gamma.read().reshape(h):null;return ls(s,L,N,C,R,r.epsilon)}else return ls(s,r.movingMean.read(),r.movingVariance.read(),r.beta==null?null:r.beta.read(),r.gamma==null?null:r.gamma.read(),r.epsilon)};if(!a)return f();var m=tF(s,r.gamma.read(),r.beta.read(),u,r.epsilon),g=m[0],v=m[1],b=m[2],w=function(L,N,C){y.tidy(function(){var R=1-C,D=L.read(),k=D.sub(N).mul(R);L.write(D.sub(k))})},S=function(){w(r.movingMean,v,r.momentum),w(r.movingVariance,b,r.momentum)};return S(),g})},t.prototype.getConfig=function(){var e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:at(this.betaInitializer),gammaInitializer:at(this.gammaInitializer),movingMeanInitializer:at(this.movingMeanInitializer),movingVarianceInitializer:at(this.movingVarianceInitializer),betaRegularizer:Ke(this.betaRegularizer),gammaRegularizer:Ke(this.gammaRegularizer),betaConstraint:yt(this.betaConstraint),gammaConstraint:yt(this.gammaConstraint)},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="BatchNormalization",t}(De);y.serialization.registerClass(Zb);var Qb=function(n){Q(t,n);function t(e){var i=this;if(e==null&&(e={}),i=n.call(this,e)||this,i.axis=e.axis==null?-1:e.axis,typeof i.axis=="number"){if(!Number.isInteger(i.axis))throw new Error("Expected axis to be an integer, but received "+i.axis)}else if(Array.isArray(i.axis))for(var r=0,a=i.axis;r=i)throw new Error("Invalid axis: "+o)}if(this.axis.length!==Li(this.axis).length)throw new Error("Found duplicate axes in: "+this.axis);var l=this.axis.map(function(c){return e[c]}),u=!0;this.scale?this.gamma=this.addWeight("gamma",l,"float32",this.gammaInitializer,this.gammaRegularizer,u):this.gamma=null,this.center?this.beta=this.addWeight("beta",l,"float32",this.betaInitializer,this.betaRegularizer,u):this.beta=null,this.built=!0},t.prototype.call=function(e,i){var r=this,a=xe(e),s=a.shape,o=s.length;return y.tidy(function(){for(var l=!0,u=y.moments(a,r.axis,l),c=u.mean,h=u.variance,d=ar(1,o),p=0,f=r.axis;p=0?i=e[2]+this.padding[0][0]+this.padding[0][1]:i=null,e[3]!=null&&e[3]>=0?r=e[3]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],e[1],i,r]):(e[1]!=null&&e[1]>=0?i=e[1]+this.padding[0][0]+this.padding[0][1]:i=null,e[2]!=null&&e[2]>=0?r=e[2]+this.padding[1][0]+this.padding[1][1]:r=null,[e[0],i,r,e[3]])},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return nF(xe(e),r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={padding:this.padding,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.className="ZeroPadding2D",t}(De);y.serialization.registerClass(ew);function Mo(n,t,e,i,r,a){return y.tidy(function(){ct(r),cy(a),rn(i),e==null&&(e=[1,1]),i==null&&(i="valid"),r==null&&(r=yn()),a==null&&(a="max"),n=pd(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool(n,t,e,o):s=y.avgPool(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,3,1,2])),s})}function tw(n,t,e,i,r,a){return y.tidy(function(){ct(r),cy(a),rn(i),e==null&&(e=[1,1,1]),i==null&&(i="valid"),r==null&&(r=yn()),a==null&&(a="max"),n=vb(n,r);var s,o=i==="same"?"same":"valid";return a==="max"?s=y.maxPool3d(n,t,e,o):s=y.avgPool3d(n,t,e,o),r==="channelsFirst"&&(s=y.transpose(s,[0,4,1,2,3])),s})}var nw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=2),i=n.call(this,e)||this,typeof e.poolSize=="number")i.poolSize=[e.poolSize];else if(Array.isArray(e.poolSize)&&e.poolSize.length===1&&typeof e.poolSize[0]=="number")i.poolSize=e.poolSize;else throw new M("poolSize for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.poolSize)));if(At(i.poolSize,"poolSize"),e.strides==null)i.strides=i.poolSize;else if(typeof e.strides=="number")i.strides=[e.strides];else if(Array.isArray(e.strides)&&e.strides.length===1&&typeof e.strides[0]=="number")i.strides=e.strides;else throw new M("strides for 1D convolutional layer must be a number or an Array of a single number, but received "+(""+JSON.stringify(e.strides)));return At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,rn(i.padding),i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){e=Ye(e);var i=Ln(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],i,e[2]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){r.invokeCallHook(e,i),e=Qa(xe(e),2);var a=r.poolingFunction(xe(e),[r.poolSize[0],1],[r.strides[0],1],r.padding,"channelsLast");return y.squeeze(a,[2])})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),iw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),rn(a),Mo(e,i,r,a,s,"max")},t.className="MaxPooling1D",t}(nw);y.serialization.registerClass(iw);var rw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),rn(a),Mo(e,i,r,a,s,"avg")},t.className="AveragePooling1D",t}(nw);y.serialization.registerClass(rw);var aw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==2)throw new M("If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),rn(i.padding),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){e=Ye(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2];return i=Ln(i,this.poolSize[0],this.padding,this.strides[0]),r=Ln(r,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r]:[e[0],i,r,e[3]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),sw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),rn(a),Mo(e,i,r,a,s,"max")},t.className="MaxPooling2D",t}(aw);y.serialization.registerClass(sw);var ow=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),rn(a),Mo(e,i,r,a,s,"avg")},t.className="AveragePooling2D",t}(aw);y.serialization.registerClass(ow);var lw=function(n){Q(t,n);function t(e){var i=this;if(e.poolSize==null&&(e.poolSize=[2,2,2]),i=n.call(this,e)||this,i.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize,e.poolSize],e.strides==null)i.strides=i.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==3)throw new M("If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length "+(e.strides.length+"."));i.strides=e.strides}else i.strides=[e.strides,e.strides,e.strides];return At(i.poolSize,"poolSize"),At(i.strides,"strides"),i.padding=e.padding==null?"valid":e.padding,i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),rn(i.padding),i.inputSpec=[new Tt({ndim:5})],i}return t.prototype.computeOutputShape=function(e){e=Ye(e);var i=this.dataFormat==="channelsFirst"?e[2]:e[1],r=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[4]:e[3];return i=Ln(i,this.poolSize[0],this.padding,this.strides[0]),r=Ln(r,this.poolSize[1],this.padding,this.strides[1]),a=Ln(a,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],i,r,a]:[e[0],i,r,a,e[4]]},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){return r.invokeCallHook(e,i),r.poolingFunction(xe(e),r.poolSize,r.strides,r.padding,r.dataFormat)})},t.prototype.getConfig=function(){var e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),uw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),rn(a),tw(e,i,r,a,s,"max")},t.className="MaxPooling3D",t}(lw);y.serialization.registerClass(uw);var cw=function(n){Q(t,n);function t(e){return n.call(this,e)||this}return t.prototype.poolingFunction=function(e,i,r,a,s){return ct(s),rn(a),tw(e,i,r,a,s,"avg")},t.className="AveragePooling3D",t}(lw);y.serialization.registerClass(cw);var hw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.inputSpec=[new Tt({ndim:3})],i}return t.prototype.computeOutputShape=function(e){return[e[0],e[2]]},t.prototype.call=function(e,i){throw new Te},t}(De),dw=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.mean(r,1)})},t.className="GlobalAveragePooling1D",t}(hw);y.serialization.registerClass(dw);var pw=function(n){Q(t,n);function t(e){return n.call(this,e||{})||this}return t.prototype.call=function(e,i){return y.tidy(function(){var r=xe(e);return y.max(r,1)})},t.className="GlobalMaxPooling1D",t}(hw);y.serialization.registerClass(pw);var fw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,ct(i.dataFormat),i.inputSpec=[new Tt({ndim:4})],i}return t.prototype.computeOutputShape=function(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]},t.prototype.call=function(e,i){throw new Te},t.prototype.getConfig=function(){var e={dataFormat:this.dataFormat},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t}(De),mw=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.mean(a,[1,2]):y.mean(a,[2,3])})},t.className="GlobalAveragePooling2D",t}(fw);y.serialization.registerClass(mw);var gw=function(n){Q(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.call=function(e,i){var r=this;return y.tidy(function(){var a=xe(e);return r.dataFormat==="channelsLast"?y.max(a,[1,2]):y.max(a,[2,3])})},t.className="GlobalMaxPooling2D",t}(fw);y.serialization.registerClass(gw);var vw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.layer=e.layer,i}return t.prototype.build=function(e){this.built=!0},Object.defineProperty(t.prototype,"trainable",{get:function(){return this.layer!=null?this.layer.trainable:!1},set:function(e){this.layer!=null&&(this.layer.trainable=e)},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"trainableWeights",{get:function(){return this.layer.trainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"nonTrainableWeights",{get:function(){return this.layer.nonTrainableWeights},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"updates",{get:function(){return this.layer._updates},enumerable:!0,configurable:!0}),Object.defineProperty(t.prototype,"losses",{get:function(){return this.layer.losses},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.layer.getWeights()},t.prototype.setWeights=function(e){this.layer.setWeights(e)},t.prototype.getConfig=function(){var e={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},i=n.prototype.getConfig.call(this);return Object.assign(e,i),e},t.prototype.setFastWeightInitDuringBuild=function(e){n.prototype.setFastWeightInitDuringBuild.call(this,e),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(e)},t.fromConfig=function(e,i,r){r===void 0&&(r={});var a=i.layer,s=Sn(a,r);delete i.layer;var o={layer:s};return Object.assign(o,i),new e(o)},t}(De),yw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this;return i.supportsMasking=!0,i}return t.prototype.build=function(e){if(e=Ye(e),e.length<3)throw new M("TimeDistributed layer expects an input shape >= 3D, but received "+("input shape "+JSON.stringify(e)));this.inputSpec=[{shape:e}];var i=[e[0]].concat(e.slice(2));this.layer.built||(this.layer.build(i),this.layer.built=!0),n.prototype.build.call(this,e)},t.prototype.computeOutputShape=function(e){e=Ye(e);var i=[e[0]].concat(e.slice(2)),r=this.layer.computeOutputShape(i),a=e[1];return[r[0],a].concat(r.slice(1))},t.prototype.call=function(e,i){var r=this;return y.tidy(function(){e=xe(e);var a=function(l,u){var c=xe(r.layer.call(l,i));return[c,[]]},s=Cb(a,e,[],!1,null,null,!1,!0),o=s[1];return o})},t.className="TimeDistributed",t}(vw);y.serialization.registerClass(yw);function iF(n){Jr(jk,"BidirectionalMergeMode",n)}var rF="concat",bw=function(n){Q(t,n);function t(e){var i=n.call(this,e)||this,r=e.layer.getConfig(),a={};a.className=e.layer.getClassName(),a.config=r,i.forwardLayer=Sn(a),r.goBackwards=!(r.goBackwards===!0);var s={};if(s.className=e.layer.getClassName(),s.config=r,i.backwardLayer=Sn(s),i.forwardLayer.name="forward_"+i.forwardLayer.name,i.backwardLayer.name="backward_"+i.backwardLayer.name,i.mergeMode=e.mergeMode===void 0?rF:e.mergeMode,iF(i.mergeMode),e.weights)throw new Te("weights support is not implemented for Bidirectional layer yet.");return i._stateful=e.layer.stateful,i.returnSequences=e.layer.returnSequences,i.returnState=e.layer.returnState,i.supportsMasking=!0,i._trainable=!0,i.inputSpec=e.layer.inputSpec,i.numConstants=null,i}return Object.defineProperty(t.prototype,"trainable",{get:function(){return this._trainable},set:function(e){this._trainable=e,this.forwardLayer!=null&&(this.forwardLayer.trainable=e),this.backwardLayer!=null&&(this.backwardLayer.trainable=e)},enumerable:!0,configurable:!0}),t.prototype.getWeights=function(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())},t.prototype.setWeights=function(e){var i=e.length,r=Math.floor(i/2);this.forwardLayer.setWeights(e.slice(0,r)),this.backwardLayer.setWeights(e.slice(r))},t.prototype.computeOutputShape=function(e){var i=this.forwardLayer.computeOutputShape(e);Array.isArray(i)&&Array.isArray(i[0])||(i=[i]),i=i;var r,a,s;return this.returnState&&(s=i.slice(1)),r=i[0],r=r,this.mergeMode==="concat"?(r[r.length-1]*=2,a=[r]):this.mergeMode==null?a=[r,r.slice()]:a=[r],this.returnState?this.mergeMode==null?a.concat(s).concat(s.slice()):[r].concat(s).concat(s.slice()):Ht(a)},t.prototype.apply=function(e,i){var r=i==null?null:i.initialState,a=i==null?null:i.constants;i==null&&(i={});var s=xb(e,r,a,this.numConstants);if(e=s.inputs,r=s.initialState,a=s.constants,Array.isArray(e)&&(r=e.slice(1),e=e[0]),(r==null||r.length===0)&&a==null)return n.prototype.apply.call(this,e,i);var o=[],l=[];if(r!=null){var u=r.length;if(u%2>0)throw new M("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");i.initialState=r,o.push.apply(o,r);var c=r.map(function(w){return new Tt({shape:w.shape})});this.forwardLayer.stateSpec=c.slice(0,u/2),this.backwardLayer.stateSpec=c.slice(u/2),l.push.apply(l,c)}if(a!=null)throw new Te("Support for constants in Bidirectional layers is not implemented yet.");for(var h=o[0]instanceof wn,d=0,p=o;dt}var Nw=function(n){Q(t,n);function t(e){var i=n.call(this)||this;if(e==null&&(e={}),e.restoreBestWeights)throw new Te("restoreBestWeights = True is not implemented in EarlyStopping yet.");return i.monitor=e.monitor||"val_loss",i.minDelta=Math.abs(e.minDelta||0),i.patience=e.patience||0,i.verbose=e.verbose||0,i.mode=e.mode||"auto",i.baseline=e.baseline,["auto","min","max"].indexOf(i.mode)===-1&&(console.warn("EarlyStopping mode '"+i.mode+"' is invalid. Falling back to mode 'auto'."),i.mode="auto"),i.mode==="min"?i.monitorFunc=Ho:i.mode==="max"||i.monitor.indexOf("acc")!==-1?i.monitorFunc=Tw:i.monitorFunc=Ho,i.monitorFunc===Ho&&(i.minDelta*=-1),i}return t.prototype.onTrainBegin=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===Ho?Infinity:-Infinity,[2]})})},t.prototype.onEpochEnd=function(e,i){return Se(this,void 0,void 0,function(){var r;return ve(this,function(a){switch(a.label){case 0:return[4,Ti(i)];case 1:return a.sent(),r=this.getMonitorValue(i),r==null?[2]:(this.monitorFunc(r-this.minDelta,this.best)?(this.best=r,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=e,this.model.stopTraining=!0)),[2])}})})},t.prototype.onTrainEnd=function(e){return Se(this,void 0,void 0,function(){return ve(this,function(i){return this.stoppedEpoch>0&&this.verbose&&console.log("Epoch "+this.stoppedEpoch+": early stopping."),[2]})})},t.prototype.getMonitorValue=function(e){e==null&&(e={});var i=e[this.monitor];return i==null&&console.warn("Metric for EarlyStopping "+this.monitor+" is not available. "+("Available metrics are: "+Object.keys(e))),i},t}(Aw);function UW(n){return new Nw(n)}var BW={earlyStopping:UW};$e.Callback=Aw;$e.CallbackList=Wy;$e.CustomCallback=By;$e.EarlyStopping=Nw;$e.History=Uy;$e.InputSpec=Tt;$e.LayerVariable=Oy;$e.LayersModel=Ni;$e.RNN=Ri;$e.Sequential=cd;$e.SymbolicTensor=wn;$e.callbacks=BW;$e.constraints=qk;$e.initializers=A3;$e.input=sb;$e.layers=fW;$e.loadLayersModel=E4;$e.metrics=OW;$e.model=R4;$e.models=EW;$e.registerCallbackConstructor=D4;$e.regularizers=WW;$e.sequential=O4;$e.version_layers=rd});var Hw=Ae(dr=>{"use strict";Object.defineProperty(dr,"__esModule",{value:!0});var B=rr();var Cw=Object.assign||function(t){for(var e,i=1,r=arguments.length;i0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]0?Object.keys(h).forEach(function(g){var v=ii(g)[0],b=l[v];b&&(b.signatureKey=h[g],u.push(b))}):u=a;var f={};t.library!=null&&t.library.function!=null&&(f=t.library.function.reduce(function(g,v){return g[v.signature.name]=i.mapFunction(v),g},{}));var m={nodes:l,inputs:u,outputs:c,weights:s,placeholders:a,signature:e,functions:f};return o.length>0&&(m.initNodes=o),m},n.prototype.mapSignatureEntries=function(t){return Object.keys(t||{}).reduce(function(e,i){return e[t[i].name]=i,e},{})},n.prototype.mapNode=function(t){var e=Ow(t.op)||this.opMappers[t.op]||{};t.attr==null&&(t.attr={});var i={name:t.name,op:t.op,category:e.category,inputNames:(t.input||[]).map(function(r){return r.startsWith("^")?r.substr(1):r}),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:t.attr};return e.inputs!=null&&(i.inputParams=e.inputs.reduce(function(r,a){return r[a.name]={type:a.type,inputIndexStart:a.start,inputIndexEnd:a.end},r},{})),e.attrs!=null&&(i.attrParams=e.attrs.reduce(function(r,a){var s=a.type,o=void 0;switch(a.type){case"string":o=Ad(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Ad(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"string[]":o=Dd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Dd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number":o=Nd(t.attr,a.tfName,a.defaultValue||0),o===void 0&&!!a.tfDeprecatedName&&(o=Nd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"number[]":o=Ed(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Ed(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool":o=Td(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Td(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"bool[]":o=Fd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Fd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape":o=Od(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Od(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"shape[]":o=kd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=kd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype":o=Cd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Cd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"dtype[]":o=Rd(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Rd(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"func":o=Ew(t.attr,a.tfName,a.defaultValue),o===void 0&&!!a.tfDeprecatedName&&(o=Ew(t.attr,a.tfDeprecatedName,a.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error("Unsupported param type: "+a.type+" for op: "+t.op)}return r[a.name]={value:o,type:s},r},{})),i},n.prototype.mapFunction=function(t){var e=this,i=t.nodeDef,r=[],a=[],s={};i!=null&&(s=i.reduce(function(d,p){return d[p.name]=e.mapNode(p),p.op==="Const"&&a.push(d[p.name]),d},{}));var o=[],l=[];t.signature.inputArg.forEach(function(d){var p=ii(d.name)[0],f={name:p,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:xd(d.type),type:"dtype"}},children:[]};f.signatureKey=d.name,o.push(f),s[p]=f});var u=Object.keys(s);u.forEach(function(d){var p=s[d];p.inputNames.forEach(function(f){var m=ii(f)[0];p.inputs.push(s[m]),s[m].children.push(p)})});var c=t.ret;t.signature.outputArg.forEach(function(d){var p=ii(c[d.name]),f=p[0],m=p[1],g=s[f];g!=null&&(g.defaultOutput=m,l.push(g))});var h=this.mapArgsToSignature(t);return{nodes:s,inputs:o,outputs:l,weights:a,placeholders:r,signature:h}},n.prototype.mapArgsToSignature=function(t){var e=this;return{methodName:t.signature.name,inputs:t.signature.inputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r),i},{}),outputs:t.signature.outputArg.reduce(function(i,r){return i[r.name]=e.mapArgToTensorInfo(r,t.ret),i},{})}},n.prototype.mapArgToTensorInfo=function(t,e){var i=t.name;return e!=null&&(i=e[i]),{name:i,dtype:t.type}},n}();function bU(n){var t=B.env().global;if(typeof t.atob!="undefined")return t.atob(n);if(typeof Buffer!="undefined")return new Buffer(n,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function kw(n,t){var e=Array.isArray(n)?String.fromCharCode.apply(null,n):bU(n);return t?e:e.toLowerCase()}function Ad(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r!=null?kw(r.s,i):e}function Td(n,t,e){var i=n[t];return i?i.b:e}function Nd(n,t,e){var i=n[t]||{},r=i.i!=null?i.i:i.f!=null?i.f:e;return typeof r=="number"?r:parseInt(r,10)}function xd(n){typeof n=="string"&&(n=An[n]);switch(n){case An.DT_FLOAT:return"float32";case An.DT_INT32:case An.DT_INT64:case An.DT_INT8:case An.DT_UINT8:return"int32";case An.DT_BOOL:return"bool";case An.DT_DOUBLE:return"float32";case An.DT_STRING:return"string";default:return null}}function Ew(n,t,e){var i=n[t];return i&&i.func?i.func.name:e}function Cd(n,t,e){var i=n[t];return i&&i.type?xd(i.type):e}function Rd(n,t,e){var i=n[t];return i&&i.list&&i.list.type?i.list.type.map(function(r){return xd(r)}):e}function Fw(n){return n.unknownRank?void 0:n.dim!=null?n.dim.map(function(t){return typeof t.size=="number"?t.size:parseInt(t.size,10)}):[]}function Od(n,t,e){var i=n[t];return i&&i.shape?Fw(i.shape):e}function Ed(n,t,e){var i=n[t];return i?((i.list.f&&i.list.f.length?i.list.f:i.list.i)||[]).map(function(r){return typeof r=="number"?r:parseInt(r,10)}):e}function Dd(n,t,e,i){i===void 0&&(i=!1);var r=n[t];return r&&r.list&&r.list.s?r.list.s.map(function(a){return kw(a,i)}):e}function kd(n,t,e){var i=n[t];return i&&i.list&&i.list.shape?i.list.shape.map(function(r){return Fw(r)}):e}function Fd(n,t,e){var i=n[t];return i&&i.list&&i.list.b?i.list.b:e}var wU=function(){function n(t,e,i){var r=this;this.node=t,this.tensorMap=e,this.context=i,this.inputs=[],this.attrs={},this.inputs=t.inputNames.map(function(a){return r.getInput(a)}),t.rawAttrs!=null&&(this.attrs=Object.keys(t.rawAttrs).reduce(function(a,s){return a[s]=r.getAttr(s),a},{}))}return n.prototype.getInput=function(t){return Vt(t,this.tensorMap,this.context)},n.prototype.getAttr=function(t,e){var i=this.node.rawAttrs[t];if(i.tensor!=null)return Vt(t,this.tensorMap,this.context);if(i.i!=null||i.f!=null)return Nd(this.node.rawAttrs,t,e);if(i.s!=null)return Ad(this.node.rawAttrs,t,e);if(i.b!=null)return Td(this.node.rawAttrs,t,e);if(i.shape!=null)return Od(this.node.rawAttrs,t,e);if(i.type!=null)return Cd(this.node.rawAttrs,t,e);if(i.list!=null){if(i.list.i!=null||i.list.f!=null)return Ed(this.node.rawAttrs,t,e);if(i.list.s!=null)return Dd(this.node.rawAttrs,t,e);if(i.list.shape!=null)return kd(this.node.rawAttrs,t,e);if(i.list.b!=null)return Fd(this.node.rawAttrs,t,e);if(i.list.type!=null)return Rd(this.node.rawAttrs,t,e)}return e},n}();var SU=function(n,t,e){switch(n.op){case"BiasAdd":case"AddV2":case"Add":return[B.add(A("a",n,t,e),A("b",n,t,e))];case"AddN":return[B.addN(A("tensors",n,t,e))];case"FloorMod":case"Mod":return[B.mod(A("a",n,t,e),A("b",n,t,e))];case"Mul":return[B.mul(A("a",n,t,e),A("b",n,t,e))];case"RealDiv":case"Div":return[B.div(A("a",n,t,e),A("b",n,t,e))];case"DivNoNan":return[B.divNoNan(A("a",n,t,e),A("b",n,t,e))];case"FloorDiv":return[B.floorDiv(A("a",n,t,e),A("b",n,t,e))];case"Sub":return[B.sub(A("a",n,t,e),A("b",n,t,e))];case"Minimum":return[B.minimum(A("a",n,t,e),A("b",n,t,e))];case"Maximum":return[B.maximum(A("a",n,t,e),A("b",n,t,e))];case"Pow":return[B.pow(A("a",n,t,e),A("b",n,t,e))];case"SquaredDifference":return[B.squaredDifference(A("a",n,t,e),A("b",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};var LU=function(n,t,e){switch(n.op){case"Abs":case"ComplexAbs":return[B.abs(A("x",n,t,e))];case"Acos":return[B.acos(A("x",n,t,e))];case"Acosh":return[B.acosh(A("x",n,t,e))];case"Asin":return[B.asin(A("x",n,t,e))];case"Asinh":return[B.asinh(A("x",n,t,e))];case"Atan":return[B.atan(A("x",n,t,e))];case"Atan2":return[B.atan2(A("x",n,t,e),A("y",n,t,e))];case"Atanh":return[B.atanh(A("x",n,t,e))];case"Ceil":return[B.ceil(A("x",n,t,e))];case"Complex":return[B.complex(A("real",n,t,e),A("imag",n,t,e))];case"Cos":return[B.cos(A("x",n,t,e))];case"Cosh":return[B.cosh(A("x",n,t,e))];case"Elu":return[B.elu(A("x",n,t,e))];case"Erf":return[B.erf(A("x",n,t,e))];case"Exp":return[B.exp(A("x",n,t,e))];case"Expm1":return[B.expm1(A("x",n,t,e))];case"Floor":return[B.floor(A("x",n,t,e))];case"Log":return[B.log(A("x",n,t,e))];case"Log1p":return[B.log1p(A("x",n,t,e))];case"Imag":return[B.imag(A("x",n,t,e))];case"Neg":return[B.neg(A("x",n,t,e))];case"Reciprocal":return[B.reciprocal(A("x",n,t,e))];case"Real":return[B.real(A("x",n,t,e))];case"Relu":return[B.relu(A("x",n,t,e))];case"Round":return[B.round(A("x",n,t,e))];case"Selu":return[B.selu(A("x",n,t,e))];case"Sigmoid":return[B.sigmoid(A("x",n,t,e))];case"Sin":return[B.sin(A("x",n,t,e))];case"Sign":return[B.sign(A("x",n,t,e))];case"Sinh":return[B.sinh(A("x",n,t,e))];case"Softplus":return[B.softplus(A("x",n,t,e))];case"Sqrt":return[B.sqrt(A("x",n,t,e))];case"Square":return[B.square(A("x",n,t,e))];case"Tanh":return[B.tanh(A("x",n,t,e))];case"Tan":return[B.tan(A("x",n,t,e))];case"Relu6":case"ClipByValue":return[B.clipByValue(A("x",n,t,e),A("clipValueMin",n,t,e),A("clipValueMax",n,t,e))];case"Rsqrt":return[B.rsqrt(Vt(n.inputNames[0],t,e))];case"Prod":return[B.prod(A("x",n,t,e),A("axes",n,t,e))];case"LeakyRelu":return[B.leakyRelu(A("x",n,t,e),A("alpha",n,t,e))];case"Prelu":return[B.prelu(A("x",n,t,e),A("alpha",n,t,e))];default:throw TypeError("Node type "+n.op+" is not implemented")}};function dn(n,t,e){e===void 0&&(e=""),B.util.assert(IU(n,t),function(){return e+(" Shapes "+n+" and "+t+" must match")})}function IU(n,t){if(n.length!==t.length)return!1;for(var e=0;e=this.size())throw new Error("Tried to read from index "+t+", but array size is: "+this.size());var e=this.tensors[t];if(e.cleared)throw new Error("TensorArray "+this.name+": Could not read index "+t+" twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).");return this.clearAfterRead&&(e.cleared=!0),e.read=!0,e.tensor},n.prototype.readMany=function(t){var e=this;return t.map(function(i){return e.read(i)})},n.prototype.write=function(t,e){if(this.closed_)throw new Error("TensorArray "+this.name+" has already been closed.");if(t<0||!this.dynamicSize&&t>=this.maxSize)throw new Error("Tried to write to index "+t+", but array is not resizeable and size is: "+this.maxSize);var i=this.tensors[t]||{};if(e.dtype!==this.dtype)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+`, + because the value dtype is `+e.dtype+", but TensorArray dtype is "+this.dtype+".");if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=e.shape),dn(this.elementShape,e.shape,"TensorArray "+this.name+": Could not write to TensorArray index "+t+"."),i.read)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been read.");if(i.written)throw new Error("TensorArray "+this.name+": Could not write to TensorArray index "+t+", because it has already been written.");i.tensor=e,B.keep(e),i.written=!0,this.tensors[t]=i},n.prototype.writeMany=function(t,e){var i=this;if(t.length!==e.length)throw new Error("TensorArray "+this.name+": could not write multiple tensors,"+("because the index size: "+t.length+" is not the same as tensors size: "+e.length+"."));t.forEach(function(r,a){return i.write(r,e[a])})},n.prototype.gather=function(t,e){if(!!e&&e!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but gather requested dtype "+e);if(t)t=t.slice(0,this.size());else{t=[];for(var i=0;i=this.maxSize)throw new Error("Max index must be < array size ("+i+" vs. "+this.maxSize+")");this.writeMany(t,B.unstack(e,0))},n.prototype.split=function(t,e){var i=this;if(e.dtype!==this.dtype)throw new Error("TensorArray dtype is "+this.dtype+" but tensor has dtype "+e.dtype);var r=0,a=t.map(function(c){return r+=c,r});if(r!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to + tensor.shape[0], but sum of lengths is + `+r+", and tensor's shape is: "+e.shape);if(!this.dynamicSize&&t.length!==this.maxSize)throw new Error("TensorArray's size is not equal to the size of lengths ("+this.maxSize+" vs. "+t.length+"), and the TensorArray is not marked as dynamically resizeable");var s=r===0?0:e.size/r,o=[];B.tidy(function(){e=B.reshape(e,[1,r,s]);for(var c=0;cthis.maxNumElements)throw new Error("TensorListResize input size "+t+" is greater maxNumElement "+this.maxNumElements+".");this.tensors.length=t},n.prototype.getItem=function(t,e,i){if(i!==this.elementDtype)throw new Error("Invalid data types; op elements "+i+", but list elements "+this.elementDtype);if(t<0||t>this.tensors.length)throw new Error("Trying to access element "+t+" in a list with "+this.tensors.length+" elements.");if(this.tensors[t]==null)throw new Error("element at index "+t+" is null.");return dn(this.tensors[t].shape,e,"TensorList shape mismatch: "),this.tensors[t]},n.prototype.setItem=function(t,e){if(e.dtype!==this.elementDtype)throw new Error("Invalid data types; op elements "+e.dtype+", but list elements "+this.elementDtype);if(t<0||this.maxNumElements!==-1&&t>=this.maxNumElements)throw new Error("Trying to set element "+t+" in a list with max "+this.maxNumElements+" elements.");dn(this.elementShape,e.shape,"TensorList shape mismatch: "),B.keep(e),this.tensors[t]=e},n.prototype.gather=function(t,e,i){var r=this;if(e!==this.elementDtype)throw new Error("Invalid data types; op elements "+e+", but list elements "+this.elementDtype);return dn(this.elementShape,i,"TensorList shape mismatch: "),t=t.slice(0,this.size()),t.length===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var a=t.map(function(s){return B.reshape(r.tensors[s],i)});return B.stack(a,0)})},n.prototype.concat=function(t,e){var i=this;if(!!t&&t!==this.elementDtype)throw new Error("TensorList dtype is "+this.elementDtype+" but concat requested dtype "+t);return dn(this.elementShape,e,"TensorList shape mismatch: "),this.size()===0?B.tensor([],[0].concat(this.elementShape)):B.tidy(function(){var r=i.tensors.map(function(a){return B.reshape(a,e)});return B.concat(r,0)})},n}();function TU(n,t,e){var i=n.dtype;if(n.shape.length<1)throw new Error("Tensor must be at least a vector, but saw shape: "+n.shape);if(n.dtype!==e)throw new Error("Invalid data types; op elements "+n.dtype+", but list elements "+e);var r=n.shape.slice(1);dn(r,t,"TensorList shape mismatch: ");var a=B.unstack(n);return new Go(a,t,i)}function NU(n,t,e){return new Go([],n,t,e)}function xU(n,t,e,i){if(t.length!==n.shape[0])throw new Error("Expected len(indices) == tensor.shape[0], but saw: "+t.length+" vs. "+n.shape[0]);var r=Math.max.apply(Math,t);if(i!=null&&i!==-1&&r>=i)throw new Error("Max index must be < array size ("+r+" vs. "+i+")");var a=new Go([],e,n.dtype,i),s=B.unstack(n,0);return t.forEach(function(o,l){a.setItem(o,s[l])}),a}function CU(n,t,e){var i=0,r=t.map(function(u){return i+=u,i});if(i!==n.shape[0])throw new Error(`Expected sum of lengths to be equal to + tensor.shape[0], but sum of lengths is + `+i+", and tensor's shape is: "+n.shape);for(var a=i===0?0:n.size/i,s=B.tidy(function(){var u=[];n=B.reshape(n,[1,i,a]);for(var c=0;c1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")},n.prototype.nextIteration=function(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;var t=Object.assign({},this.contexts[this.contexts.length-1]);t.iterationId+=1,t.id=this.lastId,this.contexts.splice(-1,1,t),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")},n.prototype.getWeight=function(t){return this.weightMap[t]},n.prototype.addTensorArray=function(t){this.tensorArrayMap[t.id]=t},n.prototype.getTensorArray=function(t){return this.tensorArrayMap[t]},n.prototype.addTensorList=function(t){this.tensorListMap[t.id]=t},n.prototype.getTensorList=function(t){return this.tensorListMap[t]},n.prototype.dispose=function(t){for(var e in this.tensorArrayMap)this.tensorArrayMap[e].clearAndClose(t);for(var e in this.tensorListMap)this.tensorListMap[e].clearAndClose(t)},n}();function Pw(n,t,e,i){var r=new Set,a=[],s=null,o=null,l=new Set,u=Object.keys(n).map(function(p){return $t(p)[0]}),c=[];i!=null&&(c=i.map(function(p){return $t(p.name)[0]}));for(var h=t.slice();h.length>0;){var d=h.pop();if((zw(d)||GU(d))&&(s==null&&(s=d,o=s.children.map(function(p){return p.name}).filter(function(p){return r.has(p)}))),r.add(d.name),e[d.name]!=null)continue;if(u.indexOf(d.name)!==-1)continue;if(c.indexOf(d.name)!==-1)continue;if(d.inputs.length===0){a.push(d.name);continue}d.inputs.forEach(function(p){if(l.has(p.name))return;l.add(p.name),h.push(p)})}return{inputs:n,outputs:t,usedNodes:r,missingInputs:a,dynamicNode:s,syncInputs:o}}function YU(n,t,e){var i=e.usedNodes,r=e.inputs,a=[],s=Object.keys(r).map(function(h){return $t(h)[0]}).map(function(h){return n.nodes[h]}),o=n.initNodes;s.forEach(function(h){i.has(h.name)&&a.push(h)}),n.weights.forEach(function(h){i.has(h.name)&&a.push(h)}),o!=null&&o.forEach(function(h){i.has(h.name)&&a.push(h)});for(var l=new Set,u=[];a.length>0;){var c=a.pop();l.add(c.name),t[c.name]||u.push(c),c.children.forEach(function(h){!l.has(h.name)&&i.has(h.name)&&h.inputs.every(function(d){return l.has(d.name)})&&a.push(h)})}return u}var KU=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],jU=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"];function zw(n){return KU.indexOf(n.op)>=0}function GU(n){return jU.indexOf(n.op)>=0}var _w=function(){function n(t,e){var i=this;this.graph=t,this.parent=e,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this._outputs=t.outputs,this._inputs=t.inputs,this._initNodes=t.initNodes,this._signature=t.signature,this._functions=t.functions,t.functions!=null&&Object.keys(t.functions).forEach(function(r){i._functionExecutorMap[r]=new n(t.functions[r],i)})}return Object.defineProperty(n.prototype,"weightIds",{get:function(){return this.parent?this.parent.weightIds:this._weightIds},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functionExecutorMap",{get:function(){return this.parent?this.parent.functionExecutorMap:this._functionExecutorMap},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weightMap",{get:function(){return this.parent?this.parent.weightMap:this._weightMap},set:function(t){var e=Object.keys(t).map(function(i){return t[i].map(function(r){return r.id})});this._weightIds=[].concat.apply([],e),this._weightMap=t},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this._inputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this._outputs.map(function(t){return{name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this._inputs.map(function(t){return t.signatureKey||t.name})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this._outputs.map(function(t){var e=t.signatureKey||t.name;return t.defaultOutput?e+":"+t.defaultOutput:e})},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"functions",{get:function(){var t=this;return Object.keys(this._functions).reduce(function(e,i){return e[i]=t._functions[i].signature,e},{})},enumerable:!0,configurable:!0}),n.prototype.getCompilationKey=function(t,e){var i=t.map(function(a){return a.name}).sort(),r=e.map(function(a){return a.name}).sort();return i.join(this.SEPERATOR)+"--"+r.join(this.SEPERATOR)},n.prototype.compile=function(t,e){var i=Pw(t,e,this.weightMap,this._initNodes),r=i.missingInputs,a=i.dynamicNode,s=i.syncInputs;if(a!=null)throw new Error("This execution contains the node '"+a.name+"', which has "+("the dynamic op '"+a.op+"'. Please use ")+"model.executeAsync() instead. Alternatively, to avoid the "+("dynamic ops, specify the inputs ["+s+"]"));if(r.length>0){var o=e.map(function(u){return u.name}),l=Object.keys(t);throw new Error("Cannot compute the outputs ["+o+"] from the provided inputs "+("["+l+"]. Missing the following inputs: ["+r+"]"))}return YU(this.graph,this.weightMap,i)},n.prototype.execute=function(t,e){var i=this;t=this.mapInputs(t);var r=Object.keys(t).sort();this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e);var a=r.map(function(d){return i.graph.nodes[$t(d)[0]]}),s=e.map(function(d){return $t(d)[0]}),o=s.map(function(d){return i.graph.nodes[d]});o.length===0&&(o=this._outputs);var l=this.getCompilationKey(a,o),u=this.compiledMap.get(l);u==null&&(u=this.compile(t,o),this.compiledMap.set(l,u));var c={},h={};return B.tidy(function(){var d=new Bw(i.weightMap,c,h,i.functionExecutorMap),p=Cw({},i.weightMap);Object.keys(t).forEach(function(w){var S=$t(w),L=S[0],N=S[1],C=[];C[N]=t[w],p[L]=C});for(var f=i.getFrozenTensorIds(p),m={},g=0;g0?(w=this.processStack(s,f,e,m,b,v,o,g,c),[4,Promise.all(w)]):[3,3];case 2:return C.sent(),[3,1];case 3:if(d==null&&!r&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead."),S=l.filter(function(R){return!zw(R)&&!Vt(R.name,m,e)}).map(function(R){return R.name}),S.length>0)throw L="",d!=null&&(L="Alternatively, to avoid the dynamic ops, use model.execute() "+("and specify the inputs ["+p+"]")),new Error("Cannot compute the outputs ["+S+"] from the provided "+("inputs ["+a+"]. Consider providing the following inputs: ")+("["+h+"]. "+L));return[2,m]}})})},n.prototype.processStack=function(t,e,i,r,a,s,o,l,u){for(var c=this,h=[],d=function(){var f=e.pop();i.currentContext=f.contexts;var m="";if(f.node.op==="Enter"&&A("isConstant",f.node,r,i)&&(m=ii(f.node.name,i)[0]),t.indexOf(f.node)===-1){var g=Uw(f.node,r,i);m||(m=ii(f.node.name,i)[0]);var v=i.currentContext;g instanceof Promise?h.push(g.then(function(b){return r[m]=b,i.currentContext=v,c.checkTensorForDisposal(m,f.node,r,i,s,o,l),c.processChildNodes(f.node,e,i,r,a,u),b})):(r[m]=g,p.checkTensorForDisposal(m,f.node,r,i,s,o,l),p.processChildNodes(f.node,e,i,r,a,u))}else p.processChildNodes(f.node,e,i,r,a,u)},p=this;e.length>0;)d();return h},n.prototype.processChildNodes=function(t,e,i,r,a,s){t.children.forEach(function(o){var l=ii(o.name,i)[0];if(a[l]||!s.has(o.name))return;o.op==="Merge"?o.inputNames.some(function(u){return!!Vt(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o})):o.inputNames.every(function(u){return!!Vt(u,r,i)})&&(a[l]=!0,e.push({contexts:i.currentContext,node:o}))})},n.prototype.dispose=function(){var t=this;Object.keys(this.weightMap).forEach(function(e){return t.weightMap[e].forEach(function(i){return i.dispose()})})},n.prototype.checkInputShapeAndType=function(t){var e=this;Object.keys(t).forEach(function(i){var r=t[i],a=$t(i)[0],s=e.graph.nodes[a];if(s.attrParams.shape&&s.attrParams.shape.value){var o=s.attrParams.shape.value,l=o.length===r.shape.length&&r.shape.every(function(u,c){return o[c]===-1||o[c]===u});B.util.assert(l,function(){return"The shape of dict['"+s.name+"'] provided in "+("model.execute(dict) must be ["+o+"], but was ")+("["+r.shape+"]")})}s.attrParams.dtype&&s.attrParams.dtype.value&&B.util.assert(r.dtype===s.attrParams.dtype.value,function(){return"The dtype of dict['"+s.name+"'] provided in model.execute(dict) must be "+(s.attrParams.dtype.value+", but was "+r.dtype)})})},n.prototype.mapInputs=function(t){var e={};for(var i in t)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[i]!=null){var r=this._signature.inputs[i];e[r.name]=t[i]}else e[i]=t[i];return e},n.prototype.checkInputs=function(t){var e=this,i=Object.keys(t).filter(function(r){var a=$t(r)[0];return e.graph.nodes[a]==null});if(i.length>0)throw new Error("The dict provided in model.execute(dict) has "+("keys: ["+i+"] that are not part of graph"))},n.prototype.mapOutputs=function(t){var e=this;return t.map(function(i){if(e._signature!=null&&e._signature.outputs!=null&&e._signature.outputs[i]!=null){var r=e._signature.outputs[i];return r.name}return i},{})},n.prototype.checkOutputs=function(t){var e=this;t.forEach(function(i){var r=$t(i)[0];if(!e.graph.nodes[r])throw new Error("The output '"+i+"' is not found in the graph")})},n}();var $U="?tfjs-format=file",XU="model.json",Mw=function(){function n(t,e){e===void 0&&(e={}),this.modelUrl=t,this.loadOptions=e,this.version="n/a",e==null&&(this.loadOptions={})}return Object.defineProperty(n.prototype,"modelVersion",{get:function(){return this.version},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputNodes",{get:function(){return this.executor.inputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputNodes",{get:function(){return this.executor.outputNodes},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"inputs",{get:function(){return this.executor.inputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"outputs",{get:function(){return this.executor.outputs},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"weights",{get:function(){return this.executor.weightMap},enumerable:!0,configurable:!0}),n.prototype.findIOHandler=function(){var t=this.modelUrl;if(t.load!=null)this.handler=t;else if(this.loadOptions.requestInit!=null)this.handler=B.io.browserHTTPRequest(t,this.loadOptions);else{var e=B.io.getLoadHandlers(t,this.loadOptions);if(e.length===0)e.push(B.io.browserHTTPRequest(t,this.loadOptions));else if(e.length>1)throw new Error("Found more than one ("+e.length+") load handlers for "+("URL '"+[t]+"'"));this.handler=e[0]}},n.prototype.load=function(){return _n(this,void 0,void 0,function(){var t;return In(this,function(e){switch(e.label){case 0:if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");return[4,this.handler.load()];case 1:return t=e.sent(),[2,this.loadSync(t)]}})})},n.prototype.loadSync=function(t){this.artifacts=t;var e=this.artifacts.modelTopology,i={};this.artifacts.userDefinedMetadata!=null&&(i=this.artifacts.userDefinedMetadata.signature),this.version=e.versions.producer+"."+e.versions.minConsumer;var r=B.io.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new _w(Dw.Instance.transformGraph(e,i)),this.executor.weightMap=this.convertTensorMapToTensorsMap(r),t.modelInitializer!=null){var a=Dw.Instance.transformGraph(t.modelInitializer);this.initializer=new _w(a),this.initializer.weightMap=this.executor.weightMap,this.initializer.execute({},[])}return!0},n.prototype.save=function(t,e){return _n(this,void 0,void 0,function(){var i;return In(this,function(r){if(typeof t=="string"){if(i=B.io.getSaveHandlers(t),i.length===0)throw new Error("Cannot find any save handlers for URL '"+t+"'");if(i.length>1)throw new Error("Found more than one ("+i.length+") save handlers for "+("URL '"+t+"'"));t=i[0]}if(t.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return[2,t.save(this.artifacts)]})})},n.prototype.predict=function(t,e){return this.execute(t,this.outputNodes)},n.prototype.normalizeInputs=function(t){if(!(t instanceof B.Tensor)&&!Array.isArray(t))return t;if(t=Array.isArray(t)?t:[t],t.length!==this.inputNodes.length)throw new Error("Input tensor count mismatch,"+("the graph model has "+this.inputNodes.length+" placeholders, ")+("while there are "+t.length+" input tensors."));return this.inputNodes.reduce(function(e,i,r){return e[i]=t[r],e},{})},n.prototype.normalizeOutputs=function(t){return t=t||this.outputNodes,Array.isArray(t)?t:[t]},n.prototype.execute=function(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);var i=this.executor.execute(t,e);return i.length>1?i:i[0]},n.prototype.executeAsync=function(t,e){return _n(this,void 0,void 0,function(){var i;return In(this,function(r){switch(r.label){case 0:return t=this.normalizeInputs(t),e=this.normalizeOutputs(e),[4,this.executor.executeAsync(t,e)];case 1:return i=r.sent(),[2,i.length>1?i:i[0]]}})})},n.prototype.convertTensorMapToTensorsMap=function(t){return Object.keys(t).reduce(function(e,i){return e[i]=[t[i]],e},{})},n.prototype.dispose=function(){this.executor.dispose(),this.initializer&&this.initializer.dispose()},n}();function JU(n,t){return t===void 0&&(t={}),_n(this,void 0,void 0,function(){var e;return In(this,function(i){switch(i.label){case 0:if(n==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");return t==null&&(t={}),t.fromTFHub&&(n.load==null&&(n.endsWith("/")||(n=n+"/"),n=""+n+XU+$U)),e=new Mw(n,t),[4,e.load()];case 1:return i.sent(),[2,e]}})})}var ZU="2.6.0";dr.GraphModel=Mw;dr.deregisterOp=PW;dr.loadGraphModel=JU;dr.registerOp=zW;dr.version_converter=ZU});var l0=Ae(qt=>{"use strict";Object.defineProperty(qt,"__esModule",{value:!0});var Ne=rr();var Ud=function(n,t){return Ud=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Ud(n,t)};function He(n,t){Ud(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function se(n,t,e,i){return new(e||(e=Promise))(function(r,a){function s(u){try{l(i.next(u))}catch(c){a(c)}}function o(u){try{l(i.throw(u))}catch(c){a(c)}}function l(u){u.done?r(u.value):new e(function(c){c(u.value)}).then(s,o)}l((i=i.apply(n,t||[])).next())})}function oe(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]>>0,d-=l,d*=l,l=d>>>0,d-=l,l+=d*4294967296}return(l>>>0)*23283064365386963e-26};return u}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.alea=s})(aa,n,!1)}),eB=pr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.x=0,l.y=0,l.z=0,l.w=0,l.next=function(){var h=l.x^l.x<<11;return l.x=l.y,l.y=l.z,l.z=l.w,l.w^=l.w>>>19^h^h>>>8},o===(o|0)?l.x=o:u+=o;for(var c=0;c>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor128=s})(aa,n,!1)}),tB=pr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.x^l.x>>>2;return l.x=l.y,l.y=l.z,l.z=l.w,l.w=l.v,(l.d=l.d+362437|0)+(l.v=l.v^l.v<<4^(h^h<<1))|0},l.x=0,l.y=0,l.z=0,l.w=0,l.v=0,o===(o|0)?l.x=o:u+=o;for(var c=0;c>>4),l.next()}function a(o,l){return l.x=o.x,l.y=o.y,l.z=o.z,l.w=o.w,l.v=o.v,l.d=o.d,l}function s(o,l){var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorwow=s})(aa,n,!1)}),nB=pr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.x,h=l.i,d,p;return d=c[h],d^=d>>>7,p=d^d<<24,d=c[h+1&7],p^=d^d>>>10,d=c[h+3&7],p^=d^d>>>3,d=c[h+4&7],p^=d^d<<7,d=c[h+7&7],d=d^d<<13,p^=d^d<<9,c[h]=p,l.i=h+1&7,p};function u(c,h){var d,p,f=[];if(h===(h|0))p=f[0]=h;else for(h=""+h,d=0;d0;--d)c.next()}u(l,o)}function a(o,l){return l.x=o.x.slice(),l.i=o.i,l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.x&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xorshift7=s})(aa,n,!1)}),iB=pr(function(n){(function(t,e,i){function r(o){var l=this;l.next=function(){var c=l.w,h=l.X,d=l.i,p,f;return l.w=c=c+1640531527|0,f=h[d+34&127],p=h[d=d+1&127],f^=f<<13,p^=p<<17,f^=f>>>15,p^=p>>>12,f=h[d]=f^p,l.i=d,f+(c^c>>>16)|0};function u(c,h){var d,p,f,m,g,v=[],b=128;for(h===(h|0)?(p=h,h=null):(h=h+"\0",p=0,b=Math.max(b,h.length)),f=0,m=-32;m>>15,p^=p<<4,p^=p>>>13,m>=0&&(g=g+1640531527|0,d=v[m&127]^=p+g,f=d==0?f+1:0);for(f>=128&&(v[(h&&h.length||0)&127]=-1),f=127,m=4*128;m>0;--m)p=v[f+34&127],d=v[f=f+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,v[f]=p^d;c.w=g,c.X=v,c.i=f}u(l,o)}function a(o,l){return l.i=o.i,l.w=o.w,l.X=o.X.slice(),l}function s(o,l){o==null&&(o=+new Date);var u=new r(o),c=l&&l.state,h=function(){return(u.next()>>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(c.X&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.xor4096=s})(aa,n,!1)}),rB=pr(function(n){(function(t,e,i){function r(o){var l=this,u="";l.next=function(){var h=l.b,d=l.c,p=l.d,f=l.a;return h=h<<25^h>>>7^d,d=d-p|0,p=p<<24^p>>>8^f,f=f-h|0,l.b=h=h<<20^h>>>12^d,l.c=d=d-p|0,l.d=p<<16^d>>>16^f,l.a=f-h|0},l.a=0,l.b=0,l.c=2654435769|0,l.d=1367130551,o===Math.floor(o)?(l.a=o/4294967296|0,l.b=o|0):u+=o;for(var c=0;c>>0)/4294967296};return h.double=function(){do var d=u.next()>>>11,p=(u.next()>>>0)/4294967296,f=(d+p)/(1<<21);while(f===0);return f},h.int32=u.next,h.quick=h,c&&(typeof c=="object"&&a(c,u),h.state=function(){return a(u,{})}),h}e&&e.exports?e.exports=s:i&&i.amd?i(function(){return s}):this.tychei=s})(aa,n,!1)}),fr=pr(function(n){(function(t,e){var i=this,r=256,a=6,s=52,o="random",l=e.pow(r,a),u=e.pow(2,s),c=u*2,h=r-1,d;function p(S,L,N){var C=[];L=L==!0?{entropy:!0}:L||{};var R=v(g(L.entropy?[S,w(t)]:S??b(),3),C),D=new f(C),k=function(){for(var W=D.g(a),F=l,P=0;W=c;)W/=2,F/=2,P>>>=1;return(W+P)/F};return k.int32=function(){return D.g(4)|0},k.quick=function(){return D.g(4)/4294967296},k.double=k,v(w(D.S),t),(L.pass||N||function(W,F,P,H){return H&&(H.S&&m(H,D),W.state=function(){return m(D,{})}),P?(e[o]=W,F):W})(k,R,"global"in L?L.global:this==e,L.state)}e["seed"+o]=p;function f(S){var L,N=S.length,C=this,R=0,D=C.i=C.j=0,k=C.S=[];for(N||(S=[N++]);R=this.items.length?[2,{value:null,done:!0}]:(e=this.items[this.trav],this.trav++,[2,{value:hB(e),done:!1}])})})},t}(Nt),fB=function(n){He(t,n);function t(e){var i=n.call(this)||this;return i.nextFn=e,i}return t.prototype.summary=function(){return"Function call"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){try{return[2,this.nextFn()]}catch(i){throw i.message="Error thrown while iterating through a dataset: "+i.message,i}return[2]})})},t}(Nt),yB=function(n){He(t,n);function t(e){var i=n.call(this)||this;return i.upstream=e,i.lastRead=Promise.resolve({value:null,done:!1}),i}return t.prototype.summary=function(){return this.upstream.summary()+" -> Serial"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.upstream.next()]})})},t}(Nt),bB=function(n){He(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.maxCount=i,r.count=0,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Skip"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return this.count++ Take"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.count++>=this.maxCount?[2,{value:null,done:!0}]:[2,this.upstream.next()]})})},t}(Nt),SB=function(n){He(t,n);function t(e,i,r){r===void 0&&(r=!0);var a=n.call(this)||this;return a.upstream=e,a.batchSize=i,a.enableSmallLastBatch=r,a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.summary=function(){return this.upstream.summary()+" -> RowMajorBatch"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:e=[],r.label=1;case 1:return e.length0?[2,{value:e,done:!1}]:[2,{value:null,done:!0}]:(e.push(i.value),[3,1]);case 3:return[2,{value:e,done:!1}]}})})},t}(Nt),LB=function(n){He(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.predicate=i,r.lastRead=Promise.resolve({value:null,done:!1}),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Filter"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,this.upstream.next()];case 1:return e=i.sent(),e.done||this.predicate(e.value)?[2,e]:(Ne.dispose(e.value),[3,0]);case 2:return[2]}})})},t}(Nt),IB=function(n){He(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Map"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,{value:null,done:!0}];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s handleErrors"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:i.label=1;case 1:return i.trys.push([1,3,,4]),[4,this.upstream.next()];case 2:return[2,i.sent()];case 3:return e=i.sent(),this.handler(e)?[3,4]:[2,{value:null,done:!0}];case 4:return[3,0];case 5:return[2]}})})},t}(Nt),Xw=function(n){He(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.transform=i,r}return t.prototype.summary=function(){return this.upstream.summary()+" -> AsyncMap"},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:return e=u.sent(),e.done?[2,{value:null,done:!0}]:(i=Ne.tensor_util.getTensorsInContainer(e.value),[4,this.transform(e.value)]);case 2:for(r=u.sent(),a=Ne.tensor_util.getTensorsInContainer(r),s=0,o=i;s Flatmap"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s,o,l;return oe(this,function(u){switch(u.label){case 0:return[4,this.upstream.next()];case 1:if(e=u.sent(),e.done)return[2,!1];for(i=Ne.tensor_util.getTensorsInContainer(e.value),r=this.transform(e.value),a=Ne.tensor_util.getTensorsInContainer(r),this.outputQueue.pushAll(r),s=0,o=i;s Chained"},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.lastRead=this.readFromChain(this.lastRead),[2,this.lastRead]})})},t.prototype.readFromChain=function(e){return se(this,void 0,void 0,function(){var i,r;return oe(this,function(a){switch(a.label){case 0:return[4,e];case 1:return a.sent(),this.iterator==null?[4,this.moreIterators.next()]:[3,3];case 2:if(i=a.sent(),i.done)return[2,{value:null,done:!0}];this.iterator=i.value,this.baseErrorHandler!=null&&(this.iterator=this.iterator.handleErrors(this.baseErrorHandler)),a.label=3;case 3:return[4,this.iterator.next()];case 4:return r=a.sent(),r.done?(this.iterator=null,[2,this.readFromChain(e)]):[2,r]}})})},t}(Nt),Ei;(function(n){n[n.FAIL=0]="FAIL",n[n.SHORTEST=1]="SHORTEST",n[n.LONGEST=2]="LONGEST"})(Ei||(Ei={}));var gB=function(n){He(t,n);function t(e,i){i===void 0&&(i=Ei.FAIL);var r=n.call(this)||this;return r.iterators=e,r.mismatchMode=i,r.count=0,r.currentPromise=null,r}return t.prototype.summary=function(){var e="TODO: fill in upstream of zip summaries";return"{"+e+"} -> Zip"},t.prototype.nextState=function(e){return se(this,void 0,void 0,function(){function i(o){if(o instanceof Nt){var l=o.next();return{value:l.then(function(u){return r++,u.done&&a++,u.value}),recurse:!1}}else return{value:null,recurse:!0}}var r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,e];case 1:return o.sent(),r=0,a=0,[4,Yw(this.iterators,i)];case 2:if(s=o.sent(),r===a)return[2,{value:null,done:!0}];if(a>0)switch(this.mismatchMode){case Ei.FAIL:throw new Error("Zipped streams should have the same length. "+("Mismatched at element "+this.count+"."));case Ei.SHORTEST:return[2,{value:null,done:!0}];case Ei.LONGEST:}return this.count++,[2,{value:s,done:!1}]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return this.currentPromise=this.nextState(this.currentPromise),[2,this.currentPromise]})})},t}(Nt),Jw=function(n){He(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.bufferSize=i,r.buffer=new Kw(i),r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Prefetch"},t.prototype.refill=function(){for(;!this.buffer.isFull();){var e=this.upstream.next();this.buffer.push(e)}},t.prototype.next=function(){return this.refill(),this.buffer.shift()},t}(Nt),NB=function(n){He(t,n);function t(e,i,r){var a=n.call(this,e,i)||this;return a.upstream=e,a.windowSize=i,a.upstreamExhausted=!1,a.random=Vw(r||Ne.util.now().toString()),a.lastRead=Promise.resolve({value:null,done:!1}),a}return t.prototype.next=function(){return se(this,void 0,void 0,function(){var e=this;return oe(this,function(i){return this.lastRead=this.lastRead.then(function(){return e.serialNext()}),[2,this.lastRead]})})},t.prototype.randomInt=function(e){return Math.floor(this.random()*e)},t.prototype.chooseIndex=function(){return this.randomInt(this.buffer.length())},t.prototype.serialNext=function(){return se(this,void 0,void 0,function(){var e,i;return oe(this,function(r){switch(r.label){case 0:this.upstreamExhausted||this.refill(),r.label=1;case 1:return this.buffer.isEmpty()?[3,3]:(e=this.chooseIndex(),[4,this.buffer.shuffleExcise(e)]);case 2:if(i=r.sent(),i.done)this.upstreamExhausted=!0;else return this.refill(),[2,i];return[3,1];case 3:return[2,{value:null,done:!0}]}})})},t}(Jw);var us=function(){function n(){this.size=null}return n.prototype.batch=function(t,e){var i=this;e===void 0&&(e=!0);var r=this;Ne.util.assert(t>0,function(){return`batchSize needs to be positive, but it is + `+t});var a;return this.size===Infinity||this.size==null?a=this.size:e?a=Math.ceil(this.size/t):a=Math.floor(this.size/t),Xt(function(){return se(i,void 0,void 0,function(){return oe(this,function(s){switch(s.label){case 0:return[4,r.iterator()];case 1:return[2,s.sent().columnMajorBatch(t,e,xB)]}})})},a)},n.prototype.concatenate=function(t){var e=this,i=this,r;return this.size===Infinity||t.size===Infinity?r=Infinity:this.size!=null&&t.size!=null?r=this.size+t.size:r=null,Xt(function(){return se(e,void 0,void 0,function(){var a,s;return oe(this,function(o){switch(o.label){case 0:return[4,i.iterator()];case 1:return s=(a=o.sent()).concatenate,[4,t.iterator()];case 2:return[2,s.apply(a,[o.sent()])]}})})},r)},n.prototype.filter=function(t){var e=this,i=this,r;return this.size===Infinity?r=Infinity:r=null,Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().filter(function(s){return Ne.tidy(function(){return t(s)})})]}})})},r)},n.prototype.forEachAsync=function(t){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.iterator()];case 1:return[2,e.sent().forEachAsync(t)]}})})},n.prototype.map=function(t){var e=this,i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().map(function(a){return Ne.tidy(function(){return t(a)})})]}})})},this.size)},n.prototype.mapAsync=function(t){var e=this,i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().mapAsync(t)]}})})},this.size)},n.prototype.prefetch=function(t){var e=this;if(t==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");var i=this;return Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(r){switch(r.label){case 0:return[4,i.iterator()];case 1:return[2,r.sent().prefetch(t)]}})})},this.size)},n.prototype.repeat=function(t){var e=this,i=this,r;return this.size!=null&&t>0?r=this.size*t:t===0?r=0:this.size!=null&&(t===void 0||t<0)?r=Infinity:r=null,Xt(function(){return se(e,void 0,void 0,function(){var a,s=this;return oe(this,function(o){return a=Bd(function(){return se(s,void 0,void 0,function(){var l;return oe(this,function(u){switch(u.label){case 0:return l={},[4,i.iterator()];case 1:return[2,(l.value=u.sent(),l.done=!1,l)]}})})}),[2,mB(a.take(t))]})})},r)},n.prototype.skip=function(t){var e=this,i=this,r;return this.size!=null&&t>=0&&this.size>=t?r=this.size-t:this.size!=null&&(this.sizet?r=t:this.size!=null&&this.size<=t?r=this.size:r=null,Xt(function(){return se(e,void 0,void 0,function(){return oe(this,function(a){switch(a.label){case 0:return[4,i.iterator()];case 1:return[2,a.sent().take(t)]}})})},r)},n.prototype.toArray=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArray()]}})})},n.prototype.toArrayForTest=function(){return se(this,void 0,void 0,function(){return oe(this,function(t){switch(t.label){case 0:if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return[4,this.iterator()];case 1:return[2,t.sent().toArrayForTest()]}})})},n.MAX_BUFFER_SIZE=1e4,n}();function Xt(n,t){return t===void 0&&(t=null),new(function(e){He(i,e);function i(){var r=e!==null&&e.apply(this,arguments)||this;return r.size=t,r}return i.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(r){return[2,n()]})})},i}(us))}function CB(n){var t=this;return Xt(function(){return se(t,void 0,void 0,function(){return oe(this,function(e){return[2,jw(n)]})})},n.length)}function RB(n){var t=this;if(!sa(n))throw new Error("The argument to zip() must be an object or array.");var e;if(Array.isArray(n))for(var i=0;i1}),Ne.util.assert(r.length===0,function(){return"Duplicate column names found: "+r.toString()}),this.columnConfigs){for(a=0,s=Object.keys(this.columnConfigs);a14||!Number.isInteger(r))throw new Error("Invalid fftSize: it must be a power of 2 between "+("2 to 4 and 2 to 14, but got "+i.fftSize));if(i.numFrames=e.numFramesPerSpectrogram||43,i.sampleRateHz=e.sampleRateHz,i.columnTruncateLength=e.columnTruncateLength||i.fftSize,i.audioTrackConstraints=e.audioTrackConstraints,i.smoothingTimeConstant=e.smoothingTimeConstant||0,i.includeSpectrogram=!(e.includeSpectrogram===!1),i.includeWaveform=e.includeWaveform===!0,!i.includeSpectrogram&&!i.includeWaveform)throw new Error("Both includeSpectrogram and includeWaveform are false. At least one type of data should be returned.");return i}return t.prototype.summary=function(){return"microphone"},t.create=function(e){return e===void 0&&(e={}),se(this,void 0,void 0,function(){var i;return oe(this,function(r){switch(r.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");return i=new t(e),[4,i.start()];case 1:return r.sent(),[2,i]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r,a;return oe(this,function(s){switch(s.label){case 0:return s.trys.push([0,2,,3]),e=this,[4,navigator.mediaDevices.getUserMedia({audio:this.audioTrackConstraints==null?!0:this.audioTrackConstraints,video:!1})];case 1:return e.stream=s.sent(),[3,3];case 2:throw i=s.sent(),new Error("Error thrown while initializing video stream: "+i.message);case 3:if(!this.stream)throw new Error("Could not obtain audio from microphone.");if(r=window.AudioContext||window.webkitAudioContext,this.audioContext=new r,!this.sampleRateHz)this.sampleRateHz=this.audioContext.sampleRate;else if(this.audioContext.sampleRate!==this.sampleRateHz)throw new Error("Mismatch in sampling rate: "+("Expected: "+this.sampleRateHz+"; ")+("Actual: "+this.audioContext.sampleRate));return a=this.audioContext.createMediaStreamSource(this.stream),this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,a.connect(this.analyser),this.freqData=new Float32Array(this.fftSize),this.timeData=new Float32Array(this.fftSize),[2]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return this.isClosed?[2,{value:null,done:!0}]:[4,this.getAudioData()];case 1:return r=o.sent(),this.includeSpectrogram&&(a=this.flattenQueue(r.freqDataQueue),e=this.getTensorFromAudioDataArray(a,[this.numFrames,this.columnTruncateLength,1])),this.includeWaveform&&(s=this.flattenQueue(r.timeDataQueue),i=this.getTensorFromAudioDataArray(s,[this.numFrames*this.fftSize,1])),[2,{value:{spectrogram:e,waveform:i},done:!1}]}})})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.getAudioData=function(){return se(this,void 0,void 0,function(){var e,i,r,a=this;return oe(this,function(s){return e=[],i=[],r=0,[2,new Promise(function(o){var l=setInterval(function(){a.includeSpectrogram&&(a.analyser.getFloatFrequencyData(a.freqData),a.freqData[0]===-Infinity&&o({freqDataQueue:e,timeDataQueue:i}),e.push(a.freqData.slice(0,a.columnTruncateLength))),a.includeWaveform&&(a.analyser.getFloatTimeDomainData(a.timeData),i.push(a.timeData.slice())),++r===a.numFrames&&(clearInterval(l),o({freqDataQueue:e,timeDataQueue:i}))},a.fftSize/a.sampleRateHz*1e3)})]})})},t.prototype.stop=function(){this.isClosed||(this.isClosed=!0,this.analyser.disconnect(),this.audioContext.close(),this.stream!=null&&this.stream.getTracks().length>0&&this.stream.getTracks()[0].stop())},t.prototype.toArray=function(){throw new Error("Can not convert infinite audio stream to array.")},t.prototype.getSampleRate=function(){return this.sampleRateHz},t.prototype.flattenQueue=function(e){var i=e[0].length,r=new Float32Array(e.length*i);return e.forEach(function(a,s){return r.set(a,s*i)}),r},t.prototype.getTensorFromAudioDataArray=function(e,i){var r=new Float32Array(Ne.util.sizeFromShape(i));return r.set(e,r.length-e.length),Ne.tensor(r,i)},t}(Nt);var DB=function(n){He(t,n);function t(e,i){var r=n.call(this)||this;if(r.webcamVideoElement=e,r.webcamConfig=i,r.isClosed=!0,r.resize=!1,r.needToResize())if(r.resize=!0,r.cropSize=[r.webcamConfig.resizeHeight,r.webcamConfig.resizeWidth],r.cropBoxInd=Ne.tensor1d([0],"int32"),r.webcamConfig.centerCrop){var a=r.webcamConfig.resizeWidth*1/r.webcamVideoElement.width,s=r.webcamConfig.resizeHeight*1/r.webcamVideoElement.height,o=(1-a)/2,l=(1-s)/2,u=o+a,c=s+l;r.cropBox=Ne.tensor2d([l,o,c,u],[1,4])}else r.cropBox=Ne.tensor2d([0,0,1,1],[1,4]);return r}return t.prototype.summary=function(){return"webcam"},t.create=function(e,i){return i===void 0&&(i={}),se(this,void 0,void 0,function(){var r;return oe(this,function(a){switch(a.label){case 0:if(Ne.env().get("IS_NODE"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!e){if(e=document.createElement("video"),!i.resizeWidth||!i.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");e.width=i.resizeWidth,e.height=i.resizeHeight}return r=new t(e,i),[4,r.start()];case 1:return a.sent(),[2,r]}})})},t.prototype.start=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:this.webcamConfig.facingMode&&Ne.util.assert(this.webcamConfig.facingMode==="user"||this.webcamConfig.facingMode==="environment",function(){return"Invalid webcam facing mode: "+r.webcamConfig.facingMode+". Please provide 'user' or 'environment'"}),a.label=1;case 1:return a.trys.push([1,3,,4]),e=this,[4,navigator.mediaDevices.getUserMedia({video:{deviceId:this.webcamConfig.deviceId,facingMode:this.webcamConfig.facingMode?this.webcamConfig.facingMode:"user",width:this.webcamVideoElement.width,height:this.webcamVideoElement.height}})];case 2:return e.stream=a.sent(),[3,4];case 3:throw i=a.sent(),i.message="Error thrown while initializing video stream: "+i.message,i;case 4:if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(s){console.log(s),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,[2,new Promise(function(s){r.webcamVideoElement.onloadedmetadata=function(){s()}})]}})})},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){if(this.isClosed)return[2,{value:null,done:!0}];try{e=Ne.browser.fromPixels(this.webcamVideoElement)}catch(r){throw new Error("Error thrown converting video to pixels: "+JSON.stringify(r))}if(this.resize)try{return[2,{value:this.cropAndResizeFrame(e),done:!1}]}catch(r){throw new Error("Error thrown cropping the video: "+r.message)}finally{e.dispose()}else return[2,{value:e,done:!1}];return[2]})})},t.prototype.needToResize=function(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))},t.prototype.cropAndResizeFrame=function(e){var i=this;return Ne.tidy(function(){var r=e.toFloat().expandDims(0),a;a=Ne.image.cropAndResize(r,i.cropBox,i.cropBoxInd,i.cropSize,"bilinear");var s=a.shape;return a.reshape(s.slice(1))})},t.prototype.capture=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){switch(e.label){case 0:return[4,this.next()];case 1:return[2,e.sent().value]}})})},t.prototype.stop=function(){var e=this.stream.getTracks();e.forEach(function(i){return i.stop()});try{this.webcamVideoElement.srcObject=null}catch(i){console.log(i),this.webcamVideoElement.src=null}this.isClosed=!0},t.prototype.toArray=function(){throw new Error("Can not convert infinite video stream to array.")},t}(Nt);var n0=function(){function n(){}return n}();var i0=function(n){He(t,n);function t(){return n!==null&&n.apply(this,arguments)||this}return t.prototype.split=function(e){return new kB(this,e)},t}(Nt),kB=function(n){He(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.impl=new FB(e,i),r}return t.prototype.summary=function(){return this.impl.summary()},t.prototype.next=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,this.impl.next()]})})},t}(i0),FB=function(n){He(t,n);function t(e,i){var r=n.call(this)||this;return r.upstream=e,r.separator=i,r.carryover="",r}return t.prototype.summary=function(){return this.upstream.summary()+" -> Split('"+this.separator+"')"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return[4,this.upstream.next()];case 1:if(e=o.sent(),e.done)return this.carryover===""?[2,!1]:(this.outputQueue.push(this.carryover),this.carryover="",[2,!0]);for(i=e.value.split(this.separator),i[0]=this.carryover+i[0],r=0,a=i.slice(0,-1);r Utf8"},t.prototype.pump=function(){return se(this,void 0,void 0,function(){var e,i,r;return oe(this,function(a){switch(a.label){case 0:return[4,this.upstream.next()];case 1:return e=a.sent(),e.done?[2,!1]:(i=e.value,Ne.env().get("IS_BROWSER")?r=this.decoder.decode(i,{stream:!0}):r=this.decoder.write(Buffer.from(i.buffer)),this.outputQueue.push(r),[2,!0])}})})},t}(zd);var r0=function(n){He(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.file=e,r.options=i,Ne.util.assert(e instanceof Uint8Array||(Ne.env().get("IS_BROWSER")?e instanceof File||e instanceof Blob:!1),function(){return"FileChunkIterator only supports File, Blob and Uint8Array right now."}),r.offset=i.offset||0,r.chunkSize=i.chunkSize||1024*1024,r}return t.prototype.summary=function(){return"FileChunks "+this.file},t.prototype.next=function(){return se(this,void 0,void 0,function(){var e,i,r=this;return oe(this,function(a){switch(a.label){case 0:return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?[2,{value:null,done:!0}]:(e=new Promise(function(s,o){var l=r.offset+r.chunkSize;if(r.file instanceof Uint8Array)s(new Uint8Array(r.file.slice(r.offset,l)));else{var u=new FileReader;u.onload=function(h){var d=u.result;if(d instanceof ArrayBuffer&&(d=new Uint8Array(d)),!(d instanceof Uint8Array))return o(new TypeError("FileReader returned unknown type."));s(d)},u.onabort=function(h){return o(new Error("Aborted"))},u.onerror=function(h){return o(new Error(h.type))};var c=r.file.slice(r.offset,l);u.readAsArrayBuffer(c)}r.offset=l}),i={},[4,e]);case 1:return[2,(i.value=a.sent(),i.done=!1,i)]}})})},t}(UB);function PB(n,t){return t===void 0&&(t={}),se(this,void 0,void 0,function(){var e,i,r,a,s;return oe(this,function(o){switch(o.label){case 0:return typeof n=="string"?e=n:(e=n.url,i=zB(n)),[4,Ne.util.fetch(e,i)];case 1:return r=o.sent(),r.ok?(s=Uint8Array.bind,[4,r.arrayBuffer()]):[3,3];case 2:return a=new(s.apply(Uint8Array,[void 0,o.sent()])),[2,new r0(a,t)];case 3:throw new Error(r.statusText)}})})}var zB=function(n){var t={method:n.method,headers:n.headers,body:n.body,mode:n.mode,credentials:n.credentials,cache:n.cache,redirect:n.redirect,referrer:n.referrer,integrity:n.integrity};return t};function a0(n){return typeof n=="string"&&n.substr(0,7)==="file://"}var s0=function(n){He(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.input=e,r.options=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){var e;return oe(this,function(i){return a0(this.input)&&Ne.env().get("IS_NODE")&&(e=require("fs"),this.input=e.readFileSync(this.input.substr(7))),[2,new r0(this.input,this.options)]})})},t}(n0);var o0=function(n){He(t,n);function t(e,i){i===void 0&&(i={});var r=n.call(this)||this;return r.url=e,r.fileOptions=i,r}return t.prototype.iterator=function(){return se(this,void 0,void 0,function(){return oe(this,function(e){return a0(this.url)?[2,new s0(this.url,this.fileOptions).iterator()]:[2,PB(this.url,this.fileOptions)]})})},t}(n0);function _B(n,t){return t===void 0&&(t={}),new t0(new o0(n),t)}function MB(n){var t=this,e=Bd(n);return Xt(function(){return se(t,void 0,void 0,function(){return oe(this,function(i){return[2,e]})})})}function HB(n){var t=this;return Xt(function(){return se(t,void 0,void 0,function(){var e;return oe(this,function(i){switch(i.label){case 0:return[4,n()];case 1:return e=i.sent(),[2,Bd(function(){return e.next()})]}})})})}function VB(n,t){return se(this,void 0,void 0,function(){return oe(this,function(e){return[2,DB.create(n,t)]})})}function qB(n){return se(this,void 0,void 0,function(){return oe(this,function(t){return[2,EB.create(n)]})})}var GB="2.6.0";qt.CSVDataset=t0;qt.Dataset=us;qt.FileDataSource=s0;qt.TextLineDataset=Zw;qt.URLDataSource=o0;qt.array=CB;qt.csv=_B;qt.func=MB;qt.generator=HB;qt.microphone=qB;qt.version_data=GB;qt.webcam=VB;qt.zip=RB});var c0=Ae((u0,_d)=>{(function(n,t,e){function i(o){var l=this,u=s();l.next=function(){var c=2091639*l.s0+l.c*23283064365386963e-26;return l.s0=l.s1,l.s1=l.s2,l.s2=c-(l.c=c|0)},l.c=1,l.s0=u(" "),l.s1=u(" "),l.s2=u(" "),l.s0-=u(o),l.s0<0&&(l.s0+=1),l.s1-=u(o),l.s1<0&&(l.s1+=1),l.s2-=u(o),l.s2<0&&(l.s2+=1),u=null}function r(o,l){return l.c=o.c,l.s0=o.s0,l.s1=o.s1,l.s2=o.s2,l}function a(o,l){var u=new i(o),c=l&&l.state,h=u.next;return h.int32=function(){return u.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,c&&(typeof c=="object"&&r(c,u),h.state=function(){return r(u,{})}),h}function s(){var o=4022871197,l=function(u){u=u.toString();for(var c=0;c>>0,h-=o,h*=o,o=h>>>0,h-=o,o+=h*4294967296}return(o>>>0)*23283064365386963e-26};return l}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.alea=a})(u0,typeof _d=="object"&&_d,typeof define=="function"&&define)});var d0=Ae((h0,Md)=>{(function(n,t,e){function i(s){var o=this,l="";o.x=0,o.y=0,o.z=0,o.w=0,o.next=function(){var c=o.x^o.x<<11;return o.x=o.y,o.y=o.z,o.z=o.w,o.w^=o.w>>>19^c^c>>>8},s===(s|0)?o.x=s:l+=s;for(var u=0;u>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor128=a})(h0,typeof Md=="object"&&Md,typeof define=="function"&&define)});var f0=Ae((p0,Hd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.x^o.x>>>2;return o.x=o.y,o.y=o.z,o.z=o.w,o.w=o.v,(o.d=o.d+362437|0)+(o.v=o.v^o.v<<4^(c^c<<1))|0},o.x=0,o.y=0,o.z=0,o.w=0,o.v=0,s===(s|0)?o.x=s:l+=s;for(var u=0;u>>4),o.next()}function r(s,o){return o.x=s.x,o.y=s.y,o.z=s.z,o.w=s.w,o.v=s.v,o.d=s.d,o}function a(s,o){var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorwow=a})(p0,typeof Hd=="object"&&Hd,typeof define=="function"&&define)});var g0=Ae((m0,Vd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.x,c=o.i,h,d,p;return h=u[c],h^=h>>>7,d=h^h<<24,h=u[c+1&7],d^=h^h>>>10,h=u[c+3&7],d^=h^h>>>3,h=u[c+4&7],d^=h^h<<7,h=u[c+7&7],h=h^h<<13,d^=h^h<<9,u[c]=d,o.i=c+1&7,d};function l(u,c){var h,d,p=[];if(c===(c|0))d=p[0]=c;else for(c=""+c,h=0;h0;--h)u.next()}l(o,s)}function r(s,o){return o.x=s.x.slice(),o.i=s.i,o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.x&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xorshift7=a})(m0,typeof Vd=="object"&&Vd,typeof define=="function"&&define)});var y0=Ae((v0,qd)=>{(function(n,t,e){function i(s){var o=this;o.next=function(){var u=o.w,c=o.X,h=o.i,d,p;return o.w=u=u+1640531527|0,p=c[h+34&127],d=c[h=h+1&127],p^=p<<13,d^=d<<17,p^=p>>>15,d^=d>>>12,p=c[h]=p^d,o.i=h,p+(u^u>>>16)|0};function l(u,c){var h,d,p,f,m,g=[],v=128;for(c===(c|0)?(d=c,c=null):(c=c+"\0",d=0,v=Math.max(v,c.length)),p=0,f=-32;f>>15,d^=d<<4,d^=d>>>13,f>=0&&(m=m+1640531527|0,h=g[f&127]^=d+m,p=h==0?p+1:0);for(p>=128&&(g[(c&&c.length||0)&127]=-1),p=127,f=4*128;f>0;--f)d=g[p+34&127],h=g[p=p+1&127],d^=d<<13,h^=h<<17,d^=d>>>15,h^=h>>>12,g[p]=d^h;u.w=m,u.X=g,u.i=p}l(o,s)}function r(s,o){return o.i=s.i,o.w=s.w,o.X=s.X.slice(),o}function a(s,o){s==null&&(s=+new Date);var l=new i(s),u=o&&o.state,c=function(){return(l.next()>>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(u.X&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.xor4096=a})(v0,typeof qd=="object"&&qd,typeof define=="function"&&define)});var w0=Ae((b0,Gd)=>{(function(n,t,e){function i(s){var o=this,l="";o.next=function(){var c=o.b,h=o.c,d=o.d,p=o.a;return c=c<<25^c>>>7^h,h=h-d|0,d=d<<24^d>>>8^p,p=p-c|0,o.b=c=c<<20^c>>>12^h,o.c=h=h-d|0,o.d=d<<16^h>>>16^p,o.a=p-c|0},o.a=0,o.b=0,o.c=2654435769|0,o.d=1367130551,s===Math.floor(s)?(o.a=s/4294967296|0,o.b=s|0):l+=s;for(var u=0;u>>0)/4294967296};return c.double=function(){do var h=l.next()>>>11,d=(l.next()>>>0)/4294967296,p=(h+d)/(1<<21);while(p===0);return p},c.int32=l.next,c.quick=c,u&&(typeof u=="object"&&r(u,l),c.state=function(){return r(l,{})}),c}t&&t.exports?t.exports=a:e&&e.amd?e(function(){return a}):this.tychei=a})(b0,typeof Gd=="object"&&Gd,typeof define=="function"&&define)});var S0=Ae((hV,$o)=>{(function(n,t){var e=this,i=256,r=6,a=52,s="random",o=t.pow(i,r),l=t.pow(2,a),u=l*2,c=i-1,h;function d(w,S,L){var N=[];S=S==!0?{entropy:!0}:S||{};var C=g(m(S.entropy?[w,b(n)]:w??v(),3),N),R=new p(N),D=function(){for(var k=R.g(r),W=o,F=0;k=u;)k/=2,W/=2,F>>>=1;return(k+F)/W};return D.int32=function(){return R.g(4)|0},D.quick=function(){return R.g(4)/4294967296},D.double=D,g(b(R.S),n),(S.pass||L||function(k,W,F,P){return P&&(P.S&&f(P,R),k.state=function(){return f(R,{})}),F?(t[s]=k,W):k})(D,C,"global"in S?S.global:this==t,S.state)}t["seed"+s]=d;function p(w){var S,L=w.length,N=this,C=0,R=N.i=N.j=0,D=N.S=[];for(L||(w=[L++]);C{var YB=c0(),KB=d0(),jB=f0(),$B=g0(),XB=y0(),JB=w0(),mr=S0();mr.alea=YB;mr.xor128=KB;mr.xorwow=jB;mr.xorshift7=$B;mr.xor4096=XB;mr.tychei=JB;L0.exports=mr});var $0=Ae(hs=>{"use strict";Object.defineProperty(hs,"__esModule",{value:!0});var T=rr(),ZB=I0();var Yd=function(n,t){return Yd=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},Yd(n,t)};function QB(n,t){Yd(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function A0(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function T0(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]f&&(f=v,m=g)}c[d]=m}return l},t.prototype.cumsum=function(e,i,r,a){if(ae(e,"cumsum"),i!==e.rank-1)throw new Error("backend.cumsum in CPU expects an inner-most axis="+(e.rank-1)+" "+("but got axis="+i));for(var s=T.upcastType(e.dtype,"int32"),o=T.zeros(e.shape,s),l=this.readSync(o.dataId),u=this.readSync(e.dataId),c=e.shape[e.rank-1],h=a?function(g,v){return g+c-v-1}:function(g,v){return g+v},d=0;da?1:0})},t.prototype.greaterEqual=function(e,i){return ae([e,i],"greaterEqual"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r>=a?1:0})},t.prototype.logicalAnd=function(e,i){return ae([e,i],"logicalAnd"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r&&a})},t.prototype.logicalOr=function(e,i){return ae([e,i],"logicalOr"),this.broadcastedBinaryOp(e,i,"bool",function(r,a){return r||a})},t.prototype.select=function(e,i,r){ae([e,i,r],"select");for(var a=this.readSync(e.dataId),s=this.readSync(i.dataId),o=this.readSync(r.dataId),l=T.zeros(i.shape,T.upcastType(i.dtype,r.dtype)),u=this.readSync(l.dataId),c=0,h=e.rank===0||e.rank>1||i.rank===1?1:T.util.sizeFromShape(i.shape.slice(1)),d=0;d=0&&a>=0?s:(s+a)%a})},t.prototype.maximum=function(e,i){return ae([e,i],"maximum"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.max(r,a)})},t.prototype.all=function(e,i){ae(e,"all"),T.backend_util.assertAxesAreInnerMostDims("all",i,e.rank);for(var r=T.backend_util.computeOutAndReduceShapes(e.shape,i),a=r[0],s=r[1],o=T.zeros(a,e.dtype),l=T.util.sizeFromShape(s),u=this.readSync(o.dataId),c=this.readSync(e.dataId),h=0;h=1?r[o]=s[o]:r[o]=s[o]*(l+1)}return this.makeOutput(r,i.shape,"float32")},t.prototype.atan2=function(e,i){return ae([e,i],"atan2"),this.broadcastedBinaryOp(e,i,e.dtype,function(r,a){return Math.atan2(r,a)})},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=this.conv2d(i,r,a);return s&&(u=T.add(u,s)),o&&(u=Kd(this,u,o,l)),u},t.prototype.conv2d=function(e,i,r){ae([e,i],"conv2d");for(var a=r.filterHeight,s=r.filterWidth,o=r.dilationHeight,l=r.dilationWidth,u=r.padInfo.left,c=r.padInfo.top,h=r.dataFormat==="channelsLast",d=T.buffer(r.outShape,e.dtype),p=e.strides[0],f=h?e.strides[1]:e.strides[2],m=h?e.strides[2]:1,g=h?1:e.strides[1],v=d.strides[0],b=h?d.strides[1]:d.strides[2],w=h?d.strides[2]:1,S=h?1:d.strides[1],L=this.readSync(e.dataId),N=this.readSync(i.dataId),C=d.values,R=0;R=r.inHeight)continue;for(var K=H*i.strides[0],j=D+_*f,q=0;q=r.inWidth)continue;for(var ne=K+X*i.strides[1],ie=j+ee*m,te=ne,re=0;re=r.inDepth)continue;for(var k=R*i.strides[0],W=w+D*e.strides[1],F=0;F=r.inHeight)continue;for(var j=k+_*i.strides[1],q=W+K*e.strides[2],G=0;G=r.inWidth)continue;for(var ie=j+ee*i.strides[2],te=q+ne*r.inChannels,re=ie,le=0;le=r.inHeight)continue;for(var R=N*i.strides[0],D=v+C*e.strides[1],k=0;k=r.inWidth)continue;for(var _=R+P*i.strides[1],K=D+H*r.inChannels,j=W,q=_,G=0;Ghe?he=We:r==="avg"&&(ye+=We,Oe++),isNaN(he))break}if(isNaN(he))break}if(isNaN(he))break}var me=le+F;S[me]=r==="avg"?ye/Oe:he}}}return w.toTensor()},t.prototype.avgPool3d=function(e,i){return ae(e,"avgPool3d"),this.pool3d(e,i,"avg").toFloat()},t.prototype.avgPool3dBackprop=function(e,i,r){ae([e,i],"avgPool3dBackprop");for(var a=r.strideDepth,s=r.strideHeight,o=r.strideWidth,l=r.filterDepth,u=r.filterHeight,c=r.filterWidth,h=r.dilationDepth,d=r.dilationHeight,p=r.dilationWidth,f=r.effectiveFilterDepth,m=r.effectiveFilterHeight,g=r.effectiveFilterWidth,v=f-1-r.padInfo.front,b=g-1-r.padInfo.left,w=m-1-r.padInfo.top,S=T.buffer(i.shape,"float32"),L=1/(l*u*c),N=this.bufferSync(e),C=0;C=r.outDepth||Math.floor(j)!==j)continue;for(var q=0;q=r.outHeight||Math.floor(G)!==G)continue;for(var Z=0;Z=r.outWidth||Math.floor(X)!==X)continue;var ee=N.get(C,j,G,X,R);_+=ee}}}S.set(_*L,C,D,k,W,R)}return S.toTensor()},t.prototype.maxPool3d=function(e,i){return ae(e,"maxPool3d"),this.pool3d(e,i,"max").toFloat()},t.prototype.maxPool3dPositions=function(e,i){for(var r=T.buffer(i.outShape,"int32"),a=i.strideDepth,s=i.strideHeight,o=i.strideWidth,l=i.dilationDepth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterDepth,d=i.effectiveFilterHeight,p=i.effectiveFilterWidth,f=i.padInfo.front,m=i.padInfo.top,g=i.padInfo.left,v=this.bufferSync(e),b=0;b=K&&(K=ie,j=G*d*p+X*d+ne)}r.set(j,b,S,R,F,w)}}}return r.toTensor()},t.prototype.maxPool3dBackprop=function(e,i,r,a){ae([i,r],"maxPool3dBackprop");for(var s=this.maxPool3dPositions(i,a),o=a.strideDepth,l=a.strideHeight,u=a.strideWidth,c=a.dilationDepth,h=a.dilationHeight,d=a.dilationWidth,p=a.effectiveFilterDepth,f=a.effectiveFilterHeight,m=a.effectiveFilterWidth,g=p-1-a.padInfo.front,v=m-1-a.padInfo.left,b=f-1-a.padInfo.top,w=T.buffer(i.shape,"float32"),S=this.bufferSync(s),L=this.bufferSync(e),N=0;N=a.outDepth||Math.floor(K)!==K)continue;for(var j=0;j=a.outHeight||Math.floor(q)!==q)continue;for(var G=0;G=a.outWidth||Math.floor(Z)!==Z)continue;var X=p*f*m-1-S.get(N,K,q,Z,C),ee=_*f*m+j*m+G,ne=X===ee?1:0;if(ne===0)continue;var ie=L.get(N,K,q,Z,C);H+=ie*ne}}}w.set(H,N,R,D,k,C)}return w.toTensor()},t.prototype.resizeBilinear=function(e,i,r,a){ae(e,"resizeBilinear");for(var s=e.shape,o=s[0],l=s[1],u=s[2],c=s[3],h=this.readSync(e.dataId),d=new Float32Array(T.util.sizeFromShape([o,i,r,c])),p=[a&&i>1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=0,g=p[0]/f[0],v=p[1]/f[1],b=0;b1?o-1:o,r&&d>1?l-1:l],m=[r&&h>1?h-1:h,r&&d>1?d-1:d],g=f[0]/m[0],v=f[1]/m[1],b=this.readSync(e.dataId),w=0,S=0;S1?l-1:l,a&&r>1?u-1:u],f=[a&&i>1?i-1:i,a&&r>1?r-1:r],m=p[0]/f[0],g=p[1]/f[1],v=0,b=0;b1?o-1:o,r&&d>1?l-1:l],g=[r&&h>1?h-1:h,r&&d>1?d-1:d],v=m[0]/g[0],b=m[1]/g[1],w=1/v,S=1/b,L=Math.ceil(w)*2+2,N=Math.ceil(S)*2+2,C=0;C=h)continue;var X=R+Z*e.strides[1],ee=Z*v,ne=Math.min(o-1,r?Math.round(ee):Math.floor(ee));if(D!==ne)continue;for(var ie=0;ie=d)continue;var re=X+te*e.strides[2],le=te*b,he=Math.min(l-1,r?Math.round(le):Math.floor(le));P===he&&(q+=f[re+j])}}p[H+j]=q}return T.tensor4d(p,i.shape,i.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){ae(e,"localResponseNormalization4D");var o=e.shape[3],l=o-1,u=this.readSync(e.dataId),c=e.size,h=new Float32Array(c);function d(g){for(var v=g%o,b=g-v+Math.max(0,v-i),w=g-v+Math.min(v+i,l),S=0;b<=w;b++){var L=u[b];S+=L*L}return S}for(var p=0;p=0&&o[l]1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});for(var a=e.shape[0],s=e.shape[1],o=e.shape[2],l=e.shape[3],u=s*i,c=o*i,h=l/(i*i),d=this.readSync(e.dataId),p=new Float32Array(a*u*c*h),f=0,m=0;m=u)continue;for(var P=f>1?(k-R)*(c-1)/(f-1):0,H=m>1?(W-D)*(h-1)/(m-1):0,_=0;_1?R*(c-1)+_*P:.5*(R+k)*(c-1);if(K<0||K>c-1){for(var j=0;j1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q1?D*(h-1)+j*H:.5*(D+W)*(h-1);if(ne<0||ne>h-1){for(var q=0;q=e.size/u)throw new Error("Invalid indices: "+m+" does not index into "+e.shape);for(var w=0;w=a/s)throw new Error("Invalid indices: "+v+" does not index into "+r);for(var L=0;Lo&&(o=u)}r[a]=o}return r}var U0=gr(function(n,t){return n*t}),Iz=jd(function(n,t,e,i){return{real:n*e-t*i,imag:n*i+t*e}}),B0=la(T.Multiply,U0,Iz),Az={kernelName:T.Multiply,backendName:"cpu",kernelFunc:B0};var z0=ua(function(n){return 1/Math.sqrt(n)}),Tz=ca(T.Rsqrt,z0),Nz={kernelName:T.Rsqrt,backendName:"cpu",kernelFunc:Tz};function P0(n,t,e,i,r){var a=T.slice_util.isSliceContinous(i,t,e),s=T.util.sizeFromShape(e),o=T.util.computeStrides(i);if(a){var l=T.slice_util.computeFlatOffset(t,o);return n.subarray(l,l+s)}for(var u=T.util.getTypedArrayFromDType(r,s),c=0;cj?j=ie:a==="avg"&&(q+=ie,G++)}if(isNaN(j))break}var te=F+P*w+C;g[te]=a==="avg"?q/G:j}return m}function V0(n,t,e,i,r,a){r===void 0&&(r=!1),a===void 0&&(a=!1);for(var s=T.buffer(i.outShape,"int32"),o=i.strideHeight,l=i.strideWidth,u=i.dilationHeight,c=i.dilationWidth,h=i.effectiveFilterHeight,d=i.effectiveFilterWidth,p=i.padInfo.top,f=i.padInfo.left,m=T.buffer(t,e,n),g=0;gk&&(k=K,r?W=a?((g*i.inHeight+F)*i.inWidth+H)*i.inChannels+v:(F*i.inWidth+H)*i.inChannels+v:W=P*d+_)}s.set(W,g,b,N,v)}}return s}function qz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;ae(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;T.util.assert(T.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=T.backend_util.computePool2DInfo(r.shape,a,s,u,o,l),h;if(c.filterWidth===1&&c.filterHeight===1&&T.util.arraysEqual(c.inShape,c.outShape))h=oa({inputs:{x:r},backend:e});else{var d=e.data.get(r.dataId).values,p=T.util.computeStrides(r.shape),f=Jd(d,r.shape,r.dtype,p,c,"avg");h=e.makeTensorInfo(c.outShape,r.dtype,f.values)}return h}var Gz={kernelName:T.AvgPool,backendName:"cpu",kernelFunc:qz};function Yz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;ae([r,a],"avgPoolBackprop");for(var o=i.filterSize,l=i.strides,u=i.pad,c=T.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=c.strideHeight,d=c.strideWidth,p=c.filterHeight,f=c.filterWidth,m=c.dilationHeight,g=c.dilationWidth,v=c.effectiveFilterHeight,b=c.effectiveFilterWidth,w=b-1-c.padInfo.left,S=v-1-c.padInfo.top,L=T.buffer(s.shape,"float32"),N=1/(p*f),C=e.data.get(r.dataId).values,R=T.buffer(r.shape,"float32",C),D=0;D=c.outHeight||Math.floor(j)!==j)continue;for(var q=0;q=c.outWidth||Math.floor(G)!==G)continue;var Z=R.get(D,j,G,k);_+=Z}}L.set(_*N,D,W,F,k)}return e.makeTensorInfo(L.shape,L.dtype,L.values)}var Kz={kernelName:T.AvgPoolBackprop,backendName:"cpu",kernelFunc:Yz};function jz(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=t.scale,s=t.offset,o=t.mean,l=t.variance;T.util.assert(o.shape.length===l.shape.length,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),T.util.assert(s==null||o.shape.length===s.shape.length,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),T.util.assert(a==null||o.shape.length===a.shape.length,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."}),ae([r,o,l,a,s],"batchNorm");var u=i.varianceEpsilon;u==null&&(u=.001);for(var c=e.data.get(r.dataId).values,h=e.data.get(o.dataId).values,d=e.data.get(l.dataId).values,p=a?e.data.get(a.dataId).values:new Float32Array([1]),f=s?e.data.get(s.dataId).values:new Float32Array([0]),m=new Float32Array(c.length),g=f.length,v=p.length,b=d.length,w=h.length,S=0,L=0,N=0,C=0,R=0;R=g&&(S=0),L>=w&&(L=0),N>=v&&(N=0),C>=b&&(C=0);return e.makeTensorInfo(r.shape,r.dtype,m)}var $z={kernelName:T.FusedBatchNorm,backendName:"cpu",kernelFunc:jz};var Xz=je(T.ClipByValue,function(n,t){var e=t;return n>e.clipValueMax?e.clipValueMax:n0});if(o.length===1)return o[0];var l=o.map(function(S){return S.shape});if(T.backend_util.assertParamsConsistent(l,a),o[0].dtype==="complex64"){var u=o.map(function(S){return ds({inputs:{input:S},backend:e})}),c=o.map(function(S){return Xo({inputs:{input:S},backend:e})}),h=fs({inputs:u,backend:e,attrs:{axis:r}}),d=fs({inputs:c,backend:e,attrs:{axis:r}}),p=Tn({inputs:{real:h,imag:d},backend:e});return u.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),c.forEach(function(S){return e.disposeIntermediateTensorInfo(S)}),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(d),p}var f=o.map(function(S){var L=T.util.sizeFromShape(S.shape.slice(a)),N=[-1,L];return Di({inputs:{x:S},backend:e,attrs:{shape:N}})});s=T.backend_util.computeOutShape(f.map(function(S){return S.shape}),1);var m=T.util.getTypedArrayFromDType(o[0].dtype,T.util.sizeFromShape(s));if(f[0].shape[0]===1){var g=0;f.forEach(function(S){var L=e.data.get(S.dataId).values,N=T.util.sizeFromShape(S.shape);m.set(L,g),g+=N})}else{var v=0;f.forEach(function(S){for(var L=e.data.get(S.dataId).values,N=0,C=0;C=0&&re=0&&heie&&(ie=ke)}}}var Pe=T.util.locToIndex([q,G,X,ne],K,T.util.computeStrides(H));j[Pe]=ie}var _e=h.write(T.util.toTypedArray(j,a.dtype),H,a.dtype);return{dataId:_e,shape:H,dtype:a.dtype}}};var sP={kernelName:T.Dilation2DBackpropFilter,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,N=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropFilter+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(s.shape,s.dtype),K=0;K=0&&re=0&&heee&&(ee=ye,ne=te,ie=le)}}}_[ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),s.shape,s.dtype);return{dataId:Oe,shape:s.shape,dtype:s.dtype}}};var oP={kernelName:T.Dilation2DBackpropInput,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.x,s=r.filter,o=r.dy,l=i,u=l.strides,c=l.pad,h=l.dilations,d=e,p=T.util.toNestedArray(a.shape,d.data.get(a.dataId).values),f=T.util.toNestedArray(s.shape,d.data.get(s.dataId).values),m=T.backend_util.computeDilation2DInfo(a.shape,s.shape,u,c,"NHWC",h),g=m.batchSize,v=m.inHeight,b=m.inWidth,w=m.inChannels,S=m.outHeight,L=m.outWidth,N=m.padInfo,C=m.strideHeight,R=m.strideWidth,D=m.filterHeight,k=m.filterWidth,W=m.dilationHeight,F=m.dilationWidth,P=m.outShape;T.util.assert(o.rank===P.length,function(){return"Error in "+T.Dilation2DBackpropInput+", dy "+("must have the same rank as output "+P.length+", but got ")+(""+o.rank)});for(var H=T.util.toNestedArray(P,d.data.get(o.dataId).values),_=T.util.makeZerosNestedTypedArray(a.shape,a.dtype),K=0;K=0&&re=0&&heee&&(ee=ye,ne=re,ie=he)}}}_[K][ne][ie][X]+=H[K][j][G][X]}var Oe=d.write(T.util.toTypedArray(_,a.dtype),a.shape,a.dtype);return{dataId:Oe,shape:a.shape,dtype:a.dtype}}};var lP=gr(function(n,t){return n/t}),uP=la(T.Div,lP),Zd={kernelName:T.Div,backendName:"cpu",kernelFunc:uP};var cP=je(T.Elu,function(n){return n>=0?n:Math.exp(n)-1}),hP={kernelName:T.Elu,backendName:"cpu",kernelFunc:cP};var dP=T.backend_util.ERF_P,pP=T.backend_util.ERF_A1,fP=T.backend_util.ERF_A2,mP=T.backend_util.ERF_A3,gP=T.backend_util.ERF_A4,vP=T.backend_util.ERF_A5,yP=je(T.Erf,function(n){var t=Math.sign(n),e=Math.abs(n),i=1/(1+dP*e);return t*(1-((((vP*i+gP)*i+mP)*i+fP)*i+pP)*i*Math.exp(-e*e))}),bP={kernelName:T.Erf,backendName:"cpu",kernelFunc:yP};function q0(n,t,e){for(var i=n.shape,r=i[0],a=i[1],s=e.data.get(n.dataId),o=s.complexTensorInfos.real,l=s.complexTensorInfos.imag,u=[r,a],c=T.util.sizeFromShape(u),h=T.util.getTypedArrayFromDType("float32",c),d=T.util.getTypedArrayFromDType("float32",c),p=0;p=0&&N=d.outHeight||Math.floor(q)!==q)continue;for(var G=0;G=d.outWidth||Math.floor(Z)!==Z)continue;var X=w*S-1-f.get(k,q,Z,W),ee=j*S+G,ne=X===ee?1:0;if(ne===0)continue;var ie=D.get(k,q,Z,W);K+=ie*ne}}C.set(K,k,F,P,W)}return e.makeTensorInfo(C.shape,C.dtype,C.values)}var HP={kernelName:T.MaxPoolBackprop,backendName:"cpu",kernelFunc:MP};function VP(n,t,e,i,r){var a=T.util.computeStrides(t),s=Jd(n,t,e,a,r,"max"),o=V0(n,t,e,r,!0,i);return[s.values,o.values]}var qP={kernelName:T.MaxPoolWithArgmax,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e,s=a.filterSize,o=a.strides,l=a.pad,u=a.includeBatchInIndex,c=i;ae(r,"MaxPoolWithArgmax");var h=c.data.get(r.dataId).values,d=T.backend_util.computePool2DInfo(r.shape,s,o,[1,1],l),p=VP(h,r.shape,r.dtype,u,d),f=p[0],m=p[1],g=c.write(f,d.outShape,r.dtype),v=c.write(m,d.outShape,r.dtype);return[{dataId:g,shape:d.outShape,dtype:r.dtype},{dataId:v,shape:d.outShape,dtype:"int32"}]}};var GP=T.kernel_impls.nonMaxSuppressionV4Impl,YP={kernelName:T.NonMaxSuppressionV4,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.padToMaxOutputSize,d=e;ae(a,"NonMaxSuppressionPadded");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=GP(p,f,l,u,c,h),g=m.selectedIndices,v=m.validOutputs;return[g,v]}};var KP=T.kernel_impls.nonMaxSuppressionV5Impl,jP={kernelName:T.NonMaxSuppressionV5,backendName:"cpu",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t,a=r.boxes,s=r.scores,o=i,l=o.maxOutputSize,u=o.iouThreshold,c=o.scoreThreshold,h=o.softNmsSigma,d=e;ae(a,"NonMaxSuppressionWithScore");var p=d.data.get(a.dataId).values,f=d.data.get(s.dataId).values,m=l,g=u,v=c,b=h,w=KP(p,f,m,g,v,b),S=w.selectedIndices,L=w.selectedScores;return[S,L]}};var $P=gr(function(n,t){return n!==t?1:0}),XP=la(T.NotEqual,$P,null,"bool"),JP={kernelName:T.NotEqual,backendName:"cpu",kernelFunc:XP};function ZP(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=i.paddings,s=i.constantValue;ae(r,"pad");var o=a.map(function(N,C){return N[0]+r.shape[C]+N[1]}),l=a.map(function(N){return N[0]}),u=e.data.get(r.dataId).values,c=T.util.sizeFromShape(r.shape),h=r.shape.length,d=T.util.computeStrides(r.shape),p=T.util.sizeFromShape(o),f=o.length,m=T.util.computeStrides(o),g=T.util.getTypedArrayFromDType(r.dtype,p);s!==0&&g.fill(s);for(var v=0;v=0&&j=0&&q.5?Math.ceil(n):t%2===0?t:t+1}),i_={kernelName:T.Round,backendName:"cpu",kernelFunc:n_};var r_=T.backend_util.SELU_SCALEALPHA,a_=T.backend_util.SELU_SCALE,s_=je(T.Selu,function(n){return n>=0?a_*n:r_*(Math.exp(n)-1)}),o_={kernelName:T.Selu,backendName:"cpu",kernelFunc:s_};var l_=je(T.Sigmoid,function(n){return 1/(1+Math.exp(-n))}),u_={kernelName:T.Sigmoid,backendName:"cpu",kernelFunc:l_};var c_=je(T.Sign,function(n){return n<0?-1:n>0?1:0}),h_={kernelName:T.Sign,backendName:"cpu",kernelFunc:c_};var d_=je(T.Sin,function(n){return Math.sin(n)}),p_={kernelName:T.Sin,backendName:"cpu",kernelFunc:d_};var f_=je(T.Sinh,function(n){return Math.sinh(n)}),m_={kernelName:T.Sinh,backendName:"cpu",kernelFunc:f_};var g_=11920928955078125e-23,Y0=Math.log(g_)+2,v_=je(T.Softplus,function(n){var t=n>-Y0,e=n0?1:e.alpha}),R_={kernelName:T.Step,backendName:"cpu",kernelFunc:C_};var O_=je(T.Tan,function(n){return Math.tan(n)}),E_={kernelName:T.Tan,backendName:"cpu",kernelFunc:O_};var D_=je(T.Tanh,function(n){return Math.tanh(n)}),k_={kernelName:T.Tanh,backendName:"cpu",kernelFunc:D_};function F_(n){var t=n.inputs,e=n.attrs,i=n.backend,r=e.axis,a=t.x;ae(a,"unique");var s=i.data.get(a.dataId).values,o=H0(s,r,a.shape,a.dtype),l=o.outputValues,u=o.outputShape,c=o.indices;return[i.makeTensorInfo(u,a.dtype,l),i.makeTensorInfo([c.length],"int32",c)]}var W_={kernelName:T.Unique,backendName:"cpu",kernelFunc:F_};var U_=[sz,kz,Wz,dz,Bz,Pz,Mz,Vz,Gz,Kz,$z,cz,fz,Jz,oz,eP,nP,rP,aP,oP,sP,Zd,hP,bP,gz,yz,AP,TP,wz,lz,xP,Zz,RP,EP,kP,Lz,WP,BP,_P,HP,qP,zP,Az,YP,jP,JP,G0,uz,e_,Qz,t_,i_,Nz,o_,u_,h_,p_,m_,xz,y_,S_,I_,A_,x_,R_,Rz,E_,k_,b_,W_];for(var ep=0,j0=U_;ep{"use strict";Object.defineProperty(Mn,"__esModule",{value:!0});var x=rr();var tp=function(n,t){return tp=Object.setPrototypeOf||{__proto__:[]}instanceof Array&&function(e,i){e.__proto__=i}||function(e,i){for(var r in i)i.hasOwnProperty(r)&&(e[r]=i[r])},tp(n,t)};function z_(n,t){tp(n,t);function e(){this.constructor=n}n.prototype=t===null?Object.create(t):(e.prototype=t.prototype,new e)}function Jo(n,t,e,i){function r(a){return a instanceof e?a:new e(function(s){s(a)})}return new(e||(e=Promise))(function(a,s){function o(c){try{u(i.next(c))}catch(h){s(h)}}function l(c){try{u(i.throw(c))}catch(h){s(h)}}function u(c){c.done?a(c.value):r(c.value).then(o,l)}u((i=i.apply(n,t||[])).next())})}function Zo(n,t){var e={label:0,sent:function(){if(a[0]&1)throw a[1];return a[1]},trys:[],ops:[]},i,r,a,s;return s={next:o(0),throw:o(1),return:o(2)},typeof Symbol=="function"&&(s[Symbol.iterator]=function(){return this}),s;function o(u){return function(c){return l([u,c])}}function l(u){if(i)throw new TypeError("Generator is already executing.");for(;e;)try{if(i=1,r&&(a=u[0]&2?r.return:u[0]?r.throw||((a=r.return)&&a.call(r),0):r.next)&&!(a=a.call(r,u[1])).done)return a;(r=0,a)&&(u=[u[0]&2,a.value]);switch(u[0]){case 0:case 1:a=u;break;case 4:return e.label++,{value:u[1],done:!1};case 5:e.label++,r=u[1],u=[0];continue;case 7:u=e.ops.pop(),e.trys.pop();continue;default:if(!(a=e.trys,a=a.length>0&&a[a.length-1])&&(u[0]===6||u[0]===2)){e=0;continue}if(u[0]===3&&(!a||u[1]>a[0]&&u[1]e||t>e){var i="["+n+"x"+t+"]",r="["+e+"x"+e+"]";throw new Error("Requested texture size "+i+" greater than WebGL maximum on this browser / GPU "+r+".")}}function oS(n){return ai(n,function(){return n.createFramebuffer()},"Unable to create WebGLFramebuffer.")}function rp(n,t,e,i,r,a,s){var o=n.getAttribLocation(t,e);return o===-1?!1:(ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,i)}),ce(n,function(){return n.vertexAttribPointer(o,r,n.FLOAT,!1,a,s)}),ce(n,function(){return n.enableVertexAttribArray(o)}),!0)}function uS(n,t,e){lS(n,e),ce(n,function(){return n.activeTexture(n.TEXTURE0+e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)})}function $_(n,t){lS(n,t),ce(n,function(){return n.activeTexture(n.TEXTURE0+t)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function cS(n,t,e){return ai(n,function(){return n.getUniformLocation(t,e)},'uniform "'+e+'" not present in program.')}function hS(n,t,e){return n.getUniformLocation(t,e)}function dS(n,t,e,i){ce(n,function(){return uS(n,t,i)}),ce(n,function(){return n.uniform1i(e,i)})}function X_(n){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,null)}),ce(n,function(){return n.viewport(0,0,n.canvas.width,n.canvas.height)}),ce(n,function(){return n.scissor(0,0,n.canvas.width,n.canvas.height)})}function el(n,t,e){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,e)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,t,0)})}function ap(n,t){ce(n,function(){return n.bindFramebuffer(n.FRAMEBUFFER,t)}),ce(n,function(){return n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,null,0)})}function bs(n){var t=n.checkFramebufferStatus(n.FRAMEBUFFER);if(t!==n.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+pS(n,t))}function pS(n,t){switch(t){case n.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case n.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case n.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return"unknown error "+t}}function ai(n,t,e){var i=ce(n,function(){return t()});if(i==null)throw new Error(e);return i}function lS(n,t){var e=n.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,i=t+n.TEXTURE0;if(ie){var r="[gl.TEXTURE0, gl.TEXTURE"+e+"]";throw new Error("textureUnit must be in "+r+".")}}function yr(n,t){return t===void 0&&(t=2),x.util.sizeFromShape(n.slice(0,n.length-t))}function br(n){if(n.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[n.length>1?n[n.length-2]:1,n[n.length-1]]}function tl(n){var t=[1,1,1],e=n.length===0||n.length===1&&n[0]===1;return e||(t=[yr(n)].concat(br(n))),t}function fS(n,t){var e;t===void 0&&(t=!1);var i=x.env().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(t&&(i=i*2,n=n.map(function(u,c){return c>=n.length-2?x.util.nearestLargerEven(n[c]):n[c]}),n.length===1&&(n=[2,n[0]])),n.length!==2){var r=x.util.squeezeShape(n);n=r.newShape}var a=x.util.sizeFromShape(n);if(n.length<=1&&a<=i)return[1,a];if(n.length===2&&n[0]<=i&&n[1]<=i)return n;if(n.length===3&&n[0]*n[1]<=i&&n[2]<=i)return[n[0]*n[1],n[2]];if(n.length===3&&n[0]<=i&&n[1]*n[2]<=i)return[n[0],n[1]*n[2]];if(n.length===4&&n[0]*n[1]*n[2]<=i&&n[3]<=i)return[n[0]*n[1]*n[2],n[3]];if(n.length===4&&n[0]<=i&&n[1]*n[2]*n[3]<=i)return[n[0],n[1]*n[2]*n[3]];if(t){var s=yr(n),o=2,l=2;return n.length&&(e=br(n),o=e[0],l=e[1]),a=s*(o/2)*(l/2),x.util.sizeToSquarishShape(a).map(function(u){return u*2})}return x.util.sizeToSquarishShape(a)}function nl(n){return n%2===0}function ws(n,t){if(n=n.slice(-2),t=t.slice(-2),x.util.arraysEqual(n,t))return!0;if(!n.length||!t.length)return!0;if(n[0]===0||n[1]===0||t[0]===0||t[1]===0)return!0;if(n.length!==t.length){var e=n.slice(-1)[0],i=t.slice(-1)[0];if(e===i)return!0;if(nl(e)&&nl(i)&&(n[0]===1||t[0]===1))return!0}return n[1]===t[1]&&nl(n[0])&&nl(t[0])}var il,rl;function mS(n){if(il==null){var t=Hn(n);il=t.getParameter(t.MAX_TEXTURE_SIZE)}return il}function J_(){il=null}function Z_(){rl=null}function gS(n){if(rl==null){var t=Hn(n);rl=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,rl)}function vS(n){if(n===0)return 0;var t,e=Hn(n);return sn(e,"EXT_disjoint_timer_query_webgl2")&&n===2?t=2:sn(e,"EXT_disjoint_timer_query")?t=1:t=0,t}function sn(n,t){var e=n.getExtension(t);return e!=null}function sp(n){try{var t=Hn(n);if(t!=null)return!0}catch(e){return console.log("Error when getting WebGL context: ",e),!1}return!1}function yS(n){if(n===0)return!1;var t=Hn(n);if(n===1){if(!sn(t,"OES_texture_float"))return!1}else if(!sn(t,"EXT_color_buffer_float"))return!1;var e=op(t);return e}function bS(n){if(n===0)return!1;var t=Hn(n);if(n===1){if(!sn(t,"OES_texture_float"))return!1;if(!sn(t,"WEBGL_color_buffer_float"))return!1}else{if(sn(t,"EXT_color_buffer_float"))return op(t);var e="EXT_color_buffer_half_float";if(sn(t,e)){var i=t.getExtension(e);return Q_(t,i)}return!1}var r=op(t);return r}function op(n){var t=ip(n),e=n.createTexture();n.bindTexture(n.TEXTURE_2D,e);var i=1,r=1;n.texImage2D(n.TEXTURE_2D,0,t.internalFormatFloat,i,r,0,t.textureFormatFloat,t.textureTypeFloat,null);var a=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,a),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,e,0);var s=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(e),n.deleteFramebuffer(a),s}function Q_(n,t){var e=ip(n,t),i=n.createTexture();n.bindTexture(n.TEXTURE_2D,i);var r=1,a=1;n.texImage2D(n.TEXTURE_2D,0,e.internalFormatHalfFloat,r,a,0,e.textureFormatFloat,e.textureTypeHalfFloat,null);var s=n.createFramebuffer();n.bindFramebuffer(n.FRAMEBUFFER,s),n.framebufferTexture2D(n.FRAMEBUFFER,n.COLOR_ATTACHMENT0,n.TEXTURE_2D,i,0);var o=n.checkFramebufferStatus(n.FRAMEBUFFER)===n.FRAMEBUFFER_COMPLETE;return n.bindTexture(n.TEXTURE_2D,null),n.bindFramebuffer(n.FRAMEBUFFER,null),n.deleteTexture(i),n.deleteFramebuffer(s),o}function wS(n){if(n!==2)return!1;var t=Hn(n),e=t.fenceSync!=null;return e}function da(n,t){Array.isArray(n)||(n=[n]),n.forEach(function(e){e!=null&&x.util.assert(e.dtype!=="complex64",function(){return t+" does not support complex64 tensors in the WebGL backend."})})}var eM={__proto__:null,callAndCheck:ce,canBeRepresented:Z0,getWebGLErrorMessage:J0,getExtensionOrThrow:ys,createVertexShader:Q0,createFragmentShader:eS,createProgram:tS,linkProgram:nS,validateProgram:Qo,createStaticVertexBuffer:iS,createStaticIndexBuffer:rS,getNumChannels:j_,createTexture:aS,validateTextureSize:sS,createFramebuffer:oS,bindVertexBufferToProgramAttribute:rp,bindTextureUnit:uS,unbindTextureUnit:$_,getProgramUniformLocationOrThrow:cS,getProgramUniformLocation:hS,bindTextureToProgramUniformSampler:dS,bindCanvasToFramebuffer:X_,bindColorTextureToFramebuffer:el,unbindColorTextureFromFramebuffer:ap,validateFramebuffer:bs,getFramebufferErrorMessage:pS,getBatchDim:yr,getRowsCols:br,getShapeAs3D:tl,getTextureShapeFromLogicalShape:fS,isReshapeFree:ws,getWebGLMaxTextureSize:mS,resetMaxTextureSize:J_,resetMaxTexturesInShader:Z_,getMaxTexturesInShader:gS,getWebGLDisjointQueryTimerVersion:vS,hasExtension:sn,isWebGLVersionEnabled:sp,isCapableOfRenderingToFloatTexture:yS,isDownloadFloatTextureEnabled:bS,isWebGLFenceEnabled:wS,assertNotComplex:da};var we=x.env();we.registerFlag("HAS_WEBGL",function(){return we.getNumber("WEBGL_VERSION")>0});we.registerFlag("WEBGL_VERSION",function(){return sp(2)?2:sp(1)?1:0});we.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",function(){return!1});we.registerFlag("WEBGL_BUFFER_SUPPORTED",function(){return we.get("WEBGL_VERSION")===2});we.registerFlag("WEBGL_CPU_FORWARD",function(){return!0});we.registerFlag("WEBGL_FORCE_F16_TEXTURES",function(){return!1});we.registerFlag("WEBGL_PACK",function(){return we.getBool("HAS_WEBGL")});we.registerFlag("WEBGL_PACK_NORMALIZATION",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_CLIP",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_DEPTHWISECONV",function(){return!1});we.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_PACK_REDUCE",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_LAZILY_UNPACK",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_CONV_IM2COL",function(){return we.getBool("WEBGL_PACK")});we.registerFlag("WEBGL_MAX_TEXTURE_SIZE",function(){return mS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",function(){return gS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",function(){var n=we.getNumber("WEBGL_VERSION");return n===0?0:vS(n)});we.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",function(){return we.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!x.device_util.isMobile()});we.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",function(){return yS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",function(){return we.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:we.getBool("WEBGL_RENDER_FLOAT32_CAPABLE")});we.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",function(){return bS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_FENCE_API_ENABLED",function(){return wS(we.getNumber("WEBGL_VERSION"))});we.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",function(){var n=we.getBool("WEBGL_RENDER_FLOAT32_ENABLED");return n?4:0});we.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",function(){return-1},function(n){if(n<0&&n!==-1)throw new Error("WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never "+("delete) or at least 0, but got "+n+"."))});function tM(n){const t=new Float32Array(n.length);for(let e=0;e{const s=x.backend_util.assertAndGetBroadcastShape(t,e),o=s.length,l=x.util.computeStrides(s),u=x.util.sizeFromShape(s),c=x.util.getTypedArrayFromDType(a,u),h=t.length,d=e.length,p=x.util.computeStrides(t),f=x.util.computeStrides(e),m=x.backend_util.getBroadcastDims(t,s),g=x.backend_util.getBroadcastDims(e,s);if(m.length+g.length===0)for(let v=0;vw[C]=0);const S=x.util.locToIndex(w,h,p),L=b.slice(-d);g.forEach(C=>L[C]=0);const N=x.util.locToIndex(L,d,f);c[v]=n(i[S],r[N])}return[c,s]}}const nM=lp((n,t)=>n+t);function pa(n){return(t,e,i)=>{const r=x.util.getTypedArrayFromDType(e,t.length);for(let a=0;aMath.ceil(n));const rM=pa(n=>Math.exp(n));const aM=pa(n=>Math.expm1(n));const sM=pa(n=>Math.floor(n));const oM=pa(n=>Math.log(n));function lM(n,t,e,i){const r=x.util.getTypedArrayFromDType(i,x.util.sizeFromShape(e));for(let a=0;ao&&(o=u)}r[a]=o}return r}const uM=lp((n,t)=>n*t);const cM=pa(n=>1/Math.sqrt(n));function hM(n,t,e,i,r){const a=x.slice_util.isSliceContinous(i,t,e),s=x.util.sizeFromShape(e),o=x.util.computeStrides(i);if(a){const u=x.slice_util.computeFlatOffset(t,o);return n.subarray(u,u+s)}const l=x.util.getTypedArrayFromDType(r,s);for(let u=0;um+t[g]),f=x.util.locToIndex(p,i.length,o);l[u]=n[f]}return l}const dM=lp((n,t)=>n-t);function pM(n,t,e,i,r){const a=t.length,s=x.util.sizeFromShape(t),o=x.util.computeStrides(t),l=x.util.computeStrides(r),u=x.util.getTypedArrayFromDType(e,x.util.sizeFromShape(r));for(let c=0;c{for(let g=0;g":"<",l=i?"inOffset + i;":"round(getBestIndicesA(batch, inOffset + i));";this.userCode=` + void main() { + ivec2 coords = getOutputCoords(); + int batch = coords[0]; + int outIdx = coords[1]; + int inOffset = outIdx * `+r+`; + + int bestIndex = inOffset; + float bestValue = getA(batch, bestIndex); + + for (int i = 0; i < `+r+`; i++) { + int inIdx = `+l+`; + float candidate = getA(batch, inIdx); + if (candidate `+o+` bestValue) { + bestValue = candidate; + bestIndex = inIdx; + } + } + setOutput(float(bestIndex)); + } + `}return n}();function LS(n,t){return["x","y","z","w","u","v"].slice(0,t).map(function(e){return n+"."+e})}function Jt(n,t){return t===1?[n]:LS(n,t)}function EM(n,t){if(n===1)return"rc";for(var e="",i=0;i 0.0 || val < 0.0) ? false : val != 0.0; + } + + bvec4 isnan_custom(vec4 val) { + return bvec4(isnan_custom(val.x), + isnan_custom(val.y), isnan_custom(val.z), isnan_custom(val.w)); + } + + #define isnan(value) isnan_custom(value) + `,l="",u=` + #define round(value) newRound(value) + int newRound(float value) { + return int(floor(value + 0.5)); + } + + ivec4 newRound(vec4 value) { + return ivec4(floor(value + vec4(0.5))); + } + `):(n="",t="attribute",e="varying",i="varying",r="texture2D",a="gl_FragColor",s="",o=` + #define isnan(value) isnan_custom(value) + bool isnan_custom(float val) { + return (val > 0. || val < 1. || val == 0.) ? false : true; + } + bvec4 isnan_custom(vec4 val) { + return bvec4(isnan(val.x), isnan(val.y), isnan(val.z), isnan(val.w)); + } + `,l=` + uniform float INFINITY; + + bool isinf(float val) { + return abs(val) == INFINITY; + } + bvec4 isinf(vec4 val) { + return equal(abs(val), vec4(INFINITY)); + } + `,u=` + int round(float value) { + return int(floor(value + 0.5)); + } + + ivec4 round(vec4 value) { + return ivec4(floor(value + vec4(0.5))); + } + `),{version:n,attribute:t,varyingVs:e,varyingFs:i,texture2D:r,output:a,defineOutput:s,defineSpecialNaN:o,defineSpecialInf:l,defineRound:u}}function wr(n,t,e){e===void 0&&(e="index");var i=x.util.computeStrides(t);return i.map(function(r,a){var s="int "+n[a]+" = "+e+" / "+r,o=a===i.length-1?"int "+n[a+1]+" = "+e+" - "+n[a]+" * "+r:"index -= "+n[a]+" * "+r;return s+"; "+o+";"}).join("")}function up(n){var t=x.util.computeStrides(n).map(function(e){return e.toString()});return` + int getFlatIndex(ivec3 coords) { + return coords.x * `+t[0]+" + coords.y * "+t[1]+` + coords.z; + } +`}var IS=` + const float FLOAT_MAX = 1.70141184e38; + const float FLOAT_MIN = 1.17549435e-38; + + lowp vec4 encode_float(highp float v) { + if (isnan(v)) { + return vec4(255, 255, 255, 255); + } + + highp float av = abs(v); + + if(av < FLOAT_MIN) { + return vec4(0.0, 0.0, 0.0, 0.0); + } else if(v > FLOAT_MAX) { + return vec4(0.0, 0.0, 128.0, 127.0) / 255.0; + } else if(v < -FLOAT_MAX) { + return vec4(0.0, 0.0, 128.0, 255.0) / 255.0; + } + + highp vec4 c = vec4(0,0,0,0); + + highp float e = floor(log2(av)); + highp float m = exp2(fract(log2(av))) - 1.0; + + c[2] = floor(128.0 * m); + m -= c[2] / 128.0; + c[1] = floor(32768.0 * m); + m -= c[1] / 32768.0; + c[0] = floor(8388608.0 * m); + + highp float ebias = e + 127.0; + c[3] = floor(ebias / 2.0); + ebias -= c[3] * 2.0; + c[2] += floor(ebias) * 128.0; + + c[3] += 128.0 * step(0.0, -v); + + return c / 255.0; + } +`;var AS=x.backend_util.getBroadcastDims;function _M(n,t,e,i){var r=[];n.forEach(function(f){var m=x.util.sizeFromShape(f.shapeInfo.logicalShape);f.shapeInfo.isUniform?r.push("uniform float "+f.name+(m>1?"["+m+"]":"")+";"):(r.push("uniform sampler2D "+f.name+";"),r.push("uniform int offset"+f.name+";"))});var a=r.join(` +`),s=n.map(function(f){return DM(f,t,i)}).join(` +`),o=t.texShape,l=Wt(),u=WM(l),c,h,d=zM(l);t.isPacked?(c=kM(t.logicalShape,o),h=BM(l)):(c=FM(t.logicalShape,o),h=UM(l)),i&&(d+=PM);var p=[d,u,h,a,c,s,e].join(` +`);return p}function fa(n){var t=n.shapeInfo.logicalShape;switch(t.length){case 0:return MM(n);case 1:return HM(n);case 2:return VM(n);case 3:return qM(n);case 4:return GM(n);case 5:return YM(n);case 6:return KM(n);default:throw new Error(t.length+"-D input sampling is not yet supported")}}function TS(n){var t=n.shapeInfo.logicalShape;switch(t.length){case 0:return jM(n);case 1:return $M(n);case 2:return XM(n);case 3:return JM(n);default:return ZM(n)}}function DM(n,t,e){e===void 0&&(e=!1);var i="";e?i+=TS(n):i+=fa(n);var r=n.shapeInfo.logicalShape,a=t.logicalShape;return r.length<=a.length&&(e?i+=QM(n,t):i+=e5(n,t)),i}function kM(n,t){switch(n.length){case 0:return NS();case 1:return t5(n,t);case 2:return r5(n,t);case 3:return n5(n,t);default:return i5(n,t)}}function FM(n,t){switch(n.length){case 0:return NS();case 1:return a5(n,t);case 2:return c5(n,t);case 3:return s5(n,t);case 4:return o5(n,t);case 5:return l5(n,t);case 6:return u5(n,t);default:throw new Error(n.length+"-D output sampling is not yet supported")}}function WM(n){return` + float sampleTexture(sampler2D textureSampler, vec2 uv) { + return `+n.texture2D+`(textureSampler, uv).r; + } + `}function UM(n){return` + void setOutput(float val) { + `+n.output+` = vec4(val, 0, 0, 0); + } + `}function BM(n){return` + void setOutput(vec4 val) { + `+n.output+` = val; + } + `}function zM(n){var t=n.version+` + precision highp float; + precision highp int; + precision highp sampler2D; + `+n.varyingFs+` vec2 resultUV; + `+n.defineOutput+` + const vec2 halfCR = vec2(0.5, 0.5); + + struct ivec5 + { + int x; + int y; + int z; + int w; + int u; + }; + + struct ivec6 + { + int x; + int y; + int z; + int w; + int u; + int v; + }; + + uniform float NAN; + `+n.defineSpecialNaN+` + `+n.defineSpecialInf+` + `+n.defineRound+` + + int imod(int x, int y) { + return x - y * (x / y); + } + + int idiv(int a, int b, float sign) { + int res = a / b; + int mod = imod(a, b); + if (sign < 0. && mod != 0) { + res -= 1; + } + return res; + } + + //Based on the work of Dave Hoskins + //https://www.shadertoy.com/view/4djSRW + #define HASHSCALE1 443.8975 + float random(float seed){ + vec2 p = resultUV * seed; + vec3 p3 = fract(vec3(p.xyx) * HASHSCALE1); + p3 += dot(p3, p3.yzx + 19.19); + return fract((p3.x + p3.y) * p3.z); + } + + `+h5+` + `+d5+` + `+p5+` + `;return t}var h5=` +vec2 uvFromFlat(int texNumR, int texNumC, int index) { + int texR = index / texNumC; + int texC = index - texR * texNumC; + return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR); +} +vec2 packedUVfrom1D(int texNumR, int texNumC, int index) { + int texelIndex = index / 2; + int texR = texelIndex / texNumC; + int texC = texelIndex - texR * texNumC; + return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR); +} +`,d5=` +vec2 packedUVfrom2D(int texelsInLogicalRow, int texNumR, + int texNumC, int row, int col) { + int texelIndex = (row / 2) * texelsInLogicalRow + (col / 2); + int texR = texelIndex / texNumC; + int texC = texelIndex - texR * texNumC; + return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR); +} +`,p5=` +vec2 packedUVfrom3D(int texNumR, int texNumC, + int texelsInBatch, int texelsInLogicalRow, int b, + int row, int col) { + int index = b * texelsInBatch + (row / 2) * texelsInLogicalRow + (col / 2); + int texR = index / texNumC; + int texC = index - texR * texNumC; + return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR); +} +`,PM=` + float getChannel(vec4 frag, vec2 innerDims) { + vec2 modCoord = mod(innerDims, 2.); + return modCoord.x == 0. ? + (modCoord.y == 0. ? frag.r : frag.g) : + (modCoord.y == 0. ? frag.b : frag.a); + } + float getChannel(vec4 frag, int dim) { + float modCoord = mod(float(dim), 2.); + return modCoord == 0. ? frag.r : frag.g; + } +`;function NS(){return` + int getOutputCoords() { + return 0; + } + `}function t5(n,t){var e=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)];return e[0]===1?` + int getOutputCoords() { + return 2 * int(resultUV.x * `+e[1]+`.0); + } + `:e[1]===1?` + int getOutputCoords() { + return 2 * int(resultUV.y * `+e[0]+`.0); + } + `:` + int getOutputCoords() { + ivec2 resTexRC = ivec2(resultUV.yx * + vec2(`+e[0]+", "+e[1]+`)); + return 2 * (resTexRC.x * `+e[1]+` + resTexRC.y); + } + `}function a5(n,t){return t[0]===1?` + int getOutputCoords() { + return int(resultUV.x * `+t[1]+`.0); + } + `:t[1]===1?` + int getOutputCoords() { + return int(resultUV.y * `+t[0]+`.0); + } + `:` + int getOutputCoords() { + ivec2 resTexRC = ivec2(resultUV.yx * + vec2(`+t[0]+", "+t[1]+`)); + return resTexRC.x * `+t[1]+` + resTexRC.y; + } + `}function n5(n,t){var e=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)],i=Math.ceil(n[2]/2),r=i*Math.ceil(n[1]/2);return` + ivec3 getOutputCoords() { + ivec2 resTexRC = ivec2(resultUV.yx * + vec2(`+e[0]+", "+e[1]+`)); + int index = resTexRC.x * `+e[1]+` + resTexRC.y; + + int b = index / `+r+`; + index -= b * `+r+`; + + int r = 2 * (index / `+i+`); + int c = imod(index, `+i+`) * 2; + + return ivec3(b, r, c); + } + `}function s5(n,t){var e=wr(["r","c","d"],n);return` + ivec3 getOutputCoords() { + ivec2 resTexRC = ivec2(resultUV.yx * + vec2(`+t[0]+", "+t[1]+`)); + int index = resTexRC.x * `+t[1]+` + resTexRC.y; + `+e+` + return ivec3(r, c, d); + } + `}function i5(n,t){for(var e=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)],i=Math.ceil(n[n.length-1]/2),r=i*Math.ceil(n[n.length-2]/2),a=r,s="",o="b, r, c",l=2;l=1?c="coords = 0;":c=o.map(function(S){return"coords."+h[S+u]+" = 0;"}).join(` +`);var d="";s<2&&a>0?d="coords":d=n.shapeInfo.logicalShape.map(function(S,L){return"coords."+h[L+u]}).join(", ");var p="return outputValue;",f=x.util.sizeFromShape(n.shapeInfo.logicalShape),m=f===1,g=x.util.sizeFromShape(t.logicalShape),v=g===1;if(a===1&&!m&&!v)p=` + return vec4(outputValue.xy, outputValue.xy); + `;else if(m&&!v)s===1?p=` + return vec4(outputValue.x, outputValue.x, 0., 0.); + `:p=` + return vec4(outputValue.x); + `;else if(o.length){var b=a-2,w=a-1;o.indexOf(b)>-1&&o.indexOf(w)>-1?p="return vec4(outputValue.x);":o.indexOf(b)>-1?p="return vec4(outputValue.x, outputValue.y, outputValue.x, outputValue.y);":o.indexOf(w)>-1&&(p="return vec4(outputValue.xx, outputValue.zz);")}return` + vec4 `+r+`() { + `+l+` coords = getOutputCoords(); + `+c+` + vec4 outputValue = get`+i+"("+d+`); + `+p+` + } + `}function e5(n,t){var e=n.name,i=e.charAt(0).toUpperCase()+e.slice(1),r="get"+i+"AtOutCoords",a=t.texShape,s=n.shapeInfo.texShape,o=n.shapeInfo.logicalShape.length,l=t.logicalShape.length;if(!n.shapeInfo.isUniform&&o===l&&n.shapeInfo.flatOffset==null&&x.util.arraysEqual(s,a))return` + float `+r+`() { + return sampleTexture(`+e+`, resultUV); + } + `;var u=Je(l),c=AS(n.shapeInfo.logicalShape,t.logicalShape),h=l-o,d,p=["x","y","z","w","u","v"];o===0?d="":l<2&&c.length>=1?d="coords = 0;":d=c.map(function(m){return"coords."+p[m+h]+" = 0;"}).join(` +`);var f="";return l<2&&o>0?f="coords":f=n.shapeInfo.logicalShape.map(function(m,g){return"coords."+p[g+h]}).join(", "),` + float `+r+`() { + `+u+` coords = getOutputCoords(); + `+d+` + return get`+i+"("+f+`); + } + `}function Je(n){if(n<=1)return"int";if(n===2)return"ivec2";if(n===3)return"ivec3";if(n===4)return"ivec4";if(n===5)return"ivec5";if(n===6)return"ivec6";throw Error("GPU for rank "+n+" is not yet supported")}function ga(n,t){var e=JSON.parse(JSON.stringify(n));return e.shapeInfo.logicalShape=t,e}function va(n,t){return t.map(function(e){return n[e]}).join(", ")}var f5=function(){function n(t,e,i,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,x.util.assert(t.length>2,function(){return"Packed arg"+(i.charAt(0).toUpperCase()+i.slice(1))+" supports only inputs with rank above 2."});var a=t[t.length-1],s=Math.ceil(a/e);this.outputShape=t.slice(0,-1),s>1&&this.outputShape.push(s),r||this.variableNames.push("bestIndicesA");var o=this.outputShape,l=o.length,u=Je(l),c=Jt("coords",l),h,d;if(s===1){d=l+1;var p=Je(d);h=` + `+p+" sourceLocR = "+p+"("+c.join()+`, 0); + ++`+c[l-1]+`; + `+p+" sourceLocG = "+p+"("+c.join()+`, 0); + ++`+c[l-2]+`; + `+p+" sourceLocA = "+p+"("+c.join()+`, 0); + --`+c[l-1]+`; + `+p+" sourceLocB = "+p+"("+c.join()+`, 0); + --`+c[l-2]+";"}else d=l,h=` + `+u+` sourceLocR = coords; + ++`+c[l-1]+`; + `+u+` sourceLocG = coords; + ++`+c[l-2]+`; + `+u+` sourceLocA = coords; + --`+c[l-1]+`; + `+u+` sourceLocB = coords; + --`+c[l-2]+";";var f=["x","y","z","w","u","v"].slice(0,d),m="."+f[d-1],g=f.map(function(D){return"int "+D}),v=Jt("sourceLocR",d-1).concat("inIdx.r"),b=Jt("sourceLocG",d-1).concat("inIdx.g"),w=Jt("sourceLocB",d-1).concat("inIdx.b"),S=Jt("sourceLocA",d-1).concat("inIdx.a"),L=i==="max"?"greaterThan":"lessThan",N=r?"":` + inIdx = round(vec4(getBestIndicesAChannel(`+v.join()+`), + getBestIndicesAChannel(`+b.join()+`), + getBestIndicesAChannel(`+w.join()+`), + getBestIndicesAChannel(`+S.join()+")));",C=`vec4( + getAChannel(`+v.join()+`), + hasNextCol ? getAChannel(`+b.join()+`) : 0., + hasNextRow ? getAChannel(`+w.join()+`) : 0., + hasNextRow && hasNextCol ? getAChannel(`+S.join()+") : 0.)",R=r?"":` + float getBestIndicesAChannel(`+g.join()+`) { + return getChannel(getBestIndicesA(`+f.join()+`), + vec2(`+f.slice(-2).join()+`)); + }`;this.userCode=` + float getAChannel(`+g.join()+`) { + return getChannel(getA(`+f.join()+`), + vec2(`+f.slice(-2).join()+`)); + } + `+R+` + void main() { + `+u+` coords = getOutputCoords(); + bool hasNextCol = `+c[l-1]+" < "+(o[l-1]-1)+`; + bool hasNextRow = `+c[l-2]+" < "+(o[l-2]-1)+`; + `+h+` + ivec4 srcIdx = ivec4(sourceLocR`+m+", sourceLocG"+m+`, + sourceLocB`+m+", sourceLocA"+m+") * "+e+`; + ivec4 inIdx = srcIdx; + vec4 bestIndex = vec4(inIdx); + vec4 bestValue = `+C+`; + + for (int i = 0; i < `+e+`; i++) { + inIdx = srcIdx; + `+N+` + vec4 candidate = `+C+`; + bvec4 nan = isnan(candidate); + bvec4 replace = bvec4( + vec4(`+L+`(candidate, bestValue)) * (vec4(1.0) - vec4(nan))); + + bestValue = vec4(replace.x ? candidate.x : bestValue.x, + replace.y ? candidate.y : bestValue.y, + replace.z ? candidate.z : bestValue.z, + replace.w ? candidate.w : bestValue.w); + bestIndex = mix(bestIndex, vec4(inIdx), vec4(replace)); + srcIdx++; + } + setOutput(bestIndex); + } + `}return n}();var m5=function(){function n(t){this.variableNames=["dy"],this.outputShape=t.inShape;var e=t.filterHeight,i=t.filterWidth,r=t.strideHeight,a=t.strideWidth,s=t.dilationHeight,o=t.dilationWidth,l=t.effectiveFilterHeight,u=t.effectiveFilterWidth,c=l-1-t.padInfo.top,h=u-1-t.padInfo.left,d=1/(e*i);this.userCode=` + const ivec2 pads = ivec2(`+c+", "+h+`); + const float avgMultiplier = float(`+d+`); + + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int d = coords[3]; + + ivec2 dyRCCorner = coords.yz - pads; + int dyRCorner = dyRCCorner.x; + int dyCCorner = dyRCCorner.y; + + // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d). + // ? = to be determined. : = across all values in that axis. + float dotProd = 0.0; + for (int wR = 0; wR < `+l+`; + wR += `+s+`) { + float dyR = float(dyRCorner + wR) / `+r+`.0; + + if (dyR < 0.0 || dyR >= `+t.outHeight+`.0 || fract(dyR) > 0.0) { + continue; + } + int idyR = int(dyR); + + for (int wC = 0; wC < `+u+`; + wC+= `+o+`) { + float dyC = float(dyCCorner + wC) / `+a+`.0; + + if (dyC < 0.0 || dyC >= `+t.outWidth+`.0 || + fract(dyC) > 0.0) { + continue; + } + int idyC = int(dyC); + + float dyValue = getDy(b, idyR, idyC, d); + + dotProd += dyValue * avgMultiplier; + } + } + setOutput(dotProd); + } + `}return n}(),g5=function(){function n(t){this.variableNames=["dy"],this.outputShape=t.inShape;var e=t.filterDepth,i=t.filterHeight,r=t.filterWidth,a=t.strideDepth,s=t.strideHeight,o=t.strideWidth,l=t.dilationDepth,u=t.dilationHeight,c=t.dilationWidth,h=t.effectiveFilterDepth,d=t.effectiveFilterHeight,p=t.effectiveFilterWidth,f=h-1-t.padInfo.front,m=d-1-t.padInfo.top,g=p-1-t.padInfo.left,v=1/(e*i*r);this.userCode=` + const ivec3 pads = ivec3(`+f+", "+m+", "+g+`); + const float avgMultiplier = float(`+v+`); + + void main() { + ivec5 coords = getOutputCoords(); + int batch = coords.x; + int ch = coords.u; + + ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads; + int dyDCorner = dyCorner.x; + int dyRCorner = dyCorner.y; + int dyCCorner = dyCorner.z; + + // Convolve dy(?, ?, ?, d) with pos mask(:, :, :, ch) to get + // dx(xD, xR, xC, ch). + // ? = to be determined. : = across all values in that axis. + float dotProd = 0.0; + + for (int wD = 0; wD < `+h+`; + wD += `+l+`) { + float dyD = float(dyDCorner + wD) / `+a+`.0; + + if (dyD < 0.0 || dyD >= `+t.outDepth+`.0 || fract(dyD) > 0.0) { + continue; + } + int idyD = int(dyD); + + for (int wR = 0; wR < `+d+`; + wR += `+u+`) { + float dyR = float(dyRCorner + wR) / `+s+`.0; + + if (dyR < 0.0 || dyR >= `+t.outHeight+`.0 || + fract(dyR) > 0.0) { + continue; + } + int idyR = int(dyR); + + for (int wC = 0; wC < `+p+`; + wC += `+c+`) { + float dyC = float(dyCCorner + wC) / `+o+`.0; + + if (dyC < 0.0 || dyC >= `+t.outWidth+`.0 || + fract(dyC) > 0.0) { + continue; + } + int idyC = int(dyC); + + float dyValue = getDy(batch, idyD, idyR, idyC, ch); + + dotProd += dyValue * avgMultiplier; + } + } + } + setOutput(dotProd); + } + `}return n}();var xS={REAL:"return areal * breal - aimag * bimag;",IMAG:"return areal * bimag + aimag * breal;"},CS=function(){function n(t,e,i){this.variableNames=["AReal","AImag","BReal","BImag"],this.outputShape=x.backend_util.assertAndGetBroadcastShape(e,i),this.userCode=` + float binaryOpComplex( + float areal, float aimag, float breal, float bimag) { + `+t+` + } + + void main() { + float areal = getARealAtOutCoords(); + float aimag = getAImagAtOutCoords(); + float breal = getBRealAtOutCoords(); + float bimag = getBImagAtOutCoords(); + setOutput(binaryOpComplex(areal, aimag, breal, bimag)); + } + `}return n}();var RS=` + if (isnan(a)) return a; + if (isnan(b)) return b; +`,cp="return a + b;",hp="return a - b;",OS="return a * b;",v5=` + float s = sign(a) * sign(b); + int ia = round(a); + int ib = round(b); + if (ib != 0) { + // Windows (D3D) wants guaranteed non-zero int division at compile-time. + return float(idiv(ia, ib, s)); + } else { + return NAN; + } +`,y5=` +if(a < 0.0 && floor(b) < b){ + return NAN; +} +if (b == 0.0) { + return 1.0; +} +return (round(mod(b, 2.0)) != 1) ? + pow(abs(a), b) : sign(a) * pow(abs(a), b); +`,b5="return float(a == b);",w5="return float(a != b);",S5="return float(a < b);",L5="return float(a <= b);",I5="return float(a > b);",A5="return float(a >= b);",T5="return float(a >= 1.0 && b >= 1.0);",N5="return float(a >= 1.0 || b >= 1.0);",x5=RS+` + return max(a, b); +`,C5=RS+` + return min(a, b); +`,R5=`if (b == 0.0) return NAN; + return mod(a, b);`,O5="return (b >= 1.0) ? a : a * (b + 1.0);",ES="return (a < 0.) ? b * a : a;",wt=function(){function n(t,e,i){this.variableNames=["A","B"],this.outputShape=x.backend_util.assertAndGetBroadcastShape(e,i),this.userCode=` + float binaryOperation(float a, float b) { + `+t+` + } + + void main() { + float a = getAAtOutCoords(); + float b = getBAtOutCoords(); + setOutput(binaryOperation(a, b)); + } + `}return n}();var al=` + result.r = isNaN.r > 0. ? NAN : result.r; + result.g = isNaN.g > 0. ? NAN : result.g; + result.b = isNaN.b > 0. ? NAN : result.b; + result.a = isNaN.a > 0. ? NAN : result.a; +`,E5=` + ivec4 ia = round(a); + ivec4 ib = round(b); + bvec4 cond = notEqual(ib, ivec4(0)); + ivec4 result = ivec4(0); + vec4 s = sign(a) * sign(b); + + // Windows (D3D) wants guaranteed non-zero int division at compile-time. + if (cond[0]) { + result[0] = idiv(ia[0], ib[0], s[0]); + } + if (cond[1]) { + result[1] = idiv(ia[1], ib[1], s[1]); + } + if (cond[2]) { + result[2] = idiv(ia[2], ib[2], s[2]); + } + if (cond[3]) { + result[3] = idiv(ia[3], ib[3], s[3]); + } + return vec4(result); +`,D5=` + // isModRound1 has 1 for components with round(mod(b, 2.0)) == 1, 0 otherwise. + vec4 isModRound1 = vec4(equal(round(mod(b, 2.0)), ivec4(1))); + vec4 multiplier = sign(a) * isModRound1 + (vec4(1.0) - isModRound1); + vec4 result = multiplier * pow(abs(a), b); + + // Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS + bvec4 isExpZero = equal(b, vec4(0.0)); + result.r = isExpZero.r ? 1.0 : result.r; + result.g = isExpZero.g ? 1.0 : result.g; + result.b = isExpZero.b ? 1.0 : result.b; + result.a = isExpZero.a ? 1.0 : result.a; + + vec4 isNaN = vec4(lessThan(a, vec4(0.0))) * vec4(lessThan(floor(b), b)); + `+al+` + return result; +`,DS=` + vec4 aLessThanZero = vec4(lessThan(a, vec4(0.))); + return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a); +`,k5=` + vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.))); + return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0)))); +`,F5=` + return vec4(equal(a, b)); +`,W5=` + return vec4(notEqual(a, b)); +`,U5=` + return vec4(lessThan(a, b)); +`,B5=` + return vec4(lessThanEqual(a, b)); +`,z5=` + return vec4(greaterThan(a, b)); +`,P5=` + return vec4(greaterThanEqual(a, b)); +`,_5=` + return vec4( + vec4(greaterThanEqual(a, vec4(1.0))) * + vec4(greaterThanEqual(b, vec4(1.0)))); +`,M5=` + return min( + vec4(greaterThanEqual(a, vec4(1.0))) + + vec4(greaterThanEqual(b, vec4(1.0))), + vec4(1.0)); +`,H5=` + vec4 result = vec4(max(a, b)); + vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0)); + `+al+` + return result; +`,V5=` + vec4 result = vec4(min(a, b)); + vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0)); + `+al+` + return result; +`,q5=` + vec4 result = mod(a, b); + vec4 isNaN = vec4(equal(b, vec4(0.0))); + `+al+` + return result; +`,ki=function(){function n(t,e,i,r){r===void 0&&(r=!1),this.variableNames=["A","B"],this.supportsBroadcasting=!0,this.packedInputs=!0,this.packedOutput=!0,this.outputShape=x.backend_util.assertAndGetBroadcastShape(e,i);var a=this.outputShape.length,s="";if(r)if(a===0||x.util.sizeFromShape(this.outputShape)===1)s=` + result.y = 0.; + result.z = 0.; + result.w = 0.; + `;else{var o=Je(a);if(s=` + `+o+` coords = getOutputCoords(); + `,a===1)s+=` + result.y = (coords + 1) >= `+this.outputShape[0]+` ? 0. : result.y; + result.z = 0.; + result.w = 0.; + `;else{var l=Jt("coords",a);s+=` + bool nextRowOutOfBounds = + (`+l[a-2]+" + 1) >= "+this.outputShape[a-2]+`; + bool nextColOutOfBounds = + (`+l[a-1]+" + 1) >= "+this.outputShape[a-1]+`; + result.y = nextColOutOfBounds ? 0. : result.y; + result.z = nextRowOutOfBounds ? 0. : result.z; + result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w; + `}}this.userCode=` + vec4 binaryOperation(vec4 a, vec4 b) { + `+t+` + } + + void main() { + vec4 a = getAAtOutCoords(); + vec4 b = getBAtOutCoords(); + + vec4 result = binaryOperation(a, b); + `+s+` + + setOutput(result); + } + `}return n}();var G5=function(){function n(t){this.variableNames=["A"],this.outputShape=t,this.userCode=` + uniform float minVal; + uniform float maxVal; + + void main() { + float value = getAAtOutCoords(); + if (isnan(value)) { + setOutput(value); + return; + } + + setOutput(clamp(value, minVal, maxVal)); + } + `}return n.prototype.getCustomSetupFunc=function(t,e){var i=this;return function(r,a){i.minLoc==null&&(i.minLoc=r.getUniformLocationNoThrow(a,"minVal"),i.maxLoc=r.getUniformLocationNoThrow(a,"maxVal")),r.gl.uniform1f(i.minLoc,t),r.gl.uniform1f(i.maxLoc,e)}},n}();var Y5=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.userCode=` + uniform float minVal; + uniform float maxVal; + + void main() { + vec4 value = getAAtOutCoords(); + + if (any(isnan(value))) { + setOutput(value); + return; + } + + setOutput(clamp(value, vec4(minVal), vec4(maxVal))); + } + `}return n.prototype.getCustomSetupFunc=function(t,e){var i=this;return function(r,a){i.minLoc==null&&(i.minLoc=r.getUniformLocationNoThrow(a,"minVal"),i.maxLoc=r.getUniformLocationNoThrow(a,"maxVal")),r.gl.uniform1f(i.minLoc,t),r.gl.uniform1f(i.maxLoc,e)}},n}();var K5=function(){function n(t){this.variableNames=["real","imag"],this.outputShape=t,this.userCode=` + void main() { + float re = abs(getRealAtOutCoords()); + float im = abs(getImagAtOutCoords()); + float mx = max(re, im); + + // sadly the length function in glsl is not underflow-safe + // (at least not on Intel GPUs). So the safe solution is + // to ensure underflow-safety in all cases. + setOutput( + mx == 0.0 ? 0.0 : mx * length(vec2(1, min(re, im)/mx)) + ); + } + `}return n}();var j5=function(){function n(t){this.outputShape=[],this.outputShape=x.backend_util.computeOutShape(t,1),this.variableNames=t.map(function(l,u){return"T"+u});var e=new Array(t.length-1);e[0]=t[0][1];for(var i=1;i= "+l[u-1]+`) { + return getChannel( + getT`+u+"("+sl(o,c,f)+`), + vec2(`+sl(h,c,f)+`)); + }`}var m=l.length,g=l[l.length-1];p+=` + return getChannel( + getT`+m+"("+sl(o,c,g)+`), + vec2(`+sl(h,c,g)+"));",this.userCode=` + float getValue(`+o.map(function(v){return"int "+v})+`) { + `+p+` + } + + void main() { + `+a+` coords = getOutputCoords(); + vec4 result = vec4(getValue(`+s+`), 0., 0., 0.); + + `+s[r-1]+" = "+s[r-1]+` + 1; + if (`+s[r-1]+" < "+i[r-1]+`) { + result.g = getValue(`+s+`); + } + + `+s[r-2]+" = "+s[r-2]+` + 1; + if (`+s[r-2]+" < "+i[r-2]+`) { + result.a = getValue(`+s+`); + } + + `+s[r-1]+" = "+s[r-1]+` - 1; + if (`+s[r-2]+" < "+i[r-2]+` && + `+s[r-1]+" < "+i[r-1]+`) { + result.b = getValue(`+s+`); + } + setOutput(result); + } + `}return n}();function sl(n,t,e){var i=n.indexOf(t),r=n.map(function(a,s){return s===i?a+" - "+e:a});return r.join()}var X5=function(){function n(t){this.variableNames=["x","dy"],this.outputShape=t.filterShape;var e=t.strideHeight,i=t.strideWidth,r=t.padInfo.top,a=t.padInfo.left,s=t.dataFormat==="channelsLast";this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int wR = coords.x; + int wC = coords.y; + int d1 = coords.z; + int d2 = coords.w; + + // Convolve x(?, ?, d1) with dy(:, :, d2) to get dw(wR, wC, d1, d2). + // ? = to be determined. : = across all values in that axis. + float dotProd = 0.0; + + for (int b = 0; b < `+t.batchSize+`; b++) { + for (int yR = 0; yR < `+t.outHeight+`; yR++) { + int xR = wR + yR * `+e+" - "+r+`; + + if (xR < 0 || xR >= `+t.inHeight+`) { + continue; + } + + for (int yC = 0; yC < `+t.outWidth+`; yC++) { + int xC = wC + yC * `+i+" - "+a+`; + + if (xC < 0 || xC >= `+t.inWidth+`) { + continue; + } + + if (`+s+`) { + float dyValue = getDy(b, yR, yC, d2); + float xValue = getX(b, xR, xC, d1); + dotProd += (xValue * dyValue); + } else { + float dyValue = getDy(b, d2, yR, yC); + float xValue = getX(b, d1, xR, xC); + dotProd += (xValue * dyValue); + } + + } + } + } + setOutput(dotProd); + } + `}return n}(),J5=function(){function n(t){this.variableNames=["dy","W"],this.outputShape=t.inShape;var e=t.filterHeight,i=t.filterWidth,r=t.strideHeight,a=t.strideWidth,s=t.dataFormat==="channelsLast",o=e-1-t.padInfo.top,l=i-1-t.padInfo.left,u=s?1:2,c=s?2:3,h=s?3:1;this.userCode=` + const ivec2 pads = ivec2(`+o+", "+l+`); + + void main() { + ivec4 coords = getOutputCoords(); + int batch = coords[0]; + int d1 = coords[`+h+`]; + + ivec2 dyCorner = ivec2(coords[`+u+"], coords["+c+`]) - pads; + int dyRCorner = dyCorner.x; + int dyCCorner = dyCorner.y; + + // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1). + // ? = to be determined. : = across all values in that axis. + float dotProd = 0.0; + for (int wR = 0; wR < `+e+`; wR++) { + float dyR = float(dyRCorner + wR) / `+r+`.0; + + if (dyR < 0.0 || dyR >= `+t.outHeight+`.0 || fract(dyR) > 0.0) { + continue; + } + int idyR = int(dyR); + + int wRPerm = `+e+` - 1 - wR; + + for (int wC = 0; wC < `+i+`; wC++) { + float dyC = float(dyCCorner + wC) / `+a+`.0; + + if (dyC < 0.0 || dyC >= `+t.outWidth+`.0 || + fract(dyC) > 0.0) { + continue; + } + int idyC = int(dyC); + + int wCPerm = `+i+` - 1 - wC; + + for (int d2 = 0; d2 < `+t.outChannels+`; d2++) { + + if (`+s+`) { + float xValue = getDy(batch, idyR, idyC, d2); + float wValue = getW(wRPerm, wCPerm, d1, d2); + dotProd += xValue * wValue; + } else { + float xValue = getDy(batch, d2, idyR, idyC); + float wValue = getW(wRPerm, wCPerm, d1, d2); + dotProd += xValue * wValue; + } + + } + } + } + setOutput(dotProd); + } + `}return n}(),Z5=function(){function n(t){this.variableNames=["x","dy"],this.outputShape=t.filterShape;var e=t.strideDepth,i=t.strideHeight,r=t.strideWidth,a=t.padInfo.front,s=t.padInfo.top,o=t.padInfo.left;this.userCode=` + void main() { + ivec5 coords = getOutputCoords(); + int wF = coords.x; + int wR = coords.y; + int wC = coords.z; + int d1 = coords.w; + int d2 = coords.u; + + float dotProd = 0.0; + + for (int b = 0; b < `+t.batchSize+`; b++) { + for (int yF = 0; yF < `+t.outDepth+`; yF++) { + int xF = wF + yF * `+e+" - "+a+`; + + if (xF < 0 || xF >= `+t.inDepth+`) { + continue; + } + + for (int yR = 0; yR < `+t.outHeight+`; yR++) { + int xR = wR + yR * `+i+" - "+s+`; + + if (xR < 0 || xR >= `+t.inHeight+`) { + continue; + } + + for (int yC = 0; yC < `+t.outWidth+`; yC++) { + int xC = wC + yC * `+r+" - "+o+`; + + if (xC < 0 || xC >= `+t.inWidth+`) { + continue; + } + + float dyValue = getDy(b, yF, yR, yC, d2); + float xValue = getX(b, xF, xR, xC, d1); + dotProd += (xValue * dyValue); + } + } + } + } + setOutput(dotProd); + } + `}return n}(),Q5=function(){function n(t){this.variableNames=["dy","W"],this.outputShape=t.inShape;var e=t.filterDepth,i=t.filterHeight,r=t.filterWidth,a=t.strideDepth,s=t.strideHeight,o=t.strideWidth,l=e-1-t.padInfo.front,u=i-1-t.padInfo.top,c=r-1-t.padInfo.left;this.userCode=` + const ivec3 pads = ivec3(`+l+", "+u+", "+c+`); + + void main() { + ivec5 coords = getOutputCoords(); + int batch = coords.x; + int d1 = coords.u; + + + ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads; + int dyFCorner = dyCorner.x; + int dyRCorner = dyCorner.y; + int dyCCorner = dyCorner.z; + + float dotProd = 0.0; + for (int wF = 0; wF < `+e+`; wF++) { + float dyF = float(dyFCorner + wF) / `+a+`.0; + + if (dyF < 0.0 || dyF >= `+t.outDepth+`.0 || fract(dyF) > 0.0) { + continue; + } + int idyF = int(dyF); + + int wFPerm = `+e+` - 1 - wF; + + for (int wR = 0; wR < `+i+`; wR++) { + float dyR = float(dyRCorner + wR) / `+s+`.0; + + if (dyR < 0.0 || dyR >= `+t.outHeight+`.0 || + fract(dyR) > 0.0) { + continue; + } + int idyR = int(dyR); + + int wRPerm = `+i+` - 1 - wR; + + for (int wC = 0; wC < `+r+`; wC++) { + float dyC = float(dyCCorner + wC) / `+o+`.0; + + if (dyC < 0.0 || dyC >= `+t.outWidth+`.0 || + fract(dyC) > 0.0) { + continue; + } + int idyC = int(dyC); + + int wCPerm = `+r+` - 1 - wC; + + for (int d2 = 0; d2 < `+t.outChannels+`; d2++) { + float xValue = getDy(batch, idyF, idyR, idyC, d2); + float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2); + dotProd += xValue * wValue; + } + } + } + } + setOutput(dotProd); + } + `}return n}();var e9=function(){function n(t){this.variableNames=["x","dy"],this.outputShape=t.filterShape;var e=t.strideHeight,i=t.strideWidth,r=t.padInfo.top,a=t.padInfo.left,s=t.outChannels/t.inChannels;this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int wR = coords.x; + int wC = coords.y; + int d1 = coords.z; + int dm = coords.w; + int d2 = d1 * `+s+` + dm; + + float dotProd = 0.0; + + // TO DO: Vec4 over the batch size + for (int b = 0; b < `+t.batchSize+`; b++) { + for (int yR = 0; yR < `+t.outHeight+`; yR++) { + int xR = wR + yR * `+e+" - "+r+`; + + if (xR < 0 || xR >= `+t.inHeight+`) { + continue; + } + + for (int yC = 0; yC < `+t.outWidth+`; yC++) { + int xC = wC + yC * `+i+" - "+a+`; + + if (xC < 0 || xC >= `+t.inWidth+`) { + continue; + } + + float dyValue = getDy(b, yR, yC, d2); + float xValue = getX(b, xR, xC, d1); + dotProd += (xValue * dyValue); + } + } + } + setOutput(dotProd); + } + `}return n}(),t9=function(){function n(t){this.variableNames=["dy","W"],this.outputShape=t.inShape;var e=t.filterHeight,i=t.filterWidth,r=t.strideHeight,a=t.strideWidth,s=e-1-t.padInfo.top,o=i-1-t.padInfo.left,l=t.outChannels/t.inChannels;this.userCode=` + const ivec2 pads = ivec2(`+s+", "+o+`); + + void main() { + ivec4 coords = getOutputCoords(); + int batch = coords[0]; + int d1 = coords[3]; + ivec2 dyCorner = coords.yz - pads; + int dyRCorner = dyCorner.x; + int dyCCorner = dyCorner.y; + + float dotProd = 0.0; + + for (int wR = 0; wR < `+e+`; wR++) { + float dyR = float(dyRCorner + wR) / `+r+`.0; + + if (dyR < 0.0 || dyR >= `+t.outHeight+`.0 || fract(dyR) > 0.0) { + continue; + } + int idyR = int(dyR); + + int wRPerm = `+e+` - 1 - wR; + + for (int wC = 0; wC < `+i+`; wC++) { + float dyC = float(dyCCorner + wC) / `+a+`.0; + + if (dyC < 0.0 || dyC >= `+t.outWidth+`.0 || + fract(dyC) > 0.0) { + continue; + } + int idyC = int(dyC); + + int wCPerm = `+i+` - 1 - wC; + + // TO DO: Vec4 over the channelMul + for (int dm = 0; dm < `+l+`; dm++) { + int d2 = d1 * `+l+` + dm; + float xValue = getDy(batch, idyR, idyC, d2); + float wValue = getW(wRPerm, wCPerm, d1, dm); + dotProd += xValue * wValue; + } + } + } + setOutput(dotProd); + } + `}return n}();var kS=function(){function n(t,e,i,r){e===void 0&&(e=!1),i===void 0&&(i=null),r===void 0&&(r=!1),this.variableNames=["x","W"],this.outputShape=t.outShape;var a=t.padInfo.top,s=t.padInfo.left,o=t.strideHeight,l=t.strideWidth,u=t.dilationHeight,c=t.dilationWidth,h=t.filterHeight,d=t.filterWidth,p=Math.floor(t.inChannels/4)*4,f=t.inChannels%4,m=t.dataFormat==="channelsLast",g=m?1:2,v=m?2:3,b=m?3:1,w="",S="";i&&(r?w=`float activation(float a) { + float b = getPreluActivationWeightsAtOutCoords(); + `+i+` + }`:w=` + float activation(float x) { + `+i+` + } + `,S="result = activation(result);");var L=e?"result += getBiasAtOutCoords();":"";e&&this.variableNames.push("bias"),r&&this.variableNames.push("preluActivationWeights"),this.userCode=` + `+w+` + + const ivec2 strides = ivec2(`+o+", "+l+`); + const ivec2 pads = ivec2(`+a+", "+s+`); + + void main() { + ivec4 coords = getOutputCoords(); + int batch = coords[0]; + int d2 = coords[`+b+`]; + + ivec2 xRCCorner = + ivec2(coords[`+g+"], coords["+v+`]) * strides - pads; + int xRCorner = xRCCorner.x; + int xCCorner = xRCCorner.y; + + // Convolve x(?, ?, d1) with w(:, :, d1, d2) to get y(yR, yC, d2). + // ? = to be determined. : = across all values in that axis. + float dotProd = 0.0; + for (int wR = 0; wR < `+h+`; wR++) { + int xR = xRCorner + wR * `+u+`; + + if (xR < 0 || xR >= `+t.inHeight+`) { + continue; + } + + for (int wC = 0; wC < `+d+`; wC++) { + int xC = xCCorner + wC * `+c+`; + + if (xC < 0 || xC >= `+t.inWidth+`) { + continue; + } + + for (int d1 = 0; d1 < `+p+`; d1 += 4) { + vec4 wValues = vec4( + getW(wR, wC, d1, d2), + getW(wR, wC, d1 + 1, d2), + getW(wR, wC, d1 + 2, d2), + getW(wR, wC, d1 + 3, d2) + ); + + if (`+m+`) { + vec4 xValues = vec4( + getX(batch, xR, xC, d1), + getX(batch, xR, xC, d1 + 1), + getX(batch, xR, xC, d1 + 2), + getX(batch, xR, xC, d1 + 3) + ); + dotProd += dot(xValues, wValues); + } else { + vec4 xValues = vec4( + getX(batch, d1, xR, xC), + getX(batch, d1 + 1, xR, xC), + getX(batch, d1 + 2, xR, xC), + getX(batch, d1 + 3, xR, xC) + ); + dotProd += dot(xValues, wValues); + } + } + + if (`+(f===1)+`) { + + if (`+m+`) { + dotProd += + getX(batch, xR, xC, `+p+`) * + getW(wR, wC, `+p+`, d2); + } else { + dotProd += + getX(batch, `+p+`, xR, xC) * + getW(wR, wC, `+p+`, d2); + } + + } else if (`+(f===2)+`) { + vec2 wValues = vec2( + getW(wR, wC, `+p+`, d2), + getW(wR, wC, `+p+` + 1, d2) + ); + + if (`+m+`) { + vec2 xValues = vec2( + getX(batch, xR, xC, `+p+`), + getX(batch, xR, xC, `+p+` + 1) + ); + dotProd += dot(xValues, wValues); + } else { + vec2 xValues = vec2( + getX(batch, `+p+`, xR, xC), + getX(batch, `+p+` + 1, xR, xC) + ); + dotProd += dot(xValues, wValues); + } + + } else if (`+(f===3)+`) { + vec3 wValues = vec3( + getW(wR, wC, `+p+`, d2), + getW(wR, wC, `+p+` + 1, d2), + getW(wR, wC, `+p+` + 2, d2) + ); + + if (`+m+`) { + vec3 xValues = vec3( + getX(batch, xR, xC, `+p+`), + getX(batch, xR, xC, `+p+` + 1), + getX(batch, xR, xC, `+p+` + 2) + ); + dotProd += dot(xValues, wValues); + } else { + vec3 xValues = vec3( + getX(batch, `+p+`, xR, xC), + getX(batch, `+p+` + 1, xR, xC), + getX(batch, `+p+` + 2, xR, xC) + ); + dotProd += dot(xValues, wValues); + } + + } + } + } + + float result = dotProd; + `+L+` + `+S+` + setOutput(result); + } + `}return n}(),n9=function(){function n(t){this.variableNames=["x","W"],this.outputShape=t.outShape;var e=t.padInfo.front,i=t.padInfo.top,r=t.padInfo.left,a=t.strideDepth,s=t.strideHeight,o=t.strideWidth,l=t.dilationDepth,u=t.dilationHeight,c=t.dilationWidth,h=t.filterDepth,d=t.filterHeight,p=t.filterWidth,f=Math.floor(t.inChannels/4)*4,m=t.inChannels%4;this.userCode=` + const ivec3 strides = ivec3(`+a+", "+s+", "+o+`); + const ivec3 pads = ivec3(`+e+", "+i+", "+r+`); + + void main() { + ivec5 coords = getOutputCoords(); + int batch = coords.x; + int d2 = coords.u; + + ivec3 xFRCCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads; + int xFCorner = xFRCCorner.x; + int xRCorner = xFRCCorner.y; + int xCCorner = xFRCCorner.z; + + // Convolve x(?, ?, ?, d1) with w(:, :, :, d1, d2) to get + // y(yF, yR, yC, d2). ? = to be determined. : = across all + // values in that axis. + float dotProd = 0.0; + for (int wF = 0; wF < `+h+`; wF++) { + int xF = xFCorner + wF * `+l+`; + + if (xF < 0 || xF >= `+t.inDepth+`) { + continue; + } + + for (int wR = 0; wR < `+d+`; wR++) { + int xR = xRCorner + wR * `+u+`; + + if (xR < 0 || xR >= `+t.inHeight+`) { + continue; + } + + for (int wC = 0; wC < `+p+`; wC++) { + int xC = xCCorner + wC * `+c+`; + + if (xC < 0 || xC >= `+t.inWidth+`) { + continue; + } + + for (int d1 = 0; d1 < `+f+`; d1 += 4) { + vec4 xValues = vec4( + getX(batch, xF, xR, xC, d1), + getX(batch, xF, xR, xC, d1 + 1), + getX(batch, xF, xR, xC, d1 + 2), + getX(batch, xF, xR, xC, d1 + 3) + ); + vec4 wValues = vec4( + getW(wF, wR, wC, d1, d2), + getW(wF, wR, wC, d1 + 1, d2), + getW(wF, wR, wC, d1 + 2, d2), + getW(wF, wR, wC, d1 + 3, d2) + ); + + dotProd += dot(xValues, wValues); + } + + if (`+(m===1)+`) { + dotProd += + getX(batch, xF, xR, xC, `+f+`) * + getW(wF, wR, wC, `+f+`, d2); + } else if (`+(m===2)+`) { + vec2 xValues = vec2( + getX(batch, xF, xR, xC, `+f+`), + getX(batch, xF, xR, xC, `+f+` + 1) + ); + vec2 wValues = vec2( + getW(wF, wR, wC, `+f+`, d2), + getW(wF, wR, wC, `+f+` + 1, d2) + ); + dotProd += dot(xValues, wValues); + } else if (`+(m===3)+`) { + vec3 xValues = vec3( + getX(batch, xF, xR, xC, `+f+`), + getX(batch, xF, xR, xC, `+f+` + 1), + getX(batch, xF, xR, xC, `+f+` + 2) + ); + vec3 wValues = vec3( + getW(wF, wR, wC, `+f+`, d2), + getW(wF, wR, wC, `+f+` + 1, d2), + getW(wF, wR, wC, `+f+` + 2, d2) + ); + dotProd += dot(xValues, wValues); + } + } + } + } + setOutput(dotProd); + } + `}return n}();var FS=function(){function n(t,e,i,r){e===void 0&&(e=!1),i===void 0&&(i=null),r===void 0&&(r=!1),this.variableNames=["x","W"],this.outputShape=t.outShape;var a=t.inHeight,s=t.inWidth,o=t.padInfo.top,l=t.padInfo.left,u=t.strideHeight,c=t.strideWidth,h=t.dilationHeight,d=t.dilationWidth,p=t.filterHeight,f=t.filterWidth,m=t.outChannels/t.inChannels,g="",v="";i&&(r?g=`float activation(float a) { + float b = getPreluActivationWeightsAtOutCoords(); + `+i+` + }`:g=` + float activation(float x) { + `+i+` + } + `,v="result = activation(result);");var b=e?"result += getBiasAtOutCoords();":"";e&&this.variableNames.push("bias"),r&&this.variableNames.push("preluActivationWeights"),this.userCode=` + `+g+` + + const ivec2 strides = ivec2(`+u+", "+c+`); + const ivec2 pads = ivec2(`+o+", "+l+`); + + void main() { + ivec4 coords = getOutputCoords(); + int batch = coords.x; + ivec2 xRCCorner = coords.yz * strides - pads; + int d2 = coords.w; + int d1 = d2 / `+m+`; + int q = d2 - d1 * `+m+`; + + int xRCorner = xRCCorner.x; + int xCCorner = xRCCorner.y; + + // Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2). + // ? = to be determined. : = across all values in that axis. + float dotProd = 0.0; + // TO DO(dsmilkov): Flatten the two for loops and vec4 the operations. + for (int wR = 0; wR < `+p+`; wR++) { + int xR = xRCorner + wR * `+h+`; + + if (xR < 0 || xR >= `+a+`) { + continue; + } + + for (int wC = 0; wC < `+f+`; wC++) { + int xC = xCCorner + wC * `+d+`; + + if (xC < 0 || xC >= `+s+`) { + continue; + } + + float xVal = getX(batch, xR, xC, d1); + float wVal = getW(wR, wC, d1, q); + dotProd += xVal * wVal; + } + } + + float result = dotProd; + `+b+` + `+v+` + setOutput(result); + } + `}return n}();var WS=function(){function n(t,e,i,r){e===void 0&&(e=!1),i===void 0&&(i=null),r===void 0&&(r=!1),this.variableNames=["x","W"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t.outShape;for(var a=t.inHeight,s=t.inWidth,o=t.padInfo.top,l=t.padInfo.left,u=t.strideHeight,c=t.strideWidth,h=t.dilationHeight,d=t.dilationWidth,p=t.filterHeight,f=t.filterWidth,m=f,g="int xR; int xC; int xCOffset;",v=0;v= 0 && xR < `+a+" && xCOffset >= 0 && xCOffset < "+s+`) { + xTexelR`+v+"C"+b+` = getX(batch, xR, xCOffset, d1); + + // Need to manually clear unused channels in case + // we're reading from recycled texture. + if(xCOffset + 1 >= `+s+`) { + xTexelR`+v+"C"+b+`.zw = vec2(0.); + } + } else { + xTexelR`+v+"C"+b+` = vec4(0.); + } + + xCOffset = xC + 1 - 2; + if(xR >= 0 && xR < `+a+" && xCOffset >= 0 && xCOffset < "+s+`) { + vec4 previous = getX(batch, xR, xCOffset, d1); + + // Need to manually clear unused channels in case + // we're reading from recycled texture. + if(xCOffset + 1 >= `+s+`) { + previous.zw = vec2(0.); + } + + xR`+v+"C"+b+" = vec4(previous.zw, xTexelR"+v+"C"+b+`.xy); + } else { + xR`+v+"C"+b+" = vec4(0, 0, xTexelR"+v+"C"+b+`.xy); + } + `:g+=` + if(xR >= 0 && xR < `+a+" && xC >= 0 && xC < "+s+`) { + xTexelR`+v+"C"+b+` = getX(batch, xR, xC, d1); + } else { + xTexelR`+v+"C"+b+` = vec4(0.); + } + + xR`+v+"C"+b+" = xTexelR"+v+"C"+b+`; + `,b+1= 0 && xR < `+a+` && + xCOffset >= 0 && xCOffset < `+s+`) { + xTexelR`+v+"C"+(b+2)+` = getX(batch, xR, xCOffset, d1); + } + `,d>1&&(g+=` + xCOffset -= 2; + if(xR >= 0 && xR < `+a+` && + xCOffset >= 0 && xCOffset < `+s+`) { + xTexelR`+v+"C"+b+` = getX(batch, xR, xCOffset, d1); + } else { + xTexelR`+v+"C"+b+` = vec4(0.); + } + `),g+=` + xR`+v+"C"+(b+1)+` = vec4( + xTexelR`+v+"C"+b+".zw, xTexelR"+v+"C"+(b+2)+`.xy); + `):g+=` + xCOffset = xC + `+S+`; + + if(xR >= 0 && xR < `+a+` && + xCOffset >= 0 && xCOffset < `+s+`) { + xTexelR`+v+"C"+(b+2)+` = getX(batch, xR, xCOffset, d1); + } + + xR`+v+"C"+(b+1)+" = xTexelR"+v+"C"+(b+2)+`; + `}}else b= 0 && xR < `+a+`) { + `,l%2===1?(g+=` + xCOffset = xC + 1 - `+c+`; + if(xCOffset >= 0 && xCOffset < `+s+`) { + xTexelR`+v+"C"+b+` = getX(batch, xR, xCOffset, d1); + } else { + xTexelR`+v+"C"+b+` = vec4(0.); + } + + if(xC + 1 >= 0 && xC + 1 < `+s+`) { + xTexelR`+v+"C"+(b+2)+` = getX(batch, xR, xC + 1, d1); + } else { + xTexelR`+v+"C"+(b+2)+` = vec4(0.); + } + + xR`+v+"C"+b+` = vec4( + xTexelR`+v+"C"+b+".zw, xTexelR"+v+"C"+(b+2)+`.zw); + `,b+1= 0 && xCOffset < `+s+`) { + final = getX(batch, xR, xCOffset, d1); + } + xR`+v+"C"+(b+1)+" = vec4(xTexelR"+v+"C"+(b+2)+`.xy, final.xy); + `)):(g+=` + if(xC >= 0 && xC < `+s+`) { + xTexelR`+v+"C"+b+` = getX(batch, xR, xC, d1); + } else { + xTexelR`+v+"C"+b+` = vec4(0.); + } + + xCOffset = xC + `+c+`; + if(xCOffset >= 0 && xCOffset < `+s+`) { + xTexelR`+v+"C"+(b+2)+` = getX(batch, xR, xCOffset, d1); + } else { + xTexelR`+v+"C"+(b+2)+` = vec4(0.); + } + + xR`+v+"C"+b+` = vec4( + xTexelR`+v+"C"+b+".xy, xTexelR"+v+"C"+(b+2)+`.xy); + `,b+11?[""+(o-1)/(h-1),"(y2-y1) * height_ratio","y1*"+m+" + float(y)*(height_scale)"]:["0.0","0.0","0.5 * (y1+y2) * "+m],b=v[0],w=v[1],S=v[2],L=d>1?[""+(l-1)/(d-1),"(x2-x1) * width_ratio","x1*"+g+" + float(x)*(width_scale)"]:["0.0","0.0","0.5 * (x1+x2) * "+g],N=L[0],C=L[1],R=L[2];this.userCode=` + const float height_ratio = float(`+b+`); + const float width_ratio = float(`+N+`); + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int y = coords[1]; + int x = coords[2]; + int d = coords[3]; + + // get box vals + float y1 = getBoxes(b,0); + float x1 = getBoxes(b,1); + float y2 = getBoxes(b,2); + float x2 = getBoxes(b,3); + + // get image in batch index + int bInd = round(getBoxInd(b)); + if(bInd < 0 || bInd >= `+s+`) { + return; + } + + float height_scale = `+w+`; + float width_scale = `+C+`; + + float in_y = `+S+`; + if( in_y < 0.0 || in_y > `+m+` ) { + setOutput(float(`+a+`)); + return; + } + float in_x = `+R+`; + if( in_x < 0.0 || in_x > `+g+` ) { + setOutput(float(`+a+`)); + return; + } + + vec2 sourceFracIndexCR = vec2(in_x,in_y); + if(`+p+` == 1) { + // Compute the four integer indices. + ivec2 sourceFloorCR = ivec2(sourceFracIndexCR); + ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR)); + + float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d); + float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d); + float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d); + float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d); + + vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR); + + float top = topLeft + (topRight - topLeft) * fracCR.x; + float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x; + float newValue = top + (bottom - top) * fracCR.y; + setOutput(newValue); + } else { + // Compute the coordinators of nearest neighbor point. + ivec2 sourceNearestCR = ivec2(floor( + sourceFracIndexCR + vec2(0.5,0.5))); + float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d); + setOutput(newValue); + } + } + `}return n}(),zS=function(){function n(t,e,i){this.variableNames=["x"],this.outputShape=t;var r=t.length,a=e?"0.0":"getX("+US(r,"coords")+")",s=t[t.length-1],o="",l="";e?(o=i?"end != "+(s-1):"end != 0",l=i?"end + 1":"end - 1"):(o=i?"end + pow2 < "+s:"end >= pow2",l=i?"end + pow2":"end - pow2"),this.userCode=` + uniform float index; + void main() { + `+Je(r)+` coords = getOutputCoords(); + int end = `+BS(r,"coords")+`; + float val = `+a+`; + int pow2 = int(pow(2.0, index)); + if (`+o+`) { + int idx = `+l+`; + `+BS(r,"coords")+` = idx; + val += getX(`+US(r,"coords")+`); + } + setOutput(val); + } + `}return n.prototype.getCustomSetupFunc=function(t){var e=this;return function(i,r){e.index==null&&(e.index=i.getUniformLocation(r,"index")),i.gl.uniform1f(e.index,t)}},n}();function US(n,t){if(n===1)return""+t;if(n===2)return t+".x, "+t+".y";if(n===3)return t+".x, "+t+".y, "+t+".z";if(n===4)return t+".x, "+t+".y, "+t+".z, "+t+".w";throw Error("Cumulative sum for rank "+n+" is not yet supported")}function BS(n,t){if(n===1)return""+t;if(n===2)return t+".y";if(n===3)return t+".z";if(n===4)return t+".w";throw Error("Cumulative sum for rank "+n+" is not yet supported")}var r9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.outPackingScheme=ms.DENSE;var e=vs(t),i=Wt();this.outputShape=t,this.userCode=` + ivec3 outCoordsFromFlatIndex(int index) { + `+wr(["r","c","d"],t)+` + return ivec3(r, c, d); + } + + void main() { + ivec2 resTexRC = ivec2(resultUV.yx * + vec2(`+e[0]+", "+e[1]+`)); + int index = 4 * (resTexRC.x * `+e[1]+` + resTexRC.y); + + vec4 result = vec4(0.); + + for (int i=0; i<4; i++) { + int flatIndex = index + i; + ivec3 rc = outCoordsFromFlatIndex(flatIndex); + result[i] = getA(rc.x, rc.y, rc.z); + } + + `+i.output+` = result; + } + `}return n}();var a9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outPackingScheme=ms.DENSE;var e=vs(t),i=Wt();this.outputShape=t,this.userCode=` + ivec3 outCoordsFromFlatIndex(int index) { + `+wr(["r","c","d"],t)+` + return ivec3(r, c, d); + } + + void main() { + ivec2 resTexRC = ivec2(resultUV.yx * + vec2(`+e[0]+", "+e[1]+`)); + int index = 4 * (resTexRC.x * `+e[1]+` + resTexRC.y); + + vec4 result = vec4(0.); + + for (int i=0; i<4; i++) { + int flatIndex = index + i; + ivec3 rc = outCoordsFromFlatIndex(flatIndex); + result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z)); + } + + `+i.output+` = result; + } + `}return n}();var s9=function(){function n(t,e,i){this.variableNames=["x"],this.outputShape=[],this.outputShape=t,this.blockSize=e,this.dataFormat=i,this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int h = `+this.getHeightCoordString()+`; + int w = `+this.getWidthCoordString()+`; + int d = `+this.getDepthCoordString()+`; + + int in_h = h / `+e+`; + int offset_h = imod(h, `+e+`); + int in_w = w / `+e+`; + int offset_w = imod(w, `+e+`); + int offset_d = (offset_h * `+e+` + offset_w) * + `+this.getOutputDepthSize()+`; + int in_d = d + offset_d; + + float result = `+this.getInputSamplingString()+`; + setOutput(result); + } + `}return n.prototype.getHeightCoordString=function(){return this.dataFormat==="NHWC"?"coords[1]":"coords[2]"},n.prototype.getWidthCoordString=function(){return this.dataFormat==="NHWC"?"coords[2]":"coords[3]"},n.prototype.getDepthCoordString=function(){return this.dataFormat==="NHWC"?"coords[3]":"coords[1]"},n.prototype.getOutputDepthSize=function(){return this.dataFormat==="NHWC"?this.outputShape[3]:this.outputShape[1]},n.prototype.getInputSamplingString=function(){return this.dataFormat==="NHWC"?"getX(b, in_h, in_w, in_d)":"getX(b, in_d, in_h, in_w)"},n}();var o9=function(){function n(t){this.variableNames=["X"],this.outputShape=[t,t],this.userCode=` + void main() { + ivec2 coords = getOutputCoords(); + float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0; + setOutput(val); + } + `}return n}();var l9=function(){function n(t){this.variableNames=["A"],this.outTexUsage=an.DOWNLOAD;var e=Wt();this.outputShape=t,this.userCode=` + `+IS+` + + void main() { + float x = getAAtOutCoords(); + `+e.output+` = encode_float(x); + } + `}return n}();var u9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=an.DOWNLOAD;var e=Wt();this.outputShape=t,this.userCode=` + `+IS+` + + void main() { + ivec3 coords = getOutputCoords(); + float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z)); + `+e.output+` = encode_float(x); + } + `}return n}();var c9=function(){function n(t,e,i){i===void 0&&(i=!1),this.variableNames=["A"];var r=Wt(),a=e[0],s=e[1];this.outputShape=t;var o="result";i&&(o="floor(result * 255. + 0.5)"),this.userCode=` + `+up(t)+` + + void main() { + ivec3 coords = getOutputCoords(); + + int flatIndex = getFlatIndex(coords); + int offset = imod(flatIndex, 4); + + flatIndex = idiv(flatIndex, 4, 1.); + + int r = flatIndex / `+s+`; + int c = imod(flatIndex, `+s+`); + vec2 uv = (vec2(c, r) + halfCR) / vec2(`+s+".0, "+a+`.0); + vec4 values = `+r.texture2D+`(A, uv); + + float result; + + if(offset == 0) { + result = values[0]; + } else if(offset == 1) { + result = values[1]; + } else if(offset == 2) { + result = values[2]; + } else { + result = values[3]; + } + + `+r.output+" = vec4("+o+`, 0., 0., 0.); + } + `}return n}();var h9=function(){function n(t,e,i){i===void 0&&(i=!1),this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0;var r=Wt(),a=e[0],s=e[1];this.outputShape=t;var o="",l="result";i&&(l="floor(result * 255. + 0.5)");for(var u=0;u<=1;u++)for(var c=0;c<=1;c++){var h=u*2+c;o+=` + localCoords = coords; + if(localCoords[2] + `+c+" < "+t[2]+`) { + localCoords[2] += `+c+`; + if(localCoords[1] + `+u+" < "+t[1]+`) { + localCoords[1] += `+u+`; + + flatIndex = getFlatIndex(localCoords); + offset = imod(flatIndex, 4); + + flatIndex = idiv(flatIndex, 4, 1.); + + r = flatIndex / `+s+`; + c = imod(flatIndex, `+s+`); + uv = (vec2(c, r) + halfCR) / vec2(`+s+".0, "+a+`.0); + values = `+r.texture2D+`(A, uv); + + if(offset == 0) { + result[`+h+`] = values[0]; + } else if(offset == 1) { + result[`+h+`] = values[1]; + } else if(offset == 2) { + result[`+h+`] = values[2]; + } else { + result[`+h+`] = values[3]; + } + } + } + `}this.userCode=` + `+up(t)+` + + void main() { + ivec3 coords = getOutputCoords(); + + vec4 result = vec4(0.); + int flatIndex, r, c, offset; + ivec3 localCoords; + vec2 uv; + vec4 values; + + `+o+` + + `+r.output+" = "+l+`; + } + `}return n}();var PS={REAL:"return real * expR - imag * expI;",IMAG:"return real * expI + imag * expR;"},_S=function(){function n(t,e,i){this.variableNames=["real","imag"];var r=e[1];this.outputShape=e;var a=i?"2.0 * "+Math.PI:"-2.0 * "+Math.PI,s=i?r+".0":"1.0";this.userCode=` + const float exponentMultiplier = `+a+`; + + float unaryOpComplex(float real, float expR, float imag, float expI) { + `+t+` + } + + float mulMatDFT(int batch, int index) { + float indexRatio = float(index) / float(`+r+`); + float exponentMultiplierTimesIndexRatio = + exponentMultiplier * indexRatio; + + float result = 0.0; + + for (int i = 0; i < `+r+`; i++) { + // x = (-2|2 * PI / N) * index * i; + float x = exponentMultiplierTimesIndexRatio * float(i); + float expR = cos(x); + float expI = sin(x); + float real = getReal(batch, i); + float imag = getImag(batch, i); + + result += + unaryOpComplex(real, expR, imag, expI) / `+s+`; + } + + return result; + } + + void main() { + ivec2 coords = getOutputCoords(); + setOutput(mulMatDFT(coords[0], coords[1])); + } + `}return n}();var d9=function(){function n(t,e){this.outputShape=[],this.variableNames=["x"],this.outputShape=t,this.userCode=` + uniform float value; + void main() { + // Input can be obtained from uniform value. + setOutput(value); + } + `}return n.prototype.getCustomSetupFunc=function(t){var e=this;return function(i,r){e.valueLoc==null&&(e.valueLoc=i.getUniformLocationNoThrow(r,"value")),i.gl.uniform1f(e.valueLoc,t)}},n}();var f9=function(){function n(t,e,i){this.variableNames=["A","indices"];var r=t.slice();r[i]=e,this.outputShape=r,this.rank=r.length;var a=Je(this.rank),s=p9(t,i);this.userCode=` + void main() { + `+a+` resRC = getOutputCoords(); + setOutput(getA(`+s+`)); + } + `}return n}();function p9(n,t){var e=n.length;if(e>4)throw Error("Gather for rank "+e+" is not yet supported");if(e===1)return"int(getIndices(resRC))";for(var i=["resRC.x","resRC.y","resRC.z","resRC.w"],r=[],a=0;a1?"strides[j]":"strides";this.userCode=` + `+r+" strides = "+r+"("+this.strides+`); + void main() { + `+a+` coords = getOutputCoords(); + int flattenIndex = 0; + for (int j = 0; j < `+this.sliceDim+`; j++) { + int index = round(getIndices(coords[0], j)); + flattenIndex += index * `+s+`; + } + setOutput(getX(flattenIndex, coords[1])); + } + `}return n}();function MS(n){var t=Wt(),e=t.version+` + precision highp float; + `+t.attribute+` vec3 clipSpacePos; + `+t.attribute+` vec2 uv; + `+t.varyingVs+` vec2 resultUV; + + void main() { + gl_Position = vec4(clipSpacePos, 1); + resultUV = uv; + }`;return Q0(n,e)}function HS(n){var t=new Float32Array([-1,1,0,0,1,-1,-1,0,0,0,1,1,0,1,1,1,-1,0,1,0]);return iS(n,t)}function VS(n){var t=new Uint16Array([0,1,2,2,1,3]);return rS(n,t)}function Ss(n,t,e,i,r,a){sS(t,e);var s=aS(n),o=n.TEXTURE_2D;return ce(n,function(){return n.bindTexture(o,s)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_S,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_WRAP_T,n.CLAMP_TO_EDGE)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MIN_FILTER,n.NEAREST)}),ce(n,function(){return n.texParameteri(o,n.TEXTURE_MAG_FILTER,n.NEAREST)}),ce(n,function(){return n.texImage2D(o,0,i,t,e,0,r,a,null)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)}),s}function dp(n){return n.internalFormatFloat}function qS(n,t,e,i){var r=gs(t,e),a=r[0],s=r[1];return Ss(n,a,s,dp(i),i.textureFormatFloat,n.FLOAT)}function pp(n){return n.internalFormatHalfFloat}function GS(n,t,e,i){var r=gs(t,e),a=r[0],s=r[1];return Ss(n,a,s,pp(i),i.textureFormatFloat,i.textureTypeHalfFloat)}function fp(n){return n.downloadTextureFormat}function YS(n,t,e,i){var r=gs(t,e),a=r[0],s=r[1];return Ss(n,a,s,fp(i),n.RGBA,n.UNSIGNED_BYTE)}function mp(n){return n.internalFormatPackedFloat}function KS(n,t,e,i){var r=ha(t,e),a=r[0],s=r[1];return Ss(n,a,s,mp(i),n.RGBA,n.FLOAT)}function gp(n){return n.internalFormatPackedHalfFloat}function jS(n,t,e,i){var r=ha(t,e),a=r[0],s=r[1];return Ss(n,a,s,gp(i),n.RGBA,i.textureTypeHalfFloat)}function $S(n,t,e){var i=0,r=3*4,a=3*4+2*4;ce(n,function(){return n.bindBuffer(n.ARRAY_BUFFER,e)});var s=rp(n,t,"clipSpacePos",e,3,a,i);return s&&rp(n,t,"uv",e,2,a,r)}function XS(n,t,e,i,r,a){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)});var s,o,l;r instanceof Uint8Array?(s=new Uint8Array(e*i*4),o=n.UNSIGNED_BYTE,l=n.RGBA):(s=new Float32Array(e*i*4),o=n.FLOAT,l=a.internalFormatPackedFloat),s.set(r),ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,l,e,i,0,n.RGBA,o,s)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function JS(n,t,e){ce(n,function(){return n.bindTexture(n.TEXTURE_2D,t)}),e.data instanceof Uint8Array?ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,e.width,e.height,0,n.RGBA,n.UNSIGNED_BYTE,e.data)}):ce(n,function(){return n.texImage2D(n.TEXTURE_2D,0,n.RGBA,n.RGBA,n.UNSIGNED_BYTE,e)}),ce(n,function(){return n.bindTexture(n.TEXTURE_2D,null)})}function ZS(n,t,e,i){var r=n.createBuffer();ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,r)});var a=4,s=4,o=a*s*t*e;return ce(n,function(){return n.bufferData(n.PIXEL_PACK_BUFFER,o,n.STREAM_READ)}),ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,0)}),ce(n,function(){return n.bindBuffer(n.PIXEL_PACK_BUFFER,null)}),r}function QS(n,t,e){var i=n,r=new Float32Array(e);return i.bindBuffer(i.PIXEL_PACK_BUFFER,t),i.getBufferSubData(i.PIXEL_PACK_BUFFER,0,r),i.bindBuffer(i.PIXEL_PACK_BUFFER,null),r}function eL(n,t,e,i){var r=gs(t,e),a=r[0],s=r[1],o=4,l=new Uint8Array(M_(t*e,o));return ce(n,function(){return n.readPixels(0,0,a,s,i.downloadTextureFormat,n.UNSIGNED_BYTE,l)}),new Float32Array(l.buffer)}function tL(n,t,e,i,r,a,s,o){var l=n,u=new Float32Array(H_(a,s));return l.bindBuffer(l.PIXEL_PACK_BUFFER,t),l.getBufferSubData(l.PIXEL_PACK_BUFFER,0,u),l.bindBuffer(l.PIXEL_PACK_BUFFER,null),u}function nL(n,t,e){var i=new Float32Array(t*e*4);return ce(n,function(){return n.readPixels(0,0,e,t,n.RGBA,n.FLOAT,i)}),i}var g9={__proto__:null,createVertexShader:MS,createVertexBuffer:HS,createIndexBuffer:VS,getInternalFormatForFloat32MatrixTexture:dp,createFloat32MatrixTexture:qS,getInternalFormatForFloat16MatrixTexture:pp,createFloat16MatrixTexture:GS,getInternalFormatForUnsignedBytesMatrixTexture:fp,createUnsignedBytesMatrixTexture:YS,getInternalFormatForPackedMatrixTexture:mp,createPackedMatrixTexture:KS,getInternalFormatForFloat16PackedMatrixTexture:gp,createFloat16PackedMatrixTexture:jS,bindVertexProgramAttributeStreams:$S,uploadDenseMatrixToTexture:XS,uploadPixelDataToTexture:JS,createBufferFromOutputTexture:ZS,downloadFloat32MatrixFromBuffer:QS,downloadByteEncodedFloatMatrixFromOutputTexture:eL,downloadPackedMatrixFromBuffer:tL,downloadMatrixFromPackedOutputTexture:nL};var iL=function(){function n(t){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];var e=x.env().getNumber("WEBGL_VERSION");t!=null?(this.gl=t,X0(e,t)):this.gl=Hn(e);var i="WEBGL_color_buffer_float",r="EXT_color_buffer_half_float";if(x.env().getNumber("WEBGL_VERSION")===1){var a="OES_texture_float",s="OES_texture_half_float";if(this.textureFloatExtension=ys(this.gl,a),sn(this.gl,s))this.textureHalfFloatExtension=ys(this.gl,s);else if(x.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support half float textures, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.");if(this.colorBufferFloatExtension=this.gl.getExtension(i),sn(this.gl,r))this.colorBufferHalfFloatExtension=ys(this.gl,r);else if(x.env().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support color renderable half floats, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.")}else if(i="EXT_color_buffer_float",sn(this.gl,i))this.colorBufferFloatExtension=this.gl.getExtension(i);else if(sn(this.gl,r))this.colorBufferHalfFloatExtension=this.gl.getExtension(r);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=HS(this.gl),this.indexBuffer=VS(this.gl),this.framebuffer=oS(this.gl),this.textureConfig=ip(this.gl,this.textureHalfFloatExtension)}return Object.defineProperty(n.prototype,"debug",{get:function(){return x.env().getBool("DEBUG")},enumerable:!0,configurable:!0}),n.prototype.dispose=function(){var t=this;if(this.disposed)return;this.program!=null&&console.warn("Disposing a GPGPUContext that still has a bound WebGLProgram. This is probably a resource leak, delete the program with GPGPUContext.deleteProgram before disposing."),this.outputTexture!=null&&console.warn("Disposing a GPGPUContext that still has a bound output matrix texture. This is probably a resource leak, delete the output matrix texture with GPGPUContext.deleteMatrixTexture before disposing.");var e=this.gl;ce(e,function(){return e.finish()}),ce(e,function(){return e.bindFramebuffer(e.FRAMEBUFFER,null)}),ce(e,function(){return e.deleteFramebuffer(t.framebuffer)}),ce(e,function(){return e.bindBuffer(e.ARRAY_BUFFER,null)}),ce(e,function(){return e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)}),ce(e,function(){return e.deleteBuffer(t.indexBuffer)}),this.disposed=!0},n.prototype.createFloat32MatrixTexture=function(t,e){return this.throwIfDisposed(),qS(this.gl,t,e,this.textureConfig)},n.prototype.createFloat16MatrixTexture=function(t,e){return this.throwIfDisposed(),GS(this.gl,t,e,this.textureConfig)},n.prototype.createUnsignedBytesMatrixTexture=function(t,e){return this.throwIfDisposed(),YS(this.gl,t,e,this.textureConfig)},n.prototype.uploadPixelDataToTexture=function(t,e){this.throwIfDisposed(),JS(this.gl,t,e)},n.prototype.uploadDenseMatrixToTexture=function(t,e,i,r){this.throwIfDisposed(),XS(this.gl,t,e,i,r,this.textureConfig)},n.prototype.createFloat16PackedMatrixTexture=function(t,e){return this.throwIfDisposed(),jS(this.gl,t,e,this.textureConfig)},n.prototype.createPackedMatrixTexture=function(t,e){return this.throwIfDisposed(),KS(this.gl,t,e,this.textureConfig)},n.prototype.deleteMatrixTexture=function(t){var e=this;this.throwIfDisposed(),this.outputTexture===t&&(ap(this.gl,this.framebuffer),this.outputTexture=null),ce(this.gl,function(){return e.gl.deleteTexture(t)})},n.prototype.downloadByteEncodedFloatMatrixFromOutputTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return eL(r.gl,e,i,r.textureConfig)})},n.prototype.downloadPackedMatrixFromBuffer=function(t,e,i,r,a,s){return tL(this.gl,t,e,i,r,a,s,this.textureConfig)},n.prototype.downloadFloat32MatrixFromBuffer=function(t,e){return QS(this.gl,t,e)},n.prototype.createBufferFromTexture=function(t,e,i){this.bindTextureToFrameBuffer(t);var r=ZS(this.gl,e,i,this.textureConfig);return this.unbindTextureToFrameBuffer(),r},n.prototype.createAndWaitForFence=function(){var t=this.createFence(this.gl);return this.pollFence(t)},n.prototype.createFence=function(t){var e=this,i,r;if(x.env().getBool("WEBGL_FENCE_API_ENABLED")){var a=t,s=a.fenceSync(a.SYNC_GPU_COMMANDS_COMPLETE,0);t.flush(),r=function(){var o=a.clientWaitSync(s,0,0);return o===a.ALREADY_SIGNALED||o===a.CONDITION_SATISFIED},i=s}else x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(i=this.beginQuery(),this.endQuery(),r=function(){return e.isQueryAvailable(i,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}):r=function(){return!0};return{query:i,isFencePassed:r}},n.prototype.downloadMatrixFromPackedTexture=function(t,e,i){var r=this;return this.downloadMatrixDriver(t,function(){return nL(r.gl,e,i)})},n.prototype.createProgram=function(t){this.throwIfDisposed();var e=this.gl,i=eS(e,t),r=MS(e),a=tS(e);return ce(e,function(){return e.attachShader(a,r)}),ce(e,function(){return e.attachShader(a,i)}),nS(e,a),this.debug&&Qo(e,a),this.vertexAttrsAreBound||(this.setProgram(a),this.vertexAttrsAreBound=$S(e,this.program,this.vertexBuffer)),a},n.prototype.deleteProgram=function(t){var e=this;this.throwIfDisposed(),t===this.program&&(this.program=null),t!=null&&ce(this.gl,function(){return e.gl.deleteProgram(t)})},n.prototype.setProgram=function(t){var e=this;this.throwIfDisposed(),this.program=t,this.program!=null&&this.debug&&Qo(this.gl,this.program),ce(this.gl,function(){return e.gl.useProgram(t)})},n.prototype.getUniformLocation=function(t,e,i){return i===void 0&&(i=!0),this.throwIfDisposed(),i?cS(this.gl,t,e):hS(this.gl,t,e)},n.prototype.getAttributeLocation=function(t,e){var i=this;return this.throwIfDisposed(),ce(this.gl,function(){return i.gl.getAttribLocation(t,e)})},n.prototype.getUniformLocationNoThrow=function(t,e){return this.throwIfDisposed(),this.gl.getUniformLocation(t,e)},n.prototype.setInputMatrixTexture=function(t,e,i){this.throwIfDisposed(),this.throwIfNoProgram(),dS(this.gl,t,e,i)},n.prototype.setOutputMatrixTexture=function(t,e,i){this.setOutputMatrixTextureDriver(t,i,e)},n.prototype.setOutputPackedMatrixTexture=function(t,e,i){this.throwIfDisposed();var r=ha(e,i),a=r[0],s=r[1];this.setOutputMatrixTextureDriver(t,a,s)},n.prototype.setOutputMatrixWriteRegion=function(t,e,i,r){this.setOutputMatrixWriteRegionDriver(i,t,r,e)},n.prototype.setOutputPackedMatrixWriteRegion=function(t,e,i,r){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")},n.prototype.debugValidate=function(){this.program!=null&&Qo(this.gl,this.program),bs(this.gl)},n.prototype.executeProgram=function(){this.throwIfDisposed(),this.throwIfNoProgram();var t=this.gl;this.debug&&this.debugValidate(),ce(t,function(){return t.drawElements(t.TRIANGLES,6,t.UNSIGNED_SHORT,0)})},n.prototype.blockUntilAllProgramsCompleted=function(){var t=this;this.throwIfDisposed(),ce(this.gl,function(){return t.gl.finish()})},n.prototype.getQueryTimerExtension=function(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=ys(this.gl,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2?"EXT_disjoint_timer_query_webgl2":"EXT_disjoint_timer_query")),this.disjointQueryTimerExtension},n.prototype.getQueryTimerExtensionWebGL2=function(){return this.getQueryTimerExtension()},n.prototype.getQueryTimerExtensionWebGL1=function(){return this.getQueryTimerExtension()},n.prototype.beginQuery=function(){if(x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2(),i=t.createQuery();return t.beginQuery(e.TIME_ELAPSED_EXT,i),i}var r=this.getQueryTimerExtensionWebGL1(),a=r.createQueryEXT();return r.beginQueryEXT(r.TIME_ELAPSED_EXT,a),a},n.prototype.endQuery=function(){if(x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){var t=this.gl,e=this.getQueryTimerExtensionWebGL2();t.endQuery(e.TIME_ELAPSED_EXT);return}var i=this.getQueryTimerExtensionWebGL1();i.endQueryEXT(i.TIME_ELAPSED_EXT)},n.prototype.waitForQueryAndGetTime=function(t){return Jo(this,void 0,void 0,function(){var e=this;return Zo(this,function(i){switch(i.label){case 0:return[4,x.util.repeatedTry(function(){return e.disposed||e.isQueryAvailable(t,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))})];case 1:return i.sent(),[2,this.getQueryTime(t,x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))]}})})},n.prototype.getQueryTime=function(t,e){if(e===0)return null;if(e===2){var i=this.gl,r=i.getQueryParameter(t,i.QUERY_RESULT);return r/1e6}else{var a=this.getQueryTimerExtensionWebGL1(),r=a.getQueryObjectEXT(t,a.QUERY_RESULT_EXT);return r/1e6}},n.prototype.isQueryAvailable=function(t,e){if(e===0)return!0;if(e===2){var i=this.gl,r=this.getQueryTimerExtensionWebGL2(),a=i.getQueryParameter(t,i.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}else{var r=this.getQueryTimerExtensionWebGL1(),a=r.getQueryObjectEXT(t,r.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(r.GPU_DISJOINT_EXT)),a&&!this.disjoint}},n.prototype.pollFence=function(t){var e=this;return new Promise(function(i){e.addItemToPoll(function(){return t.isFencePassed()},function(){return i()})})},n.prototype.pollItems=function(){for(var t=v9(this.itemsToPoll.map(function(r){return r.isDoneFn})),e=0;e<=t;++e){var i=this.itemsToPoll[e].resolveFn;i()}this.itemsToPoll=this.itemsToPoll.slice(t+1)},n.prototype.addItemToPoll=function(t,e){var i=this;if(this.itemsToPoll.push({isDoneFn:t,resolveFn:e}),this.itemsToPoll.length>1)return;x.util.repeatedTry(function(){return i.pollItems(),i.itemsToPoll.length===0})},n.prototype.bindTextureToFrameBuffer=function(t){this.throwIfDisposed(),el(this.gl,t,this.framebuffer),this.debug&&bs(this.gl)},n.prototype.unbindTextureToFrameBuffer=function(){this.outputTexture!=null?(el(this.gl,this.outputTexture,this.framebuffer),this.debug&&bs(this.gl)):ap(this.gl,this.framebuffer)},n.prototype.downloadMatrixDriver=function(t,e){this.bindTextureToFrameBuffer(t);var i=e();return this.unbindTextureToFrameBuffer(),i},n.prototype.setOutputMatrixTextureDriver=function(t,e,i){this.throwIfDisposed();var r=this.gl;el(r,t,this.framebuffer),this.debug&&bs(r),this.outputTexture=t,ce(r,function(){return r.viewport(0,0,e,i)}),ce(r,function(){return r.scissor(0,0,e,i)})},n.prototype.setOutputMatrixWriteRegionDriver=function(t,e,i,r){var a=this;this.throwIfDisposed(),ce(this.gl,function(){return a.gl.scissor(t,e,i,r)})},n.prototype.throwIfDisposed=function(){if(this.disposed)throw new Error("Attempted to use disposed GPGPUContext.")},n.prototype.throwIfNoProgram=function(){if(this.program==null)throw new Error("No GPU program is currently set.")},n}();function v9(n){for(var t=0;t0&&(b.flatOffset=g.texData.slice.flatOffset),{name:t.variableNames[v],shapeInfo:b}}),s=a.map(function(g){return g.shapeInfo}),o={logicalShape:i.shape,texShape:i.texData.texShape,isUniform:!1,isPacked:i.texData.isPacked,flatOffset:null},l=_M(a,o,r,t.packedInputs),u=n.createProgram(l),c=null,h=n.getUniformLocation(u,"NAN",!1);x.env().getNumber("WEBGL_VERSION")===1&&(c=n.getUniformLocation(u,"INFINITY",!1));for(var d={},p=0;p0,l=s.isUniform?"uniform":s.texData.texShape;i+=s.shape+"_"+l+"_"+o});var r=n.userCode,a=n.constructor.name;return a+="_"+i+"_"+r,a}var S9=function(){function n(t,e,i){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t;for(var r=i.filterWidth,a=i.inChannels,s=i.strideWidth,o=i.strideHeight,l=i.padInfo,u=i.outWidth,c=i.dilationWidth,h=i.dilationHeight,d=i.dataFormat,p=l.left,f=l.top,m=a*r,g=Wt(),v=d==="channelsLast",b=v?0:1,w=v?1:2,S="",L=0;L<=1;L++)for(var N=0;N<=1;N++)S+=` + blockIndex = rc.y + `+N+`; + pos = rc.x + `+L+`; + + if(blockIndex < `+t[1]+" && pos < "+t[0]+`) { + offsetY = int(blockIndex / (`+u+")) * "+o+" - "+f+`; + d0 = offsetY + `+h+" * (pos / "+m+`); + + if(d0 < `+e[b]+` && d0 >= 0) { + + offsetX = int(mod(float(blockIndex), `+u+".) * "+s+". - "+p+`.); + d1 = offsetX + `+c+" * (int(mod(float(pos), "+m+".) / "+a+`.)); + + if(d1 < `+e[w]+` && d1 >= 0) { + + ch = int(mod(float(pos), `+a+`.)); + + if (`+v+`) { + innerDims = vec2(d1, ch); + result[`+(L*2+N)+`] = getChannel( + getA(d0, int(innerDims.x), + int(innerDims.y)), innerDims); + } else { + innerDims = vec2(d0, d1); + result[`+(L*2+N)+`] = getChannel( + getA(ch, int(innerDims.x), + int(innerDims.y)), innerDims); + } + } + } + } + `;this.userCode=` + void main() { + ivec2 rc = getOutputCoords(); + + vec4 result = vec4(0); + + int blockIndex, pos, offsetY, d0, offsetX, d1, ch; + vec2 innerDims; + + `+S+` + + `+g.output+` = result; + } + `}return n}();var L9=function(){function n(t,e,i,r,a){this.variableNames=["x"],this.outputShape=[];var s=e,o=t[3]-1;this.outputShape=t;var l,u="float("+i+") + float("+r+") * sum";a===.5?l="inversesqrt("+u+")":a===1?l="1.0/("+u+")":l="exp(log("+u+") * float(-"+a+"));",this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int r = coords[1]; + int c = coords[2]; + int d = coords[3]; + float x = getX(b, r, c, d); + float sum = 0.0; + for (int j = -`+s+"; j <= "+s+`; j++) { + int idx = d + j; + if (idx >= 0 && idx <= `+o+`) { + float z = getX(b, r, c, idx); + sum += z * z; + } + } + float val = x * `+l+`; + setOutput(val); + } + `}return n}();var I9=function(){function n(t,e,i,r,a){this.variableNames=["inputImage","outputImage","dy"],this.outputShape=[],this.outputShape=t,this.depth=t[3],this.depthRadius=e,this.bias=i,this.alpha=r,this.beta=a,this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int r = coords[1]; + int c = coords[2]; + + float result = 0.0; + for (int d = 0; d < `+this.depth+`; ++d) { + int depthBegin = int(max(0.0, float(d - `+e+`))); + int depthEnd = int(min(float(`+this.depth+`), + float(d + `+e+` + 1))); + + const int MIN_DEPTH_BEGIN = 0; + const int MAX_DEPTH_END = `+this.depth+`; + + float norm = 0.0; + for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) { + if (k < depthBegin){ + continue; + } + else if (k >= depthBegin && k < depthEnd) { + norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k); + } + else { + break; + } + } + + norm = float(`+r+") * norm + float("+i+`); + + for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){ + if (k < depthBegin){ + continue; + } + else if (k >= depthBegin && k < depthEnd){ + float dyi = -2.0 * float(`+r+`) + * float(`+a+`) + * getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d) + / norm; + if (k == d) { + dyi += pow(norm, -1.0 * `+a+`); + } + if (k == coords[3]) { + dyi *= getDy(b, r, c, d); + result += dyi; + } + } + else { + break; + } + } + } + setOutput(result); + } + `}return n}();var A9=function(){function n(t,e,i,r,a){this.variableNames=["x"],this.outputShape=[],this.packedInputs=!0,this.packedOutput=!0;var s=e,o=t[3]-1;this.outputShape=t;var l,u="float("+i+") + float("+r+") * sum";a===.5?l="inversesqrt("+u+")":a===1?l="1.0/("+u+")":l="exp(log("+u+") * float(-"+a+"));",this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int b = coords.x; + int r = coords.y; + int c = coords.z; + int d = coords.w; + + bool hasNextCol = d < `+this.outputShape[3]+`; + bool hasNextRow = c < `+this.outputShape[2]+`; + + vec4 sum = vec4(0.); + vec4 xFragAtOutputCoords = getX(b, r, c, d); + + vec4 xAtOutputCoords = vec4( + getChannel(xFragAtOutputCoords, vec2(c, d)), + hasNextCol ? + getChannel(xFragAtOutputCoords, vec2(c, d + 1)) : 0.0, + hasNextRow ? + getChannel(xFragAtOutputCoords , vec2(c + 1, d)) : 0.0, + (hasNextRow && hasNextCol) ? + getChannel(xFragAtOutputCoords, vec2(c + 1, d + 1)) : 0.0 + ); + + int firstChannel = d - `+s+`; + vec2 cache = vec2(0.); + if(firstChannel >= 0){ + vec4 firstChannelFrag = getX(b, r, c, firstChannel); + cache.x = getChannel(firstChannelFrag, vec2(c, firstChannel)); + if(hasNextRow){ + cache.y = getChannel(firstChannelFrag, vec2(c + 1, firstChannel)); + } + } + + ivec2 depth = ivec2(d, d + 1); + for (int j = - `+s+"; j <= "+s+`; j++) { + ivec2 idx = depth + j; + bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0)); + bvec2 belowUpperBound = lessThanEqual(idx, ivec2(`+o+`)); + + bool depthInRange = aboveLowerBound.x && belowUpperBound.x; + bool depthPlusOneInRange = aboveLowerBound.y && belowUpperBound.y; + + if(depthInRange || depthPlusOneInRange){ + vec4 z = vec4(0.); + vec4 xFragAtCurrentDepth; + z.xz = cache.xy; + if(depthPlusOneInRange && hasNextCol){ + xFragAtCurrentDepth = idx.y != d ? + getX(b, r, c, idx.y) : xFragAtOutputCoords; + z.y = getChannel(xFragAtCurrentDepth, vec2(c, idx.y)); + if(hasNextRow){ + z.w = getChannel(xFragAtCurrentDepth, vec2(c + 1, idx.y)); + } + } + cache.xy = z.yw; + sum += z * z; + } + } + vec4 result = xAtOutputCoords * `+l+`; + setOutput(result); + } + `}return n}();var T9=function(){function n(t){this.variableNames=["dy","maxPos"],this.outputShape=t.inShape;var e=t.strideHeight,i=t.strideWidth,r=t.dilationHeight,a=t.effectiveFilterHeight,s=t.effectiveFilterWidth,o=a-1-t.padInfo.top,l=s-1-t.padInfo.left,u=a*s-1;this.userCode=` + const ivec2 pads = ivec2(`+o+", "+l+`); + + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int d = coords[3]; + + ivec2 dyRCCorner = coords.yz - pads; + int dyRCorner = dyRCCorner.x; + int dyCCorner = dyRCCorner.y; + + // Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d). + // ? = to be determined. : = across all values in that axis. + float dotProd = 0.0; + for (int wR = 0; wR < `+a+`; + wR += `+r+`) { + float dyR = float(dyRCorner + wR) / `+e+`.0; + + if (dyR < 0.0 || dyR >= `+t.outHeight+`.0 || fract(dyR) > 0.0) { + continue; + } + int idyR = int(dyR); + + for (int wC = 0; wC < `+s+`; wC++) { + float dyC = float(dyCCorner + wC) / `+i+`.0; + + if (dyC < 0.0 || dyC >= `+t.outWidth+`.0 || + fract(dyC) > 0.0) { + continue; + } + int idyC = int(dyC); + + float dyValue = getDy(b, idyR, idyC, d); + int maxPosValue = `+u+` - int(getMaxPos(b, idyR, idyC, d)); + + // Get the current value, check it against the value from the + // position matrix. + int curPosValue = wR * `+s+` + wC; + float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0); + + dotProd += dyValue * mask; + } + } + setOutput(dotProd); + } + `}return n}(),N9=function(){function n(t){this.variableNames=["dy","maxPos"],this.outputShape=t.inShape;var e=t.strideDepth,i=t.strideHeight,r=t.strideWidth,a=t.dilationDepth,s=t.dilationHeight,o=t.dilationWidth,l=t.effectiveFilterDepth,u=t.effectiveFilterHeight,c=t.effectiveFilterWidth,h=l-1-t.padInfo.front,d=u-1-t.padInfo.top,p=c-1-t.padInfo.left,f=l*u*c-1;this.userCode=` + const ivec3 pads = ivec3(`+h+", "+d+", "+p+`); + + void main() { + ivec5 coords = getOutputCoords(); + int batch = coords.x; + int ch = coords.u; + + ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads; + int dyDCorner = dyCorner.x; + int dyRCorner = dyCorner.y; + int dyCCorner = dyCorner.z; + + // Convolve dy(?, ?, ?, ch) with pos mask(:, :, :, d) to get + // dx(xD, xR, xC, ch). + // ? = to be determined. : = across all values in that axis. + float dotProd = 0.0; + + for (int wD = 0; wD < `+l+`; + wD += `+a+`) { + float dyD = float(dyDCorner + wD) / `+e+`.0; + + if (dyD < 0.0 || dyD >= `+t.outDepth+`.0 || fract(dyD) > 0.0) { + continue; + } + int idyD = int(dyD); + + for (int wR = 0; wR < `+u+`; + wR += `+s+`) { + float dyR = float(dyRCorner + wR) / `+i+`.0; + + if (dyR < 0.0 || dyR >= `+t.outHeight+`.0 || + fract(dyR) > 0.0) { + continue; + } + int idyR = int(dyR); + + for (int wC = 0; wC < `+c+`; + wC += `+o+`) { + float dyC = float(dyCCorner + wC) / `+r+`.0; + + if (dyC < 0.0 || dyC >= `+t.outWidth+`.0 || + fract(dyC) > 0.0) { + continue; + } + int idyC = int(dyC); + + float dyValue = getDy(batch, idyD, idyR, idyC, ch); + int maxPosValue = `+f+` - + int(getMaxPos(batch, idyD, idyR, idyC, ch)); + + // Get the current value, check it against the value from the + // position matrix. + int curPosValue = + wD * `+u+" * "+c+` + + wR * `+c+` + wC; + float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0); + + dotProd += dyValue * mask; + } + } + } + setOutput(dotProd); + } + `}return n}();var vp=function(){function n(t,e,i,r,a,s,o){i===void 0&&(i=!1),r===void 0&&(r=!1),a===void 0&&(a=!1),s===void 0&&(s=null),o===void 0&&(o=!1),this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;var l=i?t[1]:t[2],u=Math.ceil(l/2),c=i?"i * 2, rc.y":"rc.y, i * 2",h=r?"rc.z, i * 2":"i * 2, rc.z",d=i?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],p=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],f="",m="";s&&(o?f=`vec4 activation(vec4 a) { + vec4 b = getPreluActivationWeightsAtOutCoords(); + `+s+` + }`:f=`vec4 activation(vec4 x) { + `+s+` + }`,m="result = activation(result);");var g=a?"result += getBiasAtOutCoords();":"";a&&this.variableNames.push("bias"),o&&this.variableNames.push("preluActivationWeights"),this.userCode=` + `+f+` + + const float sharedDimension = `+u+`.0; + + vec4 dot2x2ARowBCol(ivec3 rc) { + vec4 result = vec4(0); + for (int i = 0; i < `+u+`; i++) { + vec4 a = getMatrixA(rc.x, `+c+`); + vec4 b = getMatrixB(rc.x, `+h+`); + + // These swizzled products need to be separately added. + // See: https://github.com/tensorflow/tfjs/issues/1735 + result += (`+d[0]+" * "+p[0]+`); + result += (`+d[1]+" * "+p[1]+`); + } + return result; + } + + void main() { + ivec3 rc = getOutputCoords(); + vec4 result = dot2x2ARowBCol(rc); + + `+g+` + + `+m+` + + setOutput(result); + } + `}return n}();var x9=function(){function n(t,e,i){this.variableNames=["probs"],this.outputShape=[t,i],this.userCode=` + uniform float seed; + + void main() { + ivec2 coords = getOutputCoords(); + int batch = coords[0]; + + float r = random(seed); + float cdf = 0.0; + + for (int i = 0; i < `+(e-1)+`; i++) { + cdf += getProbs(batch, i); + + if (r < cdf) { + setOutput(float(i)); + return; + } + } + + // If no other event happened, last event happened. + setOutput(float(`+(e-1)+`)); + } + `}return n.prototype.getCustomSetupFunc=function(t){var e=this;return function(i,r){e.seedLoc==null&&(e.seedLoc=i.getUniformLocation(r,"seed")),i.gl.uniform1f(e.seedLoc,t)}},n}();var C9=function(){function n(t,e,i,r){this.variableNames=["indices"],this.outputShape=[t,e],this.userCode=` + void main() { + ivec2 coords = getOutputCoords(); + int index = round(getIndices(coords.x)); + setOutput(mix(float(`+r+"), float("+i+`), + float(index == coords.y))); + } + `}return n}();var D9=function(){function n(t){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.outputShape=t;var e=t.length;if(e===0)this.userCode=` + void main() { + setOutput(vec4(getA(), 0., 0., 0.)); + } + `;else{var i=Jt("rc",e),r=Je(e),a=R9(e,t,i),s=O9(e,t[t.length-1],t[t.length-2],i),o=E9(t,i);this.userCode=` + void main() { + `+r+` rc = getOutputCoords(); + + if(`+a+`) { + setOutput(vec4(0)); + } else { + `+s+` + + setOutput(vec4(`+o+`)); + } + } + `}}return n}();function k9(n,t){for(var e=[],i=0;i<=1;i++)for(var r=0;r<=1;r++){for(var a=(i===0?"r":"rp1")+", "+(r===0?"c":"cp1"),s=2;s "+t[0];for(var i="",r=n-2;r= "+t[r],r= `+t+`; + bool rEdge = rp1 >= `+e+`; + `}function E9(n,t){var e=n.length,i=k9(e,t);return e===1?`getA(rc), + rc + 1 >= `+n[0]+` ? 0. : getA(rc + 1), + 0, 0`:"getA("+i[0]+`), + cEdge ? 0. : getA(`+i[1]+`), + rEdge ? 0. : getA(`+i[2]+`), + rEdge || cEdge ? 0. : getA(`+i[3]+")"}var F9=function(){function n(t,e,i){this.variableNames=["x"],this.outputShape=e.map(function(u,c){return u[0]+t[c]+u[1]});var r=t.length,a=Je(r),s=e.map(function(u){return u[0]}).join(","),o=e.map(function(u,c){return u[0]+t[c]}).join(","),l=["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,r);if(r===1){this.userCode=` + int start = `+s+`; + int end = `+o+`; + + void main() { + int outC = getOutputCoords(); + if (outC < start || outC >= end) { + setOutput(float(`+i+`)); + } else { + setOutput(getX(outC - start)); + } + } + `;return}this.userCode=` + `+a+" start = "+a+"("+s+`); + `+a+" end = "+a+"("+o+`); + + void main() { + `+a+` outC = getOutputCoords(); + if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) { + setOutput(float(`+i+`)); + } else { + `+a+` coords = outC - start; + setOutput(getX(`+l+`)); + } + } + `}return n}();var W9=function(){function n(t,e,i){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.map(function(v,b){return v[0]+t[b]+v[1]});for(var r=t.length,a=Je(r),s=e.map(function(v){return v[0]}).join(","),o=e.map(function(v,b){return v[0]+t[b]}).join(","),l=Jt("rc",r),u=Jt("source",r),c=l[r-1]+" < "+this.outputShape[r-1],h=r===1?"source":"vec2("+u.slice(-2).join()+")",d=[a+" rc = outputLoc;",l[r-1]+` += 1; + if(`+c+`) { + `,r===1?"":`} + rc = outputLoc; + `+l[r-2]+` += 1; + if(`+l[r-2]+" < "+this.outputShape[r-2]+") {",r===1?"":" "+l[r-1]+` += 1; + if(`+c+") {"],p=r===1?"rc < start || rc >= end":"any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))",f="",m=0,g=r===1?2:4;m= `+t.inHeight+`) { + continue; + } + + for (int wC = 0; wC < `+d+`; + wC += `+c+`) { + int xC = xCCorner + wC; + + if (xC < 0 || xC >= `+t.inWidth+`) { + continue; + } + + float value = getX(batch, xR, xC, d); + + // If a min / max value has already been found, use it. If not, + // use the current value. + float currMinMaxValue = mix( + value, minMaxValue, minMaxValueFound); + if (value `+w+` currMinMaxValue) { + minMaxValue = value; + minMaxValueFound = 1.0; + minMaxPosition = `+(r?a?g:v:"wR * "+d+" + wC")+`; + } + } + } + setOutput(float(minMaxPosition)); + } + `;return}var S="max",L=e+"("+e+"("+e+"(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])";e==="avg"&&(L="avgValue / count");var N=Math.floor(s/4)*4,C=s%4,R=` + if (`+m+`) { + avgValue += dot(values, ones); + } else { + minMaxValue = `+S+`(values, minMaxValue); + } + `;this.userCode=` + const ivec2 strides = ivec2(`+o+", "+l+`); + const ivec2 pads = ivec2(`+p+", "+f+`); + const float initializationValue = `+b+`; + const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0); + + float count = 0.0; + + float getValue(int batch, int xR, int xC, int d) { + if (xC < 0 || xC >= `+t.inWidth+`) { + return initializationValue; + } + count += 1.0; + return getX(batch, xR, xC, d); + } + + void main() { + ivec4 coords = getOutputCoords(); + int batch = coords[0]; + int d = coords[3]; + + ivec2 xRCCorner = coords.yz * strides - pads; + int xRCorner = xRCCorner.x; + int xCCorner = xRCCorner.y; + + // max/min x(?, ?, d) to get y(yR, yC, d). + // ? = to be determined + vec4 minMaxValue = vec4(`+b+`); + float avgValue = 0.0; + count = 0.0; + + for (int wR = 0; wR < `+h+`; + wR += `+u+`) { + int xR = xRCorner + wR; + + if (xR < 0 || xR >= `+t.inHeight+`) { + continue; + } + + for (int wC = 0; wC < `+N+`; wC += 4) { + int xC = xCCorner + wC * `+c+`; + + vec4 values = vec4( + getValue(batch, xR, xC, d), + getValue(batch, xR, xC + `+c+`, d), + getValue(batch, xR, xC + 2 * `+c+`, d), + getValue(batch, xR, xC + 3 * `+c+`, d) + ); + + `+R+` + } + + int xC = xCCorner + `+N+`; + if (`+(C===1)+`) { + vec4 values = vec4( + getValue(batch, xR, xC, d), + initializationValue, + initializationValue, + initializationValue + ); + + `+R+` + } else if (`+(C===2)+`) { + vec4 values = vec4( + getValue(batch, xR, xC, d), + getValue(batch, xR, xC + `+c+`, d), + initializationValue, + initializationValue + ); + + `+R+` + } else if (`+(C===3)+`) { + vec4 values = vec4( + getValue(batch, xR, xC, d), + getValue(batch, xR, xC + `+c+`, d), + getValue(batch, xR, xC + 2 * `+c+`, d), + initializationValue + ); + + `+R+` + } + } + setOutput(`+L+`); + } + `}return n}(),yp=function(){function n(t,e,i,r,a){if(r===void 0&&(r=!1),a===void 0&&(a=!1),this.variableNames=["x"],e==="avg"&&i)throw new Error("Cannot compute positions for average pool.");var s=t.filterWidth,o=t.strideDepth,l=t.strideHeight,u=t.strideWidth,c=t.dilationDepth,h=t.dilationHeight,d=t.dilationWidth,p=t.effectiveFilterDepth,f=t.effectiveFilterHeight,m=t.effectiveFilterWidth,g=t.padInfo.front,v=t.padInfo.top,b=t.padInfo.left;this.outputShape=t.outShape;var w=e==="avg",S="0.0";if(w||(S="-1.0 / 1e-20"),i){var L=">=";this.userCode=` + const ivec3 strides = + ivec3(`+o+", "+l+", "+u+`); + const ivec3 pads = ivec3(`+g+", "+v+", "+b+`); + + void main() { + ivec5 coords = getOutputCoords(); + int batch = coords.x; + int ch = coords.u; + + ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads; + int xDCorner = xCorner.x; + int xRCorner = xCorner.y; + int xCCorner = xCorner.z; + + // max/min x(?, ?, ?, ch) to get y(yD, yR, yC, ch). + // ? = to be determined + float minMaxValue = 0.0; + float minMaxValueFound = 0.0; + int minMaxPosition = 0; + + for (int wD = 0; wD < `+p+`; + wD += `+c+`) { + int xD = xDCorner + wD; + + if (xD < 0 || xD >= `+t.inDepth+`) { + continue; + } + + for (int wR = 0; wR < `+f+`; + wR += `+h+`) { + int xR = xRCorner + wR; + + if (xR < 0 || xR >= `+t.inHeight+`) { + continue; + } + + for (int wC = 0; wC < `+m+`; + wC += `+d+`) { + int xC = xCCorner + wC; + + if (xC < 0 || xC >= `+t.inWidth+`) { + continue; + } + + float value = getX(batch, xD, xR, xC, ch); + + // If a min / max value has already been found, use it. If not, + // use the current value. + float currMinMaxValue = mix( + value, minMaxValue, minMaxValueFound); + if (value `+L+` currMinMaxValue) { + minMaxValue = value; + minMaxValueFound = 1.0; + minMaxPosition = `+(r?a?"(((batch * "+t.inDepth+" + xD) * "+t.inHeight+" + xR) * "+t.inWidth+" + xC) * "+t.inChannels+" + ch":"((xD * "+t.inHeight+" + xR) * "+t.inWidth+" + xC) * "+t.inChannels+" + ch":"wD * "+f+" * "+m+` + + wR * `+m+" + wC")+`; + } + } + } + } + setOutput(float(minMaxPosition)); + } + `;return}var N="max",C=e+"("+e+"("+e+"(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])";e==="avg"&&(C="avgValue / count");var R=Math.floor(s/4)*4,D=s%4,k=` + if (`+w+`) { + avgValue += dot(values, ones); + } else { + minMaxValue = `+N+`(values, minMaxValue); + } + `;this.userCode=` + const ivec3 strides = + ivec3(`+o+", "+l+", "+u+`); + const ivec3 pads = ivec3(`+g+", "+v+", "+b+`); + const float initializationValue = `+S+`; + const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0); + + float count = 0.0; + + float getValue(int batch, int xD, int xR, int xC, int ch) { + if (xC < 0 || xC >= `+t.inWidth+`) { + return initializationValue; + } + count += 1.0; + return getX(batch, xD, xR, xC, ch); + } + + void main() { + ivec5 coords = getOutputCoords(); + int batch = coords.x; + int ch = coords.u; + + ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads; + int xDCorner = xCorner.x; + int xRCorner = xCorner.y; + int xCCorner = xCorner.z; + + // max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch). + // ? = to be determined + vec4 minMaxValue = vec4(`+S+`); + float avgValue = 0.0; + count = 0.0; + + for (int wD = 0; wD < `+p+`; + wD += `+c+`) { + int xD = xDCorner + wD; + + if (xD < 0 || xD >= `+t.inDepth+`) { + continue; + } + + for (int wR = 0; wR < `+f+`; + wR += `+h+`) { + int xR = xRCorner + wR; + + if (xR < 0 || xR >= `+t.inHeight+`) { + continue; + } + + for (int wC = 0; wC < `+R+`; wC += 4) { + int xC = xCCorner + wC * `+d+`; + + vec4 values = vec4( + getValue(batch, xD, xR, xC, ch), + getValue(batch, xD, xR, xC + `+d+`, ch), + getValue(batch, xD, xR, xC + 2 * `+d+`, ch), + getValue(batch, xD, xR, xC + 3 * `+d+`, ch) + ); + + `+k+` + } + + int xC = xCCorner + `+R+`; + if (`+(D===1)+`) { + vec4 values = vec4( + getValue(batch, xD, xR, xC, ch), + initializationValue, + initializationValue, + initializationValue + ); + + `+k+` + } else if (`+(D===2)+`) { + vec4 values = vec4( + getValue(batch, xD, xR, xC, ch), + getValue(batch, xD, xR, xC + `+d+`, ch), + initializationValue, + initializationValue + ); + + `+k+` + } else if (`+(D===3)+`) { + vec4 values = vec4( + getValue(batch, xD, xR, xC, ch), + getValue(batch, xD, xR, xC + `+d+`, ch), + getValue(batch, xD, xR, xC + 2 * `+d+`, ch), + initializationValue + ); + + `+k+` + } + } + setOutput(`+C+`); + } + } + `}return n}();var aL=function(){function n(t,e){this.variableNames=["x"];var i=t.windowSize,r=t.batchSize,a=t.inSize,s=t.outSize;this.outputShape=[r,s];var o="0.0",l="";e==="prod"?o="1.0":e==="min"?(o="1.0 / 1e-20",l="min"):e==="max"&&(o="-1.0 / 1e-20",l="max");var u=e+"("+e+"("+e+"(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])";e==="sum"?u="sumValue":e==="prod"?u="prodValue":e==="all"?u="allValue":e==="any"&&(u="anyValue");var c=Math.floor(i/4)*4,h=i%4,d=` + if (`+(e==="sum")+`) { + sumValue += dot(values, ones); + } else if (`+(e==="prod")+`) { + vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]); + prodValue *= tmp[0] * tmp[1]; + } else { + minMaxValue = `+l+`(values, minMaxValue); + } + `,p="vec4";e==="all"?(o="1.0",d=` + bool reducedAllValue = all(values); + float floatedReducedAllValue = float(reducedAllValue); + allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0); + `,p="bvec4"):e==="any"&&(o="0.0",d=` + bool reducedAnyValue = any(values); + float floatedReducedAnyValue = float(reducedAnyValue); + anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0); + `,p="bvec4");var f="";a%i>0&&(f=` + if (inIdx < 0 || inIdx >= `+a+`) { + return initializationValue; + } + `),this.userCode=` + const float initializationValue = `+o+`; + const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0); + + float getValue(int batch, int inIdx) { + `+f+` + return getX(batch, inIdx); + } + + void main() { + ivec2 coords = getOutputCoords(); + int batch = coords[0]; + int outIdx = coords[1]; + int inOffset = outIdx * `+i+`; + + vec4 minMaxValue = vec4(`+o+`); + float prodValue = 1.0; + float sumValue = 0.0; + float allValue = 1.0; + float anyValue = 0.0; + + for (int i = 0; i < `+c+`; i += 4) { + int inIdx = inOffset + i; + `+p+" values = "+p+`( + getValue(batch, inIdx), + getValue(batch, inIdx + 1), + getValue(batch, inIdx + 2), + getValue(batch, inIdx + 3) + ); + + `+d+` + } + + int inIdx = inOffset + `+c+`; + if (`+(h===1)+`) { + `+p+" values = "+p+`( + getValue(batch, inIdx), + initializationValue, + initializationValue, + initializationValue + ); + + `+d+` + } else if (`+(h===2)+`) { + `+p+" values = "+p+`( + getValue(batch, inIdx), + getValue(batch, inIdx + 1), + initializationValue, + initializationValue + ); + + `+d+` + } else if (`+(h===3)+`) { + `+p+" values = "+p+`( + getValue(batch, inIdx), + getValue(batch, inIdx + 1), + getValue(batch, inIdx + 2), + initializationValue + ); + + `+d+` + } + setOutput(`+u+`); + } + `}return n}();var sL=function(){function n(t,e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t;for(var i="",r=0;r<4;r++){var a="thisRC = rc;";r%2===1&&(a+="thisRC.z += 1;"),r>1&&(a+="thisRC.y += 1;"),i+=` + `+a+` + `+(r>0?"if(thisRC.y < rows && thisRC.z < cols){":"")+` + int flatIndex = getFlatIndex(thisRC); + + ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex); + vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z)); + + result[`+r+`] = + getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims); + `+(r>0?"}":"")+` + `}this.userCode=` + `+U9(e)+` + `+up(t)+` + + void main() { + ivec3 rc = getOutputCoords(); + + vec4 result = vec4(0.); + + ivec3 thisRC; + int rows = `+t[1]+`; + int cols = `+t[2]+`; + + `+i+` + + setOutput(result); + } + `}return n}();function U9(n){var t=wr(["r","c","d"],n);return` + ivec3 inputCoordsFromReshapedOutCoords(int index) { + `+t+` + return ivec3(r, c, d); + } + `}var B9=function(){function n(t,e,i){this.variableNames=["dy"],this.outputShape=[],this.outputShape=e.shape;var r=e.shape,a=r[1],s=r[2],o=t.shape,l=o[1],u=o[2],c=[i&&l>1?a-1:a,i&&u>1?s-1:s],h=[i&&l>1?l-1:l,i&&u>1?u-1:u],d=c[0]/h[0],p=c[1]/h[1],f=1/d,m=1/p,g=Math.ceil(f)*2+2,v=Math.ceil(m)*2+2;this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int d = coords[3]; + int r = coords[1]; + int c = coords[2]; + + float accumulator = 0.0; + + const float heightScale = float(`+d+`); + const float widthScale = float(`+p+`); + + const float invHeightScale = float(`+f+`); + const float invWidthScale = float(`+m+`); + + const int winHeight = int(`+g+`); + const int winWidth = int(`+v+`); + + // Compute bounds for where in dy we will look + float startRLerp = floor(float(r) * invHeightScale); + int startDyR = int(startRLerp - float(winHeight / 2)); + + float startCLerp = floor(float(c) * invWidthScale); + int startDyC = int(startCLerp - float(winWidth / 2)); + + // Loop over dy + for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) { + int dyR = dyROffset + startDyR; + + // Guard against the window exceeding the bounds of dy + if (dyR < 0 || dyR >= `+l+`) { + continue; + } + + for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) { + int dyC = dyCOffset + startDyC; + + // Guard against the window exceeding the bounds of dy + if (dyC < 0 || dyC >= `+u+`) { + continue; + } + + float dxR = float(dyR) * heightScale; + int topDxRIndex = int(floor(dxR)); + int bottomDxRIndex = int(min(ceil(dxR), `+(a-1)+`.0)); + float dxRLerp = dxR - float(topDxRIndex); + float inverseDxRLerp = 1.0 - dxRLerp; + + float dxC = float(dyC) * widthScale; + int leftDxCIndex = int(floor(dxC)); + int rightDxCIndex = int(min(ceil(dxC), `+(s-1)+`.0)); + float dxCLerp = dxC - float(leftDxCIndex); + float inverseDxCLerp = 1.0 - dxCLerp; + + if (r == topDxRIndex && c == leftDxCIndex) { + // topLeft + accumulator += + getDy(b, dyR, dyC, d) * inverseDxRLerp * inverseDxCLerp; + } + + if (r == topDxRIndex && c == rightDxCIndex) { + // topRight + accumulator += getDy(b, dyR, dyC, d) * inverseDxRLerp * dxCLerp; + } + + if (r == bottomDxRIndex && c == leftDxCIndex) { + // bottomLeft + accumulator += getDy(b, dyR, dyC, d) * dxRLerp * inverseDxCLerp; + } + + if (r == bottomDxRIndex && c == rightDxCIndex) { + // bottomRight + accumulator += getDy(b, dyR, dyC, d) * dxRLerp * dxCLerp; + } + } + } + // End loop over dy + + setOutput(accumulator); + } + `}return n}();var z9=function(){function n(t,e,i,r){this.variableNames=["A"],this.outputShape=[];var a=t[0],s=t[1],o=t[2],l=t[3];this.outputShape=[a,e,i,l];var u=[r&&e>1?s-1:s,r&&i>1?o-1:o],c=[r&&e>1?e-1:e,r&&i>1?i-1:i];this.userCode=` + const vec2 effectiveInputOverOutputRatioRC = vec2( + `+u[0]/c[0]+`, + `+u[1]/c[1]+`); + const vec2 inputShapeRC = vec2(`+s+".0, "+o+`.0); + + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int d = coords[3]; + ivec2 yRC = coords.yz; + + // Fractional source index. + vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC; + + // Compute the four integer indices. + ivec2 sourceFloorRC = ivec2(sourceFracIndexRC); + ivec2 sourceCeilRC = ivec2( + min(inputShapeRC - 1.0, ceil(sourceFracIndexRC))); + + float topLeft = getA(b, sourceFloorRC.x, sourceFloorRC.y, d); + float bottomLeft = getA(b, sourceCeilRC.x, sourceFloorRC.y, d); + float topRight = getA(b, sourceFloorRC.x, sourceCeilRC.y, d); + float bottomRight = getA(b, sourceCeilRC.x, sourceCeilRC.y, d); + + vec2 fracRC = sourceFracIndexRC - vec2(sourceFloorRC); + + float top = topLeft + (topRight - topLeft) * fracRC.y; + float bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y; + float newValue = top + (bottom - top) * fracRC.x; + + setOutput(newValue); + } + `}return n}();var P9=function(){function n(t,e,i,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[];var a=t[0],s=t[1],o=t[2],l=t[3];this.outputShape=[a,e,i,l];var u=[r&&e>1?s-1:s,r&&i>1?o-1:o],c=[r&&e>1?e-1:e,r&&i>1?i-1:i];this.userCode=` + const vec3 effectiveInputOverOutputRatioRC = vec3( + `+u[0]/c[0]+`, + `+u[1]/c[1]+`, + `+u[1]/c[1]+`); + const vec3 inputShapeRC = vec3(`+s+".0, "+o+`.0, + `+o+`.0); + + float getAValue(int b, int r, int c, int d) { + return getChannel(getA(b, r, c, d), vec2(c, d)); + } + + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int d = coords[3]; + // Calculate values for next column in yRC.z. + ivec3 yRC = coords.yzz + ivec3(0, 0, 1); + + // Fractional source index. + vec3 sourceFracIndexRC = vec3(yRC) * effectiveInputOverOutputRatioRC; + + // Compute the four integer indices. + ivec3 sourceFloorRC = ivec3(sourceFracIndexRC); + ivec3 sourceCeilRC = ivec3( + min(inputShapeRC - 1.0, ceil(sourceFracIndexRC))); + + // Should we calculate next column and row elements in 2x2 packed cell. + bool hasNextCol = d < `+(l-1)+`; + bool hasNextRow = coords.z < `+(i-1)+`; + + // In parallel, construct four corners for all four components in + // packed 2x2 cell. + vec4 topLeft = vec4( + getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d), + hasNextCol ? getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d + 1) + : 0.0, + hasNextRow ? getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d) + : 0.0, + (hasNextRow && hasNextCol) ? + getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d + 1) : 0.0); + + vec4 bottomLeft = vec4( + getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d), + hasNextCol ? getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d + 1) + : 0.0, + hasNextRow ? getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d) + : 0.0, + (hasNextRow && hasNextCol) ? + getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d + 1) : 0.0); + + vec4 topRight = vec4( + getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d), + hasNextCol ? getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d + 1) + : 0.0, + hasNextRow ? getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d) + : 0.0, + (hasNextRow && hasNextCol) ? + getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d + 1) : 0.0); + + vec4 bottomRight = vec4( + getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d), + hasNextCol ? getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d + 1) + : 0.0, + hasNextRow ? getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d) + : 0.0, + (hasNextRow && hasNextCol) ? + getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d + 1) : 0.0); + + vec3 fracRC = sourceFracIndexRC - vec3(sourceFloorRC); + + vec4 top = mix(topLeft, topRight, fracRC.yyzz); + vec4 bottom = mix(bottomLeft, bottomRight, fracRC.yyzz); + vec4 newValue = mix(top, bottom, fracRC.x); + + setOutput(newValue); + } + `}return n}();var _9=function(){function n(t,e,i){this.variableNames=["dy"],this.outputShape=[],this.outputShape=e.shape;var r=e.shape,a=r[1],s=r[2],o=t.shape,l=o[1],u=o[2],c=[i&&l>1?a-1:a,i&&u>1?s-1:s],h=[i&&l>1?l-1:l,i&&u>1?u-1:u],d=c[0]/h[0],p=c[1]/h[1],f=1/d,m=1/p,g=Math.ceil(f)*2+2,v=Math.ceil(m)*2+2;this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int d = coords[3]; + int r = coords[1]; + int c = coords[2]; + + float accumulator = 0.0; + + const float heightScale = float(`+d+`); + const float widthScale = float(`+p+`); + + const float invHeightScale = float(`+f+`); + const float invWidthScale = float(`+m+`); + + const int winHeight = int(`+g+`); + const int winWidth = int(`+v+`); + + // Compute bounds for where in dy we will look + float startRLerp = floor(float(r) * invHeightScale); + int startDyR = int(floor(startRLerp - float(winHeight / 2))); + + float startCLerp = floor(float(c) * invWidthScale); + int startDyC = int(floor(startCLerp - float(winWidth / 2))); + + // Loop over dy + for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) { + int dyR = dyROffset + startDyR; + + // Guard against the window exceeding the bounds of dy + if (dyR < 0 || dyR >= `+l+`) { + continue; + } + + for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) { + int dyC = dyCOffset + startDyC; + + // Guard against the window exceeding the bounds of dy + if (dyC < 0 || dyC >= `+u+`) { + continue; + } + + float sourceFracRow = + float(`+c[0]+`) * + (float(dyR) / float(`+h[0]+`)); + + float sourceFracCol = + float(`+c[1]+`) * + (float(dyC) / float(`+h[1]+`)); + + int sourceNearestRow = int(min( + float(int(`+a+`) - 1), + `+i+` ? float(round(sourceFracRow)) : + float(floor(sourceFracRow)))); + + int sourceNearestCol = int(min( + float(int(`+s+`) - 1), + `+i+` ? float(round(sourceFracCol)) : + float(floor(sourceFracCol)))); + + if (r == sourceNearestRow && c == sourceNearestCol) { + accumulator += getDy(b, dyR, dyC, d); + } + } + } + // End loop over dy + + setOutput(accumulator); + } + `}return n}();var M9=function(){function n(t,e,i,r){this.variableNames=["A"],this.outputShape=[];var a=t[0],s=t[1],o=t[2],l=t[3];this.outputShape=[a,e,i,l];var u=[r&&e>1?s-1:s,r&&i>1?o-1:o],c=[r&&e>1?e-1:e,r&&i>1?i-1:i],h=r?"0.5":"0.0";this.userCode=` + const vec2 effectiveInputOverOutputRatioRC = vec2( + `+u[0]/c[0]+`, + `+u[1]/c[1]+`); + const vec2 inputShapeRC = vec2(`+s+".0, "+o+`.0); + + void main() { + ivec4 coords = getOutputCoords(); + int b = coords[0]; + int d = coords[3]; + ivec2 yRC = coords.yz; + + // Fractional source index. + vec2 sourceFracIndexRC = vec2(yRC) * effectiveInputOverOutputRatioRC; + + // Compute the coordinators of nearest neighbor point. + ivec2 sourceNearestRC = ivec2( + min(inputShapeRC - 1.0, floor(sourceFracIndexRC + `+h+`))); + + float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d); + + setOutput(newValue); + } + `}return n}();var H9=function(){function n(t,e){this.variableNames=["x"];var i=t.length;if(i>4)throw new Error("WebGL backend: Reverse of rank-"+i+" tensor is not yet supported");if(this.outputShape=t,i===1){this.userCode=` + void main() { + int coord = getOutputCoords(); + setOutput(getX(`+t[0]+` - coord - 1)); + } + `;return}var r=function(o){return e.indexOf(o)!==-1&&t[o]!==1?t[o]+" - coords["+o+"] - 1":"coords["+o+"]"},a=t.map(function(o,l){return r(l)}).join(","),s=Je(i);this.userCode=` + void main() { + `+s+` coords = getOutputCoords(); + setOutput(getX(`+a+`)); + } + `}return n}();var V9=function(){function n(t,e){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;var i=t.length;if(i>4)throw new Error("WebGL backend: Reverse of rank-"+i+" tensor is not yet supported");this.outputShape=t;var r=Jt("rc",i),a=r[i-1]+" + 1 < "+this.outputShape[i-1],s=r[i-2]+" + 1 < "+this.outputShape[i-2],o=Je(i);i===1?this.userCode=` + void main(){ + int rc = getOutputCoords(); + vec4 result = vec4(0.); + result.r = getChannel(getX(`+t[0]+` - rc - 1), + `+t[0]+` - rc - 1); + if(`+a+`){ + result.g = getChannel(getX(`+t[0]+` - (rc + 1) - 1), + `+t[0]+` - (rc + 1) - 1); + } + setOutput(result); + } + `:this.userCode=` + void main() { + `+o+` rc = getOutputCoords(); + vec4 result = vec4(0.); + result.r = `+l(r.slice())+`; + if(`+a+`){ + result.g = `+u(r.slice())+`; + } + if(`+s+`) { + result.b = `+c(r.slice())+`; + if(`+a+`) { + result.a = `+h(r.slice())+`; + } + } + setOutput(result); + } + `;function l(f){return d(f)}function u(f){return f[i-1]="("+f[i-1]+" + 1)",d(f)}function c(f){return f[i-2]="("+f[i-2]+" + 1)",d(f)}function h(f){return f[i-1]="("+f[i-1]+" + 1)",f[i-2]="("+f[i-2]+" + 1)",d(f)}function d(f){var m=t.map(function(b,w){return p(w,f)}),g=m.join(","),v=m.slice(-2).join(",");return"getChannel(getX("+g+"), vec2("+v+"))"}function p(f,m){return e.indexOf(f)!==-1&&t[f]!==1?t[f]+" - "+m[f]+" - 1":""+m[f]}}return n}();var oL=function(){function n(t,e,i,r,a,s,o){this.variableNames=["updates","indices","defaultValue"],this.outputShape=s;var l=Je(a.length),u=Je(s.length),c="";i===1?c="i":i===2&&(c="i, j");var h="getIndices("+c+")",d="";r===1?d="i":r===2&&(d="i, coords[1]");var p="getUpdates("+d+")",f=e>1?"strides[j]":"strides";this.userCode=` + `+l+" strides = "+l+"("+a+`); + + void main() { + `+u+` coords = getOutputCoords(); + float sum = 0.0; + bool found = false; + for (int i = 0; i < `+t+`; i++) { + int flattenedIndex = 0; + for (int j = 0; j < `+e+`; j++) { + int index = round(`+h+`); + flattenedIndex += index * `+f+`; + } + if (flattenedIndex == coords[0]) { + sum += `+p+`; + found = true; + } + } + setOutput(mix(getDefaultValue(), sum, float(found))); + } + `}return n}();var q9=function(){function n(t,e){this.variableNames=["x","segmentIds"];var i=t.windowSize,r=t.batchSize,a=t.inSize,s=t.numSegments,o=s*Math.ceil(a/i);this.outputShape=[r,o];var l="0.0",u="sumValue",c=Math.floor(i/4)*4,h=i%4,d=` + sumValue += dot(values, segFilter); + `,p="";a%i>0&&(p=` + if (inIdx < 0 || inIdx >= `+a+`) { + return initializationValue; + } + `);var f="";a%i>0&&(f=` + if (inIdx < 0 || inIdx >= `+a+`) { + return -1.0; + } + `),this.userCode=` + const float initializationValue = `+l+`; + + float getValue(int batch, int inIdx) { + `+p+` + return getX(batch, inIdx); + } + + float getSegmentIdAtIndex(int inIdx) { + `+f+` + return getSegmentIds(inIdx); + } + + void main() { + ivec2 coords = getOutputCoords(); + int batch = coords[0]; + int outIdx = coords[1]; + int inOffset = int(floor(float(outIdx) / float( + `+s+")) * float("+i+`)); + int currentSeg = int(mod(float(outIdx), float(`+s+`))); + + float sumValue = 0.0; + + for (int i = 0; i < `+c+`; i += 4) { + int inIdx = inOffset + i; + vec4 values = vec4( + getValue(batch, inIdx), + getValue(batch, inIdx + 1), + getValue(batch, inIdx + 2), + getValue(batch, inIdx + 3) + ); + + vec4 segFilter = vec4( + int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0, + int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0, + int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0, + int(getSegmentIdAtIndex(inIdx + 3)) == currentSeg ? 1 : 0 + ); + + `+d+` + } + + int inIdx = inOffset + `+c+`; + if (`+(h===1)+`) { + vec4 values = vec4( + getValue(batch, inIdx), + initializationValue, + initializationValue, + initializationValue + ); + + int inIdxSeg = int(getSegmentIdAtIndex(inIdx)); + + vec4 segFilter = vec4( + int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0, + 0, + 0, + 0 + ); + + `+d+` + } else if (`+(h===2)+`) { + vec4 values = vec4( + getValue(batch, inIdx), + getValue(batch, inIdx + 1), + initializationValue, + initializationValue + ); + + vec4 segFilter = vec4( + int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0, + int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0, + 0, + 0 + ); + + `+d+` + } else if (`+(h===3)+`) { + vec4 values = vec4( + getValue(batch, inIdx), + getValue(batch, inIdx + 1), + getValue(batch, inIdx + 2), + initializationValue + ); + + vec4 segFilter = vec4( + int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0, + int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0, + int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0, + 0 + ); + + `+d+` + } + setOutput(`+u+`); + } + `}return n}();var G9=function(){function n(t,e,i){this.variableNames=["c","a","b"],this.outputShape=e;var r,a;if(i>4)throw Error("Where for rank "+i+" is not yet supported");if(i===1)a="resRC",r="resRC";else{for(var s=["resRC.x","resRC.y","resRC.z","resRC.w"],o=[],l=[],u=0;u= 1.0) { + setOutput(getA(`+a+`)); + } else { + setOutput(getB(`+a+`)); + } + } + `}return n}();var K9=function(){function n(t){this.variableNames=["source"],this.outputShape=t,this.rank=t.length;var e=Je(this.rank),i="uniform int start["+this.rank+"];",r=Y9(this.rank),a,s=t.map(function(o,l){return"sourceLoc."+bp[l]+" = start["+l+"] + coords."+bp[l]+";"});a=` + `+e+` sourceLoc; + `+e+` coords = getOutputCoords(); + `+s.join(` +`)+` + `,this.userCode=` + `+i+` + void main() { + `+a+` + setOutput(getSource(`+r+`)); + } + `}return n.prototype.getCustomSetupFunc=function(t){var e=this;if(t.length!==this.rank)throw Error("The rank ("+this.rank+") of the program must match the "+("length of start ("+t.length+")"));return function(i,r){if(e.startLoc==null&&(e.startLoc=i.getUniformLocationNoThrow(r,"start"),e.startLoc==null))return;i.gl.uniform1iv(e.startLoc,t)}},n}(),bp=["x","y","z","w","u","v"];function Y9(n){if(n===1)return"sourceLoc";if(n<=6)return bp.slice(0,n).map(function(t){return"sourceLoc."+t}).join(",");throw Error("Slicing for rank "+n+" is not yet supported")}var j9=function(){function n(t){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.rank=t.length;var e=Je(this.rank),i=Jt("coords",this.rank),r=Jt("sourceLoc",this.rank),a=this.rank===1?"sourceLoc":"vec2("+r.slice(-2).join()+")",s="getChannel(getSource("+r.join()+"), "+a+")",o=` + result.x = `+s+`; + if (++`+i[this.rank-1]+" < "+t[this.rank-1]+`) { + ++`+r[this.rank-1]+`; + result.y = `+s+`; + --`+r[this.rank-1]+`; + } + `,l=this.rank===1?"":` + --`+i[this.rank-1]+`; + if (++`+i[this.rank-2]+" < "+t[this.rank-2]+`) { + ++`+r[this.rank-2]+`; + result.z = `+s+`; + if (++`+i[this.rank-1]+" < "+t[this.rank-1]+`) { + ++`+r[this.rank-1]+`; + result.w = `+s+`; + } + } + `,u=this.rank<=4?`sourceLoc = coords + + `+e+"("+t.map(function(c,h){return"start["+h+"]"}).join()+");":t.map(function(c,h){return r[h]+" = "+i[h]+" + start["+h+"];"}).join(` +`);this.userCode=` + uniform int start[`+this.rank+`]; + void main() { + `+e+` coords = getOutputCoords(); + `+e+` sourceLoc; + `+u+` + vec4 result = vec4(0.); + `+o+` + `+l+` + setOutput(result); + } + `}return n.prototype.getCustomSetupFunc=function(t){var e=this;if(t.length!==this.rank)throw Error("The rank ("+this.rank+") of the program must match the "+("length of start ("+t.length+")"));return function(i,r){if(e.startLoc==null&&(e.startLoc=i.getUniformLocationNoThrow(r,"start"),e.startLoc==null))return;i.gl.uniform1iv(e.startLoc,t)}},n}();var $9=function(){function n(t,e,i){this.variableNames=["x"],this.outputShape=i;var r=i.length,a=Je(i.length),s=Je(i.length),o="";if(r===1)o="coords * strides + begin";else{var l=0;o=i.map(function(u,c){return l++,i.length===1?"coords * strides["+c+"] + begin["+c+"]":"coords["+(l-1)+"] * strides["+c+"] + begin["+c+"]"}).join(",")}this.userCode=` + `+a+" begin = "+a+"("+t+`); + `+a+" strides = "+a+"("+e+`); + + void main() { + `+s+` coords = getOutputCoords(); + setOutput(getX(`+o+`)); + } + `}return n}();var X9=function(){function n(t){this.gpgpu=t,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}return n.prototype.acquireTexture=function(t,e,i){var r=uL(e,i),a=cL(t,r,i);a in this.freeTextures||(this.freeTextures[a]=[]),a in this.usedTextures||(this.usedTextures[a]=[]);var s=lL(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,i);if(this.freeTextures[a].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=s,this.log();var o=this.freeTextures[a].shift();return this.usedTextures[a].push(o),o}var l;return r===Rt.PACKED_2X2_FLOAT32?l=this.gpgpu.createPackedMatrixTexture(t[0],t[1]):r===Rt.PACKED_2X2_FLOAT16?l=this.gpgpu.createFloat16PackedMatrixTexture(t[0],t[1]):r===Rt.UNPACKED_FLOAT32?l=this.gpgpu.createFloat32MatrixTexture(t[0],t[1]):r===Rt.UNPACKED_FLOAT16?l=this.gpgpu.createFloat16MatrixTexture(t[0],t[1]):r===Rt.PACKED_4X1_UNSIGNED_BYTE&&(l=this.gpgpu.createUnsignedBytesMatrixTexture(t[0],t[1])),this.usedTextures[a].push(l),this.numUsedTextures++,this._numBytesAllocated+=s,this.log(),l},n.prototype.releaseTexture=function(t,e,i,r){if(this.freeTextures==null)return;var a=uL(i,r),s=cL(e,a,r);s in this.freeTextures||(this.freeTextures[s]=[]);var o=lL(e,a,this.gpgpu.gl,this.gpgpu.textureConfig,r),l=x.env().get("WEBGL_DELETE_TEXTURE_THRESHOLD");l!==-1&&this._numBytesAllocated>l?(this.gpgpu.deleteMatrixTexture(t),this._numBytesAllocated-=o):(this.freeTextures[s].push(t),this.numFreeTextures++,this._numBytesFree+=o),this.numUsedTextures--;var u=this.usedTextures[s],c=u.indexOf(t);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");u.splice(c,1),this.log()},n.prototype.log=function(){if(!this.logEnabled)return;var t=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",this.numFreeTextures+" / "+this.numUsedTextures,"("+t+")");var e=this._numBytesFree/this._numBytesAllocated;console.log("Bytes allocated: "+this._numBytesAllocated),console.log("Bytes unused: "+this._numBytesFree+" ("+Math.round(100*e)+"%)")},Object.defineProperty(n.prototype,"numBytesAllocated",{get:function(){return this._numBytesAllocated},enumerable:!0,configurable:!0}),Object.defineProperty(n.prototype,"numBytesFree",{get:function(){return this._numBytesFree},enumerable:!0,configurable:!0}),n.prototype.getNumUsedTextures=function(){return this.numUsedTextures},n.prototype.getNumFreeTextures=function(){return this.numFreeTextures},n.prototype.dispose=function(){var t=this;if(this.freeTextures==null)return;for(var e in this.freeTextures)this.freeTextures[e].forEach(function(i){t.gpgpu.deleteMatrixTexture(i)});for(var e in this.usedTextures)this.usedTextures[e].forEach(function(r){t.gpgpu.deleteMatrixTexture(r)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0},n}();function J9(n,t){var e=n;if(t===e.R32F)return 4;if(t===e.R16F)return 2;if(t===e.RGBA32F)return 16;if(t===n.RGBA)return 16;if(t===e.RGBA16F)return 8;throw new Error("Unknown internal format "+t)}function lL(n,t,e,i,r){var a=Z9(t,i),s;if(r){var o=ha(n[0],n[1]),l=o[0],u=o[1];s=l*u}else{var c=gs(n[0],n[1]),h=c[0],d=c[1];s=h*d}var p=J9(e,a);return s*p}function Z9(n,t){switch(n){case Rt.PACKED_2X2_FLOAT32:return mp(t);case Rt.PACKED_2X2_FLOAT16:return gp(t);case Rt.UNPACKED_FLOAT32:return dp(t);case Rt.UNPACKED_FLOAT16:return pp(t);case Rt.PACKED_4X1_UNSIGNED_BYTE:return fp(t);default:throw new Error("Unknown physical texture type "+n)}}function Q9(n){return x.env().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?n?Rt.PACKED_2X2_FLOAT32:Rt.UNPACKED_FLOAT32:n?Rt.PACKED_2X2_FLOAT16:Rt.UNPACKED_FLOAT16}function uL(n,t){if(n===an.UPLOAD)return Rt.PACKED_2X2_FLOAT32;if(n===an.RENDER||n==null)return Q9(t);if(n===an.DOWNLOAD||n===an.PIXELS)return Rt.PACKED_4X1_UNSIGNED_BYTE;throw new Error("Unknown logical texture type "+n)}function cL(n,t,e){return n[0]+"_"+n[1]+"_"+t+"_"+e}var t6=function(){function n(t,e){this.variableNames=["A"];for(var i=new Array(t.length),r=0;r5)throw Error("Tile for rank "+t+" is not yet supported");if(t===1)return"imod(resRC, "+n[0]+")";for(var e=["resRC.x","resRC.y","resRC.z","resRC.w","resRC.u"],i=[],r=0;r= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0); +`;function r6(n){return n===void 0&&(n=0),si+(` + return x > 0.0 ? 1.0 : float(`+n+`); + `)}var mL="return -x;",gL="return ceil(x);",vL="return floor(x);",a6=` + if (isnan(x)) { return 0.0; } + return sign(x); +`,s6="return float(isnan(x));",o6="return float(isinf(x));",l6="return float(!isnan(x) && !isinf(x));",u6=` + // OpenGL ES does not support round function. + // The algorithm is based on banker's rounding. + float base = floor(x); + if ((x - base) < 0.5) { + return floor(x); + } else if ((x - base) > 0.5) { + return ceil(x); + } else { + if (mod(base, 2.0) == 0.0) { + return base; + } else { + return base + 1.0; + } + } +`,yL="return exp(x);",bL="return exp(x) - 1.0;",c6=`if (x < 0.0) return NAN; + return log(x);`,h6="return log(1.0 + x);",d6="return sqrt(x);",p6="return inversesqrt(x);",f6="return 1.0 / (1.0 + exp(-1.0 * x));",m6=` + float epsilon = 1.1920928955078125e-7; + float threshold = log(epsilon) + 2.0; + + bool too_large = x > -threshold; + bool too_small = x < threshold; + + float result; + float exp_x = exp(x); + + if (too_large){ + result = x; + } + else if (too_small){ + result = exp_x; + } + else{ + result = log(exp_x + 1.0); + } + return result; +`,g6=si+` + if (abs(x) > 1.) { + return NAN; + } + return asin(x); +`,v6=si+` + if (abs(x) > 1.) { + return NAN; + } + return acos(x); +`,y6=si+` + return atan(x); +`,b6=` + float e2x = exp(x); + return (e2x - 1.0 / e2x) / 2.0; +`,w6=` + float e2x = exp(-x); + return (e2x + 1.0 / e2x) / 2.0; +`,S6=` + float e2x = exp(-2.0 * abs(x)); + return sign(x) * (1.0 - e2x) / (1.0 + e2x); +`,L6=si+"return log(x + sqrt(x * x + 1.0));",I6=si+` + if (x < 1.0) return NAN; + return log(x + sqrt(x * x - 1.0));`,A6=si+` + if ((x < -1.0) || (x > 1.0)) return NAN; + return (log(1.0 + x) - log(1.0 - x)) / 2.0;`,T6=` + // Error function is calculated approximately with elementary function. + // See "Handbook of Mathematical Functions with Formulas, + // Graphs, and Mathematical Tables", Abramowitz and Stegun. + float p = `+x.backend_util.ERF_P+`; + float a1 = `+x.backend_util.ERF_A1+`; + float a2 = `+x.backend_util.ERF_A2+`; + float a3 = `+x.backend_util.ERF_A3+`; + float a4 = `+x.backend_util.ERF_A4+`; + float a5 = `+x.backend_util.ERF_A5+`; + + float sign = sign(x); + x = abs(x); + float t = 1.0 / (1.0 + p * x); + return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x)); +`,N6="return 1.0 / x;",x6="return float(!(x >= 1.0));",C6="return float(int(x));",ol="return x;";var R6="return x;",O6=` + vec4 result = log(x); + vec4 isNaN = vec4(lessThan(x, vec4(0.0))); + result.r = isNaN.r == 1.0 ? NAN : result.r; + result.g = isNaN.g == 1.0 ? NAN : result.g; + result.b = isNaN.b == 1.0 ? NAN : result.b; + result.a = isNaN.a == 1.0 ? NAN : result.a; + + return result; +`,wL=` + vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0))); + bvec4 isNaN = isnan(x); + + result.r = isNaN.r ? x.r : result.r; + result.g = isNaN.g ? x.g : result.g; + result.b = isNaN.b ? x.b : result.b; + result.a = isNaN.a ? x.a : result.a; + + return result; +`,SL=` + vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0))); + bvec4 isNaN = isnan(x); + + result.r = isNaN.r ? x.r : result.r; + result.g = isNaN.g ? x.g : result.g; + result.b = isNaN.b ? x.b : result.b; + result.a = isNaN.a ? x.a : result.a; + + return result; +`,LL=` + vec4 result; + + result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0); + result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0); + result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0); + result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0); + + return result; +`,Is=function(){function n(t,e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.userCode=` + vec4 unaryOperation(vec4 x) { + `+e+` + } + + void main() { + vec4 x = getAAtOutCoords(); + vec4 y = unaryOperation(x); + + setOutput(y); + } + `}return n}();var E6=function(){function n(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=t;var e=t.length,i=Jt("rc",e),r=Je(e),a=EM(e,i),s=i.slice(-2),o=e<=1?"rc":"vec2("+s.join(",")+")";this.userCode=` + void main() { + `+r+` rc = getOutputCoords(); + vec4 packedInput = getA(`+a+`); + + setOutput(getChannel(packedInput, `+o+`)); + } + `}return n}();var IL=x.backend_util.segment_util,D6=x.kernel_impls.split,k6=x.kernel_impls.tile,F6=x.kernel_impls.topkImpl,W6=x.kernel_impls.whereImpl,U6=1e-7,B6=1e-4,ll={};function z6(n){return n in ll||(ll[n]={}),ll[n]}function ul(n,t){if(t===void 0&&(t=!1),n==="linear")return t?R6:n6;if(n==="relu")return t?wL:dL;if(n==="elu")return t?LL:fL;if(n==="relu6")return t?SL:pL;if(n==="prelu")return t?DS:ES;throw new Error("Activation "+n+" has not been implemented for the WebGL backend.")}var P6=128,_6=600;function M6(){return x.env().global.screen==null?1024:x.env().global.screen.height*x.env().global.screen.width*window.devicePixelRatio*_6/1024/1024}var AL=1e3,TL=function(n){z_(t,n);function t(e){var i=n.call(this)||this;if(i.pendingRead=new WeakMap,i.pendingDisposal=new WeakSet,i.dataRefCount=new WeakMap,i.numBytesInGPU=0,i.uploadWaitMs=0,i.downloadWaitMs=0,i.warnedAboutMemory=!1,i.warnedAboutCPUBackend=!1,i.pendingDeletes=0,i.disposed=!1,!x.env().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){var r=Hn(x.env().getNumber("WEBGL_VERSION"));i.binaryCache=z6(x.env().getNumber("WEBGL_VERSION")),i.gpgpu=new iL(r),i.canvas=r.canvas,i.gpgpuCreatedLocally=!0}else i.gpgpu=e,i.binaryCache={},i.gpgpuCreatedLocally=!1,i.canvas=e.gl.canvas;return i.textureManager=new X9(i.gpgpu),i.numMBBeforeWarning=M6(),i.texData=new x.DataStorage(i,x.engine()),i}return t.prototype.numDataIds=function(){return this.texData.numDataIds()+(this.cpuBackend?this.cpuBackend.numDataIds():0)-this.pendingDeletes},t.prototype.write=function(e,i,r){if((x.env().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||x.env().getBool("DEBUG"))&&this.checkNumericalProblems(e),r==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");var a={};return this.texData.set(a,{shape:i,dtype:r,values:e,usage:an.UPLOAD,refCount:1}),a},t.prototype.incRef=function(e){var i=this.texData.get(e);i.refCount++},t.prototype.decRef=function(e){if(this.texData.has(e)){var i=this.texData.get(e);i.refCount--}},t.prototype.move=function(e,i,r,a){if(x.env().getBool("DEBUG")&&this.checkNumericalProblems(i),a==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:r,dtype:a,values:i,usage:an.UPLOAD,refCount:1})},t.prototype.disposeIntermediateTensorInfo=function(e){var i=e.dataId;if(this.texData.has(i)){var r=this.texData.get(i);r.refCount--,r.refCount<1&&this.disposeData(i)}},t.prototype.readSync=function(e){var i=this.texData.get(e),r=i.values,a=i.dtype,s=i.complexTensors,o=i.slice,l=i.shape,u=i.isPacked;if(o!=null){var c=void 0;u?c=new Is(l,ol):c=new Re(l,ol);var h=this.runWebGLProgram(c,[{dataId:e,shape:l,dtype:a}],a),d=this.readSync(h.dataId);return this.disposeIntermediateTensorInfo(h),d}if(r!=null)return this.convertAndCacheOnCPU(e);if(a==="string")return r;var p=this.activeTimers!=null,f;p&&(f=x.util.now());var m;if(a==="complex64"){var g=s.real.dataSync(),v=s.imag.dataSync();m=x.backend_util.mergeRealAndImagArrays(g,v)}else m=this.getValuesFromTexture(e);return p&&(this.downloadWaitMs+=x.util.now()-f),this.convertAndCacheOnCPU(e,m)},t.prototype.read=function(e){return Jo(this,void 0,void 0,function(){var i,r,a,s,o,l,u,c,h,d,p,f,m,g,v,b,w,S,L,N,C,R;return Zo(this,function(D){switch(D.label){case 0:if(this.pendingRead.has(e))return i=this.pendingRead.get(e),[2,new Promise(function(k){return i.push(k)})];if(r=this.texData.get(e),a=r.values,s=r.shape,o=r.slice,l=r.dtype,u=r.complexTensors,c=r.isPacked,o!=null)return h=void 0,c?h=new Is(s,ol):h=new Re(s,ol),d=this.runWebGLProgram(h,[{dataId:e,shape:s,dtype:l}],l),p=this.read(d.dataId),this.disposeIntermediateTensorInfo(d),[2,p];if(a!=null)return[2,this.convertAndCacheOnCPU(e)];if(!x.env().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&x.env().getNumber("WEBGL_VERSION")===2)throw new Error("tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and WEBGL_VERSION=2 not yet supported.");return f=null,l!=="complex64"&&x.env().get("WEBGL_BUFFER_SUPPORTED")&&(m=this.decode(e),g=this.texData.get(m.dataId),f=(R=this.gpgpu).createBufferFromTexture.apply(R,[g.texture].concat(vs(s)))),this.pendingRead.set(e,[]),l!=="complex64"?[4,this.gpgpu.createAndWaitForFence()]:[3,2];case 1:D.sent(),D.label=2;case 2:return l==="complex64"?[4,Promise.all([u.real.data(),u.imag.data()])]:[3,4];case 3:return b=D.sent(),w=b[0],S=b[1],v=x.backend_util.mergeRealAndImagArrays(w,S),[3,5];case 4:f==null?v=this.getValuesFromTexture(e):(L=x.util.sizeFromShape(s),v=this.gpgpu.downloadFloat32MatrixFromBuffer(f,L)),D.label=5;case 5:return m!=null&&this.disposeIntermediateTensorInfo(m),N=this.convertAndCacheOnCPU(e,v),C=this.pendingRead.get(e),this.pendingRead.delete(e),C.forEach(function(k){return k(N)}),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e),this.pendingDeletes--),[2,N]}})})},t.prototype.checkNumericalProblems=function(e){if(e==null)return;for(var i=0;i0?[4,Promise.all(s)]:[3,2];case 1:return u=c.sent(),l.kernelMs=x.util.sum(u),l.getExtraProfileInfo=function(){return u.map(function(h,d){return{name:o[d],ms:h}}).map(function(h){return h.name+": "+h.ms}).join(", ")},[3,3];case 2:l.kernelMs={error:"WebGL query timers are not supported in this environment."},c.label=3;case 3:return this.uploadWaitMs=0,this.downloadWaitMs=0,[2,l]}})})},t.prototype.memory=function(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}},t.prototype.startTimer=function(){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:x.util.now(),endMs:null}},t.prototype.endTimer=function(e){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=x.util.now(),e)},t.prototype.getQueryTime=function(e){return Jo(this,void 0,void 0,function(){var i;return Zo(this,function(r){return x.env().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?[2,this.gpgpu.waitForQueryAndGetTime(e)]:(i=e,[2,i.endMs-i.startMs])})})},t.prototype.disposeData=function(e){if(this.pendingDisposal.has(e))return;if(this.pendingRead.has(e)){this.pendingDisposal.add(e),this.pendingDeletes++;return}if(!this.texData.has(e))return;this.releaseGPUData(e);var i=this.texData.get(e).complexTensors;i!=null&&(i.real.dispose(),i.imag.dispose()),this.texData.delete(e)},t.prototype.releaseGPUData=function(e){var i=this.texData.get(e),r=i.texture,a=i.dtype,s=i.texShape,o=i.usage,l=i.isPacked,u=i.slice,c=u&&u.origDataId||e,h=this.dataRefCount.get(c);h>1?this.dataRefCount.set(c,h-1):(this.dataRefCount.delete(c),r!=null&&(this.numBytesInGPU-=this.computeBytes(s,a),this.textureManager.releaseTexture(r,s,o,l)));var d=this.texData.get(e);d.texture=null,d.texShape=null,d.isPacked=!1,d.slice=null},t.prototype.getTexture=function(e){return this.uploadToGPU(e),this.texData.get(e).texture},t.prototype.getDataInfo=function(e){return this.texData.get(e)},t.prototype.getCPUBackend=function(){return x.env().getBool("WEBGL_CPU_FORWARD")?(this.cpuBackend==null&&(this.cpuBackend=x.engine().findBackend("cpu")),this.cpuBackend):null},t.prototype.shouldExecuteOnCPU=function(e,i){var r=this;i===void 0&&(i=P6);var a=this.getCPUBackend();return!this.warnedAboutCPUBackend&&a==null&&(console.warn("Your application contains ops that are small enough to be executed on the CPU backend, however the CPU backend cannot be found. Consider importing the CPU backend (@tensorflow/tfjs-backend-cpu) for better performance."),this.warnedAboutCPUBackend=!0),a!=null&&e.every(function(s){return r.texData.get(s.dataId).texture==null&&x.util.sizeFromShape(s.shape)x.env().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){var s=Math.floor(e.length/2),o=this.concat(e.slice(0,s),i),l=this.concat(e.slice(s),i);return this.concat([o,l],i)}if(x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].rank>1){var u=new $5(e.map(function(f){return f.shape}),i);return this.compileAndRun(u,e)}var c=x.backend_util.computeOutShape(e.map(function(f){return f.shape}),i),h=e.map(function(f){return f.as2D(-1,x.util.sizeFromShape(f.shape.slice(i)))}),d=new j5(h.map(function(f){return f.shape})),p=this.compileAndRun(d,h);return p.reshape(c)},t.prototype.neg=function(e){var i=this,r=this.tryRunOnCpuOrThrow([e],function(){return i.cpuBackend.neg(e)});if(r)return r;if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,mL,e.dtype);var a=new Re(e.shape,mL);return this.compileAndRun(a,[e])},t.prototype.batchMatMul=function(e,i,r,a){var s=r?e.shape[2]:e.shape[1],o=a?i.shape[1]:i.shape[2],l=r?e.shape[1]:e.shape[2],u=e.shape,c=u[0];if((s===1||o===1)&&l>AL){r&&(e=x.transpose(e,[0,2,1])),a&&(i=x.transpose(i,[0,2,1]));var h=o===1?e:e.as3D(c,l,1),d=o===1?2:1,p=o===1?i.as3D(c,1,l):i;return this.multiply(h,p).sum(d,!0)}var f=x.upcastType(e.dtype,i.dtype),m=new vp(e.shape,[c,s,o],r,a);return this.compileAndRun(m,[e,i],f)},t.prototype.fusedBatchMatMul=function(e){var i=e.a,r=e.b,a=e.transposeA,s=e.transposeB,o=e.bias,l=e.activation,u=e.preluActivationWeights,c=a?i.shape[2]:i.shape[1],h=s?r.shape[1]:r.shape[2],d=i.shape,p=d[0],f=x.upcastType(i.dtype,r.dtype),m=o!=null,g=u!=null,v=l?ul(l,!0):null,b=new vp(i.shape,[p,c,h],a,s,m,v,g),w=[i,r];return o&&w.push(o),u&&w.push(u),this.compileAndRun(b,w,f)},t.prototype.multiply=function(e,i){if(e.dtype==="complex64"){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),s=new CS(xS.REAL,e.shape,i.shape),o=new CS(xS.IMAG,e.shape,i.shape),l=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag),this.makeComplexComponentTensorInfo(i,a.complexTensors.real),this.makeComplexComponentTensorInfo(i,a.complexTensors.imag)],u=this.compileAndRun(s,l),c=this.compileAndRun(o,l),h=this.complex(u,c);return u.dispose(),c.dispose(),h}var d=x.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var r=this.texData.get(e.dataId),a=this.texData.get(i.dataId),p=IM(e.shape,i.shape,r.values,a.values,d),f=p[0],m=p[1];return this.makeOutput(m,d,f)}if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,OS,e.dtype);var g=new wt(OS,e.shape,i.shape);return this.compileAndRun(g,[e,i],e.dtype)},t.prototype.localResponseNormalization4D=function(e,i,r,a,s){var o=x.env().getBool("WEBGL_PACK_NORMALIZATION")?new A9(e.shape,i,r,a,s):new L9(e.shape,i,r,a,s);return this.compileAndRun(o,[e])},t.prototype.LRNGrad=function(e,i,r,a,s,o,l){var u=new I9(i.shape,a,s,o,l);return this.compileAndRun(u,[i,r,e])},t.prototype.tile=function(e,i){if(e.dtype==="string"){var r=this.readSync(e.dataId),a=r.map(function(l){return x.util.decodeString(l)}),s=x.buffer(e.shape,e.dtype,a);return k6(s,i)}var o=new t6(e.shape,i);return this.compileAndRun(o,[e])},t.prototype.pad=function(e,i,r){var a=x.env().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new W9(e.shape,i,r):new F9(e.shape,i,r);return this.compileAndRun(a,[e])},t.prototype.gather=function(e,i,r){var a=this,s=this.tryRunOnCpuOrThrow([e,i],function(){return a.cpuBackend.gather(e,i,r)});if(s)return s;var o=new f9(e.shape,i.size,r);return this.compileAndRun(o,[e,i])},t.prototype.batchToSpaceND=function(e,i,r){x.util.assert(e.rank<=4,function(){return"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(h,d){return h*d}),s=x.backend_util.getReshaped(e.shape,i,a),o=x.backend_util.getPermuted(s.length,i.length),l=x.backend_util.getReshapedPermuted(e.shape,i,a),u=x.backend_util.getSliceBeginCoords(r,i.length),c=x.backend_util.getSliceSize(l,r,i.length);return x.transpose(e.reshape(s),o).reshape(l).slice(u,c)},t.prototype.spaceToBatchND=function(e,i,r){x.util.assert(e.rank<=4,function(){return"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet"});var a=i.reduce(function(p,f){return p*f}),s=[[0,0]];s.push.apply(s,r);for(var o=1+i.length;ox.env().get("WEBGL_MAX_TEXTURES_IN_SHADER")){var i=Math.floor(e.length/2),r=this.addN(e.slice(0,i)),a=this.addN(e.slice(i));return this.addN([r,a])}var s=e.map(function(c){return c.dtype}).reduce(function(c,h){return x.upcastType(c,h)}),o=e.map(function(c){return c.shape}),l=x.env().getBool("WEBGL_PACK"),u=l?new RM(e[0].shape,o):new CM(e[0].shape,o);return this.compileAndRun(u,e,s)},t.prototype.subtract=function(e,i){if(e.dtype==="complex64"&&i.dtype==="complex64")return this.complexSeparableBinaryOp(e,i,hp);var r=x.upcastType(e.dtype,i.dtype);if(this.shouldExecuteOnCPU([e,i])){var a=this.texData.get(e.dataId),s=this.texData.get(i.dataId),o=NM(e.shape,i.shape,a.values,s.values,r),l=o[0],u=o[1];return this.makeOutput(u,r,l)}if(x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"))return this.packedBinaryOp(e,i,hp,e.dtype);var c=new wt(hp,e.shape,i.shape);return this.compileAndRun(c,[e,i],r)},t.prototype.pow=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS"),a=r?new ki(D5,e.shape,i.shape):new wt(y5,e.shape,i.shape),s=x.upcastType(e.dtype,i.dtype);return this.compileAndRun(a,[e,i],s)},t.prototype.ceil=function(e){if(this.shouldExecuteOnCPU([e])){var i=vM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,gL,e.dtype);var r=new Re(e.shape,gL);return this.compileAndRun(r,[e])},t.prototype.floor=function(e){if(this.shouldExecuteOnCPU([e])){var i=wM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,vL,e.dtype);var r=new Re(e.shape,vL);return this.compileAndRun(r,[e])},t.prototype.sign=function(e){var i=new Re(e.shape,a6);return this.compileAndRun(i,[e])},t.prototype.isNaN=function(e){var i=new Re(e.shape,s6);return this.compileAndRun(i,[e],"bool")},t.prototype.isInf=function(e){var i=new Re(e.shape,o6);return this.compileAndRun(i,[e],"bool")},t.prototype.isFinite=function(e){var i=new Re(e.shape,l6);return this.compileAndRun(i,[e],"bool")},t.prototype.round=function(e){var i=new Re(e.shape,u6);return this.compileAndRun(i,[e])},t.prototype.exp=function(e){if(this.shouldExecuteOnCPU([e])){var i=yM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,yL,e.dtype);var r=new Re(e.shape,yL);return this.compileAndRun(r,[e])},t.prototype.expm1=function(e){if(this.shouldExecuteOnCPU([e])){var i=bM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,bL,e.dtype);var r=new Re(e.shape,bL);return this.compileAndRun(r,[e])},t.prototype.softmax=function(e,i){var r=x.util.parseAxisParam([i],e.shape),a=x.max(e,r),s=x.backend_util.expandShapeToKeepDim(a.shape,r),o=this.subtract(e,a.reshape(s)),l=this.exp(o),u=this.sum(l,r).reshape(s);return x.div(l,u)},t.prototype.log=function(e){if(this.shouldExecuteOnCPU([e])){var i=SM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,O6,e.dtype);var r=new Re(e.shape,c6);return this.compileAndRun(r,[e])},t.prototype.log1p=function(e){var i=new Re(e.shape,h6);return this.compileAndRun(i,[e])},t.prototype.sqrt=function(e){var i=new Re(e.shape,d6);return this.compileAndRun(i,[e])},t.prototype.rsqrt=function(e){if(this.shouldExecuteOnCPU([e])){var i=AM(this.texData.get(e.dataId).values,e.dtype);return this.makeOutput(e.shape,e.dtype,i)}var r=new Re(e.shape,p6);return this.compileAndRun(r,[e])},t.prototype.reciprocal=function(e){var i=new Re(e.shape,N6);return this.compileAndRun(i,[e])},t.prototype.relu=function(e){var i;return x.env().getBool("WEBGL_PACK")?i=new Is(e.shape,wL):i=new Re(e.shape,dL),this.compileAndRun(i,[e])},t.prototype.relu6=function(e){var i;return x.env().getBool("WEBGL_PACK")?i=new Is(e.shape,SL):i=new Re(e.shape,pL),this.compileAndRun(i,[e])},t.prototype.prelu=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new ki(DS,e.shape,i.shape):new wt(ES,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.elu=function(e){if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,LL,e.dtype);var i=new Re(e.shape,fL);return this.compileAndRun(i,[e])},t.prototype.eluDer=function(e,i){var r=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new ki(k5,e.shape,i.shape):new wt(O5,e.shape,i.shape);return this.compileAndRun(r,[e,i])},t.prototype.selu=function(e){var i=new Re(e.shape,i6);return this.compileAndRun(i,[e])},t.prototype.int=function(e){var i=new Re(e.shape,C6);return this.compileAndRun(i,[e],"int32")},t.prototype.clip=function(e,i,r){var a;x.env().getBool("WEBGL_PACK_CLIP")?a=new Y5(e.shape):a=new G5(e.shape);var s=a.getCustomSetupFunc(i,r);return this.compileAndRun(a,[e],null,s)},t.prototype.abs=function(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){var i=mM(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,i)}if(x.env().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,hL,e.dtype);var r=new Re(e.shape,hL);return this.compileAndRun(r,[e])},t.prototype.complexAbs=function(e){var i=this.texData.get(e.dataId),r=new K5(e.shape),a=[this.makeComplexComponentTensorInfo(e,i.complexTensors.real),this.makeComplexComponentTensorInfo(e,i.complexTensors.imag)];return this.compileAndRun(r,a)},t.prototype.sigmoid=function(e){var i=new Re(e.shape,f6);return this.compileAndRun(i,[e])},t.prototype.softplus=function(e){var i=new Re(e.shape,m6);return this.compileAndRun(i,[e])},t.prototype.asin=function(e){var i=new Re(e.shape,g6);return this.compileAndRun(i,[e])},t.prototype.acos=function(e){var i=new Re(e.shape,v6);return this.compileAndRun(i,[e])},t.prototype.atan=function(e){var i=new Re(e.shape,y6);return this.compileAndRun(i,[e])},t.prototype.sinh=function(e){var i=new Re(e.shape,b6);return this.compileAndRun(i,[e])},t.prototype.cosh=function(e){var i=new Re(e.shape,w6);return this.compileAndRun(i,[e])},t.prototype.tanh=function(e){var i=new Re(e.shape,S6);return this.compileAndRun(i,[e])},t.prototype.asinh=function(e){var i=new Re(e.shape,L6);return this.compileAndRun(i,[e])},t.prototype.acosh=function(e){var i=new Re(e.shape,I6);return this.compileAndRun(i,[e])},t.prototype.atanh=function(e){var i=new Re(e.shape,A6);return this.compileAndRun(i,[e])},t.prototype.erf=function(e){var i=new Re(e.shape,T6);return this.compileAndRun(i,[e])},t.prototype.step=function(e,i){var r=new Re(e.shape,r6(i));return this.compileAndRun(r,[e])},t.prototype.conv2dByMatMul=function(e,i,r,a,s,o){var l=e.shape,u=this.texData.get(e.dataId),c=r.inChannels,h=l[0]*l[1]*l[2],d=r.outChannels,p=r.dataFormat==="channelsLast",f=!1,m=!1,g=(h===1||d===1)&&c>AL,v=l[2]%2!==0&&!!u.isPacked;if(g||!x.env().getBool("WEBGL_LAZILY_UNPACK")||!x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")||!v){var b=p?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],w=x.reshape(e,[1,b,r.inChannels]),S=x.reshape(i,[1,r.inChannels,r.outChannels]),L=this.fusedBatchMatMul({a:w,b:S,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o});return x.reshape(L,r.outShape)}var N=p?l[0]*l[1]*(l[2]+1):l[0]*l[2]*(l[3]+1),C={dataId:e.dataId,shape:[1,N,r.inChannels],dtype:e.dtype},R=u.shape;u.shape=u.shape.slice(),u.shape[u.shape.length-2]++,x.util.assert(ws(u.shape,C.shape),function(){return"packed reshape "+u.shape+" to "+C.shape+" isn't free"});var D=x.reshape(i,[1,r.inChannels,r.outChannels]),k=this.fusedBatchMatMul({a:C,b:D,transposeA:f,transposeB:m,bias:a,activation:s,preluActivationWeights:o}),W=this.texData.get(k.dataId);return x.util.assert(W.isPacked,function(){return"batchMatMul result is expected to be packed"}),u.shape=R,W.shape=r.outShape,x.engine().makeTensorFromDataId(k.dataId,r.outShape,k.dtype)},t.prototype.conv2dWithIm2Row=function(e,i,r,a,s,o){var l=r.filterWidth,u=r.filterHeight,c=r.inChannels,h=r.outWidth,d=r.outHeight,p=r.dataFormat,f=p==="channelsLast",m=l*u*c,g=d*h,v=[m,g],b=!0,w=!1,S=e.squeeze([0]),L=i.reshape([1,m,-1]),N=new S9(v,S.shape,r),C=this.compileAndRun(N,[S]).reshape([1,v[0],v[1]]),R=a!=null,D=o!=null,k=s?ul(s,!0):null,W=new vp(C.shape,[1,g,r.outChannels],b,w,R,k,D),F=[C,L];a&&F.push(a),D&&F.push(o);var P=this.compileAndRun(W,F);return f?P.reshape([1,d,h,r.outChannels]):P.reshape([1,r.outChannels,d,h])},t.prototype.fusedConv2d=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights;if(a.filterHeight===1&&a.filterWidth===1&&a.dilationHeight===1&&a.dilationWidth===1&&a.strideHeight===1&&a.strideWidth===1&&(a.padInfo.type==="SAME"||a.padInfo.type==="VALID"))return this.conv2dByMatMul(i,r,a,s,o,l);if(x.env().getBool("WEBGL_CONV_IM2COL")&&i.shape[0]===1)return this.conv2dWithIm2Row(i,r,a,s,o,l);var u=s!=null,c=l!=null,h=o?ul(o,!1):null,d=new kS(a,u,h,c),p=[i,r];return s&&p.push(s),l&&p.push(l),this.compileAndRun(d,p)},t.prototype.conv2d=function(e,i,r){if(r.filterHeight===1&&r.filterWidth===1&&r.dilationHeight===1&&r.dilationWidth===1&&r.strideHeight===1&&r.strideWidth===1&&(r.padInfo.type==="SAME"||r.padInfo.type==="VALID"))return this.conv2dByMatMul(e,i,r);if(x.env().getBool("WEBGL_CONV_IM2COL")&&e.shape[0]===1)return this.conv2dWithIm2Row(e,i,r);var a=new kS(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerInput=function(e,i,r){var a=new J5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv2dDerFilter=function(e,i,r){var a=new X5(r);return this.compileAndRun(a,[e,i])},t.prototype.fusedDepthwiseConv2D=function(e){var i=e.input,r=e.filter,a=e.convInfo,s=e.bias,o=e.activation,l=e.preluActivationWeights,u=x.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&a.strideWidth<=2&&a.outChannels/a.inChannels===1,c=o?ul(o,u):null,h=[i,r],d=s!=null,p=l!=null;d&&h.push(s),p&&h.push(l);var f;return u?(f=new WS(a,d,c,p),this.compileAndRun(f,h)):(f=new FS(a,d,c,p),this.compileAndRun(f,h))},t.prototype.depthwiseConv2D=function(e,i,r){var a;return x.env().getBool("WEBGL_PACK_DEPTHWISECONV")&&r.strideWidth<=2&&r.outChannels/r.inChannels===1?(a=new WS(r),this.compileAndRun(a,[e,i])):(a=new FS(r),this.compileAndRun(a,[e,i]))},t.prototype.depthwiseConv2DDerInput=function(e,i,r){var a=new t9(r);return this.compileAndRun(a,[e,i])},t.prototype.depthwiseConv2DDerFilter=function(e,i,r){var a=new e9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3d=function(e,i,r){var a=new n9(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerInput=function(e,i,r){var a=new Q5(r);return this.compileAndRun(a,[e,i])},t.prototype.conv3dDerFilter=function(e,i,r){var a=new Z5(r);return this.compileAndRun(a,[e,i])},t.prototype.cast=function(e,i){return x.backend_util.castTensor(e,i,this)},t.prototype.unstack=function(e,i){for(var r=e.shape[i],a=new Array(e.rank-1),s=0,o=0;o1,function(){return"blockSize should be > 1 for depthToSpace, but was: "+i});var a=e.shape[0],s=r==="NHWC"?e.shape[1]:e.shape[2],o=r==="NHWC"?e.shape[2]:e.shape[3],l=r==="NHWC"?e.shape[3]:e.shape[1],u=s*i,c=o*i,h=l/(i*i),d=r==="NHWC"?[a,u,c,h]:[a,h,u,c],p=new s9(d,i,r);return this.compileAndRun(p,[e])},t.prototype.split=function(e,i,r){return D6(e,i,r)},t.prototype.scatterND=function(e,i,r){var a=x.backend_util.calculateShapes(i,e,r),s=a.sliceRank,o=a.numUpdates,l=a.sliceSize,u=a.strides,c=a.outputSize,h=[c/l,l],d=e.reshape([o,s]),p=i.reshape([o,l]);if(c===0)return x.backend_util.reshapeTensor(x.tensor([]),r);var f=x.scalar(0),m=new oL(o,s,d.rank,p.rank,u,h),g=this.compileAndRun(m,[p,d,f]);return g.reshape(r)},t.prototype.sparseToDense=function(e,i,r,a){var s=x.backend_util.calculateShapes(i,e,r),o=s.sliceRank,l=s.numUpdates,u=s.strides,c=s.outputSize,h=!1,d=new oL(l,o,e.rank,i.rank,u,[c,1],h),p=this.compileAndRun(d,[i,e,a]);return p.reshape(r)},t.prototype.fft=function(e){var i=!1;return this.fftImpl(e,i)},t.prototype.ifft=function(e){var i=!0;return this.fftImpl(e,i)},t.prototype.fftImpl=function(e,i){var r=this.texData.get(e.dataId),a=new _S(PS.REAL,e.shape,i),s=new _S(PS.IMAG,e.shape,i),o=[this.makeComplexComponentTensorInfo(e,r.complexTensors.real),this.makeComplexComponentTensorInfo(e,r.complexTensors.imag)],l=this.compileAndRun(a,o),u=this.compileAndRun(s,o),c=this.complex(l,u).as2D(e.shape[0],e.shape[1]);return l.dispose(),u.dispose(),c},t.prototype.gatherND=function(e,i){var r=i.shape,a=r[r.length-1],s=x.backend_util.prepareAndValidate(e,i),o=s[0],l=s[1],u=s[2],c=s[3],h=i.reshape([l,a]),d=e.reshape([e.size/u,u]),p=new m9(a,c,[l,u]),f=this.compileAndRun(p,[d,h]);return f.reshape(o)},t.prototype.fill=function(e,i,r){if(r=r||x.util.inferDtype(i),r==="string"){var a=x.util.getArrayFromDType(r,x.util.sizeFromShape(e));return a.fill(i),x.engine().makeTensor(a,e,r,this)}else{var s=new d9(e,i),o=s.getCustomSetupFunc(i);return this.compileAndRun(s,[],r,o)}},t.prototype.onesLike=function(e){if(e.dtype==="string")throw new Error("onesLike is not supported under string dtype");return this.fill(e.shape,1,e.dtype)},t.prototype.zerosLike=function(e){return this.fill(e.shape,e.dtype==="string"?"":0,e.dtype)},t.prototype.linspace=function(e,i,r){return x.backend_util.linspaceImpl(e,i,r)},t.prototype.makeTensorInfo=function(e,i,r){var a=this.write(r,e,i);return this.texData.get(a).usage=null,{dataId:a,shape:e,dtype:i}},t.prototype.makeOutput=function(e,i,r){var a=this.makeTensorInfo(e,i,r).dataId;return x.engine().makeTensorFromDataId(a,e,i,this)},t.prototype.unpackTensor=function(e){var i=new E6(e.shape);return this.runWebGLProgram(i,[e],e.dtype)},t.prototype.packTensor=function(e){var i=new D9(e.shape),r=!0;return this.runWebGLProgram(i,[e],e.dtype,null,r)},t.prototype.packedReshape=function(e,i){var r=[yr(e.shape)].concat(br(e.shape)),a={dtype:e.dtype,shape:r,dataId:e.dataId},s=[yr(i)].concat(br(i)),o=new sL(s,r),l=!0,u=this.runWebGLProgram(o,[a],e.dtype,null,l);return{dataId:u.dataId,shape:i,dtype:u.dtype}},t.prototype.decode=function(e){var i=this.texData.get(e),r=i.isPacked,a=i.shape,s=i.dtype,o=tl(a),l;r?l=new a9(o):l=new r9(o);var u=!0,c=this.runWebGLProgram(l,[{shape:o,dtype:s,dataId:e}],s,null,u);return{dtype:s,shape:a,dataId:c.dataId}},t.prototype.runWebGLProgram=function(e,i,r,a,s){var o=this;s===void 0&&(s=!1);var l=this.makeTensorInfo(e.outputShape,r),u=this.texData.get(l.dataId);if(e.packedOutput&&(u.isPacked=!0),e.outPackingScheme===ms.DENSE){var c=vs(e.outputShape);u.texShape=c.map(function(w){return w*2})}if(e.outTexUsage!=null&&(u.usage=e.outTexUsage),x.util.sizeFromShape(l.shape)===0)return u.values=x.util.getTypedArrayFromDType(l.dtype,0),l;var h=[],d=i.map(function(w){if(w.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");var S=o.texData.get(w.dataId);if(S.texture==null){if(!e.packedInputs&&x.util.sizeFromShape(w.shape)<=x.env().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:w.shape,texData:null,isUniform:!0,uniformValues:S.values};e.packedInputs&&(S.isPacked=!0,S.shape=w.shape)}else if(!!S.isPacked!==!!e.packedInputs)w=S.isPacked?o.unpackTensor(w):o.packTensor(w),h.push(w),S=o.texData.get(w.dataId);else if(S.isPacked&&!ws(S.shape,w.shape)){var L=w,N=w.shape;w.shape=S.shape,w=o.packedReshape(w,N),h.push(w),S=o.texData.get(w.dataId),L.shape=N}return o.uploadToGPU(w.dataId),{shape:w.shape,texData:S,isUniform:!1}});this.uploadToGPU(l.dataId);var p={shape:l.shape,texData:u,isUniform:!1},f=w9(e,d,p),m=this.getAndSaveBinary(f,function(){return y9(o.gpgpu,e,d,p)}),g=this.activeTimers!=null,v;if(g&&(v=this.startTimer()),b9(this.gpgpu,m,d,p,a),h.forEach(function(w){return o.disposeIntermediateTensorInfo(w)}),g&&(v=this.endTimer(v),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(v)})),!x.env().getBool("WEBGL_LAZILY_UNPACK")&&u.isPacked&&s===!1){var b=this.unpackTensor(l);return this.disposeIntermediateTensorInfo(l),b}return l},t.prototype.compileAndRun=function(e,i,r,a,s){s===void 0&&(s=!1),r=r||i[0].dtype;var o=this.runWebGLProgram(e,i,r,a,s);return x.engine().makeTensorFromDataId(o.dataId,o.shape,o.dtype)},t.prototype.getAndSaveBinary=function(e,i){return e in this.binaryCache||(this.binaryCache[e]=i()),this.binaryCache[e]},t.prototype.getTextureManager=function(){return this.textureManager},t.prototype.dispose=function(){var e=this;if(this.disposed)return;if(!x.env().getBool("IS_TEST")){var i=Object.keys(this.binaryCache);i.forEach(function(r){e.gpgpu.deleteProgram(e.binaryCache[r].webGLProgram),delete e.binaryCache[r]})}this.textureManager.dispose(),this.canvas!=null&&typeof HTMLCanvasElement!="undefined"&&this.canvas instanceof HTMLCanvasElement?this.canvas.remove():this.canvas=null,this.gpgpuCreatedLocally&&(this.gpgpu.program=null,this.gpgpu.dispose()),this.disposed=!0},t.prototype.floatPrecision=function(){var e=this;return this.floatPrecisionValue==null&&(this.floatPrecisionValue=x.tidy(function(){if(!x.env().get("WEBGL_RENDER_FLOAT32_ENABLED")){var i=x.env().getBool("DEBUG");x.env().set("DEBUG",!1);var r=e.abs(x.scalar(1e-8)).dataSync()[0];if(x.env().set("DEBUG",i),r>0)return 32}return 16})),this.floatPrecisionValue},t.prototype.epsilon=function(){return this.floatPrecision()===32?U6:B6},t.prototype.uploadToGPU=function(e){var i,r=this.texData.get(e),a=r.shape,s=r.dtype,o=r.values,l=r.texture,u=r.usage,c=r.isPacked;if(l!=null)return;var h=this.activeTimers!=null,d;h&&(d=x.util.now());var p=r.texShape;if(p==null&&(p=fS(a,c),r.texShape=p),o!=null){var f=tl(a),m=void 0,g=p[1],v=p[0],b=o instanceof Uint8Array;c?(i=ha(p[0],p[1]),g=i[0],v=i[1],m=new h9(f,[v,g],b)):m=new c9(f,[v,g],b);var w=this.makeTensorInfo([v,g],s);b?this.texData.get(w.dataId).usage=an.PIXELS:this.texData.get(w.dataId).usage=an.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(w.dataId),g,v,o);var S=!0,L=this.runWebGLProgram(m,[w],s,null,S),N=this.texData.get(L.dataId);r.texture=N.texture,r.texShape=N.texShape,r.isPacked=N.isPacked,r.usage=N.usage,this.disposeIntermediateTensorInfo(w),this.texData.delete(L.dataId),r.values=null,h&&(this.uploadWaitMs+=x.util.now()-d)}else{var C=this.acquireTexture(p,u,s,c);r.texture=C}},t.prototype.convertAndCacheOnCPU=function(e,i){var r=this.texData.get(e),a=r.dtype;return this.releaseGPUData(e),i!=null&&(r.values=H6(i,a)),r.values},t.prototype.acquireTexture=function(e,i,r,a){if(this.numBytesInGPU+=this.computeBytes(e,r),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){var s=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn("High memory usage in GPU: "+s+" MB, most likely due to a memory leak")}return this.textureManager.acquireTexture(e,i,a)},t.prototype.computeBytes=function(e,i){return e[0]*e[1]*x.util.bytesPerElement(i)},t.prototype.tryRunOnCpuOrThrow=function(e,i){if(this.shouldExecuteOnCPU(e))try{return i()}catch(r){if(x.env().getBool("IS_TEST"))throw new Error("CPU forwarding failed")}return null},t}(x.KernelBackend);function H6(n,t){if(t==="float32"||t==="complex64")return n;if(t==="int32"||t==="bool"){for(var e=t==="int32"?new Int32Array(n.length):new Uint8Array(n.length),i=0;i 0. ? NAN : result.r; + result.g = isNaN.g > 0. ? NAN : result.g; + result.b = isNaN.b > 0. ? NAN : result.b; + result.a = isNaN.a > 0. ? NAN : result.a; +`;function cl(n){return function(t){var e=t.inputs,i=t.backend,r=e.x,a=i,s=new Re(r.shape,n);return a.runWebGLProgram(s,[r],r.dtype)}}function wp(n,t,e,i){return function(r){var a=r.inputs,s=r.backend,o=a,l=o.a,u=o.b,c=s,h=x.env().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new ki(t,l.shape,u.shape,!!e):new wt(n,l.shape,u.shape),d=i||l.dtype,p=c.runWebGLProgram(h,[l,u],d);return p}}var K6=G6+` + return atan(a, b); +`,j6=` + vec4 result = atan(a, b); + vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0)); + `+Y6+` + return result; +`,$6=wp(K6,j6),X6={kernelName:x.Atan2,backendName:"webgl",kernelFunc:$6};function Sp(n){var t=n.inputs,e=n.backend,i=t.x;return e.incRef(i.dataId),{dataId:i.dataId,shape:i.shape,dtype:i.dtype}}var J6={kernelName:x.Identity,backendName:"webgl",kernelFunc:Sp};function Z6(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x;da(r,"avgPool");var a=i.filterSize,s=i.strides,o=i.pad,l=i.dimRoundingMode,u=1;x.util.assert(x.backend_util.eitherStridesOrDilationsAreOne(s,u),function(){return"Error in avgPool: Either strides or dilations must be 1. "+("Got strides "+s+" and dilations '"+u+"'")});var c=x.backend_util.computePool2DInfo(r.shape,a,s,u,o,l);if(c.filterWidth===1&&c.filterHeight===1&&x.util.arraysEqual(c.inShape,c.outShape))return Sp({inputs:{x:r},backend:e});var h=new Ls(c,"avg",!1);return e.runWebGLProgram(h,[r],"float32")}var Q6={kernelName:x.AvgPool,backendName:"webgl",kernelFunc:Z6};function e8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.dy,a=t.input,s=a;da([r,a],"avgPoolBackprop");var o=i.filterSize,l=i.strides,u=i.pad,c=x.backend_util.computePool2DInfo(s.shape,o,l,1,u),h=new m5(c);return e.runWebGLProgram(h,[r],s.dtype)}var t8={kernelName:x.AvgPoolBackprop,backendName:"webgl",kernelFunc:e8};var n8=function(){function n(t,e,i,r,a,s){this.outputShape=[],this.variableNames=["x","mean","variance"],x.backend_util.assertAndGetBroadcastShape(t,e),x.backend_util.assertAndGetBroadcastShape(t,i);var o="0.0";r!=null&&(x.backend_util.assertAndGetBroadcastShape(t,r),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");var l="1.0";a!=null&&(x.backend_util.assertAndGetBroadcastShape(t,a),this.variableNames.push("scale"),l="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=` + void main() { + float x = getXAtOutCoords(); + float mean = getMeanAtOutCoords(); + float variance = getVarianceAtOutCoords(); + float offset = `+o+`; + float scale = `+l+`; + float inv = scale * inversesqrt(variance + float(`+s+`)); + setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1))); + } + `}return n}();var i8=function(){function n(t,e,i,r,a,s){this.packedInputs=!0,this.packedOutput=!0,this.variableNames=["x","mean","variance"],x.backend_util.assertAndGetBroadcastShape(t,e),x.backend_util.assertAndGetBroadcastShape(t,i);var o="vec4(0.0)";r!=null&&(x.backend_util.assertAndGetBroadcastShape(t,r),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");var l="vec4(1.0)";a!=null&&(x.backend_util.assertAndGetBroadcastShape(t,a),this.variableNames.push("scale"),l="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=` + void main() { + vec4 offset = `+o+`; + vec4 scale = `+l+`; + + vec4 x = getXAtOutCoords(); + vec4 mean = getMeanAtOutCoords(); + vec4 variance = getVarianceAtOutCoords(); + + vec4 inv = scale * inversesqrt(variance + vec4(`+s+`)); + + setOutput((x - mean) * inv + offset); + } + `}return n}();var r8=function(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.x,a=t.mean,s=t.variance,o=t.offset,l=t.scale;x.util.assert(a.shape.length===s.shape.length,function(){return"Batch normalization gradient requires mean and variance to have equal ranks."}),x.util.assert(o==null||a.shape.length===o.shape.length,function(){return"Batch normalization gradient requires mean and offset to have equal ranks."}),x.util.assert(l==null||a.shape.length===l.shape.length,function(){return"Batch normalization gradient requires mean and scale to have equal ranks."});var u=i.varianceEpsilon;u==null&&(u=.001);var c=[r,a,s],h=null;o!=null&&(h=o.shape,c.push(o));var d=null;l!=null&&(d=l.shape,c.push(l));var p=x.env().getBool("WEBGL_PACK_NORMALIZATION")?new i8(r.shape,a.shape,s.shape,h,d,u):new n8(r.shape,a.shape,s.shape,h,d,u),f=e.runWebGLProgram(p,c,c[0].dtype);return f},a8={kernelName:x.FusedBatchNorm,backendName:"webgl",kernelFunc:r8};var s8=xL+` + return cos(x); +`,o8=cl(s8),l8={kernelName:x.Cos,backendName:"webgl",kernelFunc:o8};var u8=` +if (a == b) { + return 1.0; +}; +return a / b;`,c8=` + // vec4 one = vec4(equal(a, b)); + // return one + (vec4(1.0) - one) * a / b; + vec4 result = a / b; + if(a.x == b.x) { + result.x = 1.; + } + if(a.y == b.y) { + result.y = 1.; + } + if(a.z == b.z) { + result.z = 1.; + } + if(a.w == b.w) { + result.w = 1.; + } + + return result; +`,h8=wp(u8,c8,!0),d8={kernelName:x.Div,backendName:"webgl",kernelFunc:h8};var p8=function(){function n(t){this.variableNames=["Image"],this.outputShape=[];var e=t[2];this.outputShape=t,this.userCode=` + void main() { + ivec4 coords = getOutputCoords(); + int x = coords[2]; + + int coordX = `+e+` - x; + float outputValue; + if(coordX >= 0 && coordX < `+e+`) { + outputValue = getImage(coords[0], coords[1], coordX, coords[3]); + } else { + outputValue = getImage(coords[0], coords[1], coords[2], coords[3]); + } + setOutput(outputValue); + } + `}return n}();var f8={kernelName:x.FlipLeftRight,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.backend,i=t.image,r=e,a=new p8(i.shape),s=r.runWebGLProgram(a,[i],i.dtype);return s}};var m8=function(){function n(t){this.variableNames=["A"];var e=Wt(),i=t[0],r=t[1];this.outputShape=t,this.userCode=` + void main() { + ivec3 coords = getOutputCoords(); + int texR = coords[0]; + int texC = coords[1]; + int depth = coords[2]; + vec2 uv = (vec2(texC, texR) + halfCR) / vec2(`+r+".0, "+i+`.0); + + vec4 values = `+e.texture2D+`(A, uv); + float value; + if (depth == 0) { + value = values.r; + } else if (depth == 1) { + value = values.g; + } else if (depth == 2) { + value = values.b; + } else if (depth == 3) { + value = values.a; + } + + setOutput(floor(value * 255.0 + 0.5)); + } + `}return n}();var g8=function(){function n(t){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0;var e=Wt(),i=t[0],r=t[1];this.outputShape=t,this.userCode=` + void main() { + ivec3 coords = getOutputCoords(); + int texR = coords[0]; + int texC = coords[1]; + int depth = coords[2]; + + vec4 result = vec4(0.); + + for(int row=0; row<=1; row++) { + for(int col=0; col<=1; col++) { + texC = coords[1] + row; + depth = coords[2] + col; + + vec2 uv = (vec2(texC, texR) + halfCR) / + vec2(`+r+".0, "+i+`.0); + vec4 values = `+e.texture2D+`(A, uv); + float value; + if (depth == 0) { + value = values.r; + } else if (depth == 1) { + value = values.g; + } else if (depth == 2) { + value = values.b; + } else if (depth == 3) { + value = values.a; + } + + result[row * 2 + col] = floor(value * 255.0 + 0.5); + } + } + + `+e.output+` = result; + } + `}return n}();var y8={kernelName:x.FromPixels,backendName:"webgl",kernelFunc:v8},ya;function v8(n){var t=n.inputs,e=n.backend,i=n.attrs,r=t.pixels,a=i.numChannels,s=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,o=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=s?[r.videoWidth,r.videoHeight]:[r.width,r.height],u=l[0],c=l[1],h=[c,u],d=[c,u,a];(o||s)&&(ya==null&&(ya=document.createElement("canvas").getContext("2d")),ya.canvas.width=u,ya.canvas.height=c,ya.drawImage(r,0,0,u,c),r=ya.canvas);var p=e.makeTensorInfo(h,"int32");e.texData.get(p.dataId).usage=an.PIXELS,e.gpgpu.uploadPixelDataToTexture(e.getTexture(p.dataId),r);var f=x.env().getBool("WEBGL_PACK")?new g8(d):new m8(d),m=e.runWebGLProgram(f,[p],"int32");return e.disposeData(p.dataId),m}function b8(n){for(var t=[];t.length===0||t[t.length-1].outSize!==1;){var e=t.length?t[t.length-1].outSize:n[1],i=x.backend_util.computeOptimalWindowSize(e);t.push({inSize:e,windowSize:i,outSize:Math.ceil(e/i)})}return t}function w8(n,t,e,i){for(var r=b8(n.shape),a=n,s=0;s6)throw Error("Transpose for rank "+t+" is not yet supported");for(var e=["resRC.x","resRC.y","resRC.z","resRC.w","resRC.u","resRC.v"],i=new Array(t),r=0;r6)throw Error("Packed transpose for rank "+this.rank+" is not yet supported.");for(var a=Je(this.rank),s=LS("rc",this.rank),o=new Array(this.rank),r=0;r= 0 && coordX < `+s+" && coordY >= 0 && coordY < "+a+`) { + outputValue = getImage(coords[0], coordY, coordX, coords[3]); + } + setOutput(outputValue); + } + `}return n}();var _8={kernelName:x.RotateWithOffset,backendName:"webgl",kernelFunc:function(n){var t=n.inputs,e=n.attrs,i=n.backend,r=t.image,a=e,s=a.radians,o=a.fillValue,l=a.center,u=i,c=new P8(r.shape,s,o,l),h=u.runWebGLProgram(c,[r],r.dtype);return h}};var M8=xL+` + return sin(x); +`,H8=cl(M8),V8={kernelName:x.Sin,backendName:"webgl",kernelFunc:H8};var q8="return x * x;",G8=cl(q8),Y8={kernelName:x.Square,backendName:"webgl",kernelFunc:G8};var RL="return (a - b) * (a - b);",K8=wp(RL,RL),j8={kernelName:x.SquaredDifference,backendName:"webgl",kernelFunc:K8};var $8="return tan(x);",X8=cl($8),J8={kernelName:x.Tan,backendName:"webgl",kernelFunc:X8};var Z8={kernelName:x.Transpose,backendName:"webgl",kernelFunc:function(n){for(var t=n.inputs,e=n.attrs,i=n.backend,r=t.x,a=e.perm,s=i,o=r.shape.length,l=new Array(o),u=0;u{"use strict";Object.defineProperty(Lr,"__esModule",{value:!0});var Ap=rr(),Tp=xw(),Np=Hw(),DL=l0(),i7=$0(),r7=EL();var a7="2.6.0";var s7={"tfjs-core":Ap.version_core,"tfjs-backend-cpu":i7.version_cpu,"tfjs-backend-webgl":r7.version_webgl,"tfjs-data":DL.version_data,"tfjs-layers":Tp.version_layers,"tfjs-converter":Np.version_converter,tfjs:a7};Object.keys(Ap).forEach(function(n){n!=="default"&&Object.defineProperty(Lr,n,{enumerable:!0,get:function(){return Ap[n]}})});Object.keys(Tp).forEach(function(n){n!=="default"&&Object.defineProperty(Lr,n,{enumerable:!0,get:function(){return Tp[n]}})});Object.keys(Np).forEach(function(n){n!=="default"&&Object.defineProperty(Lr,n,{enumerable:!0,get:function(){return Np[n]}})});Lr.data=DL;Lr.version=s7});var zL=Ae(hl=>{const Be=Ut(),o7={strides:[8,16],anchors:[2,6]},kL=6;function l7(n,t,e){const i=[];for(let r=0;r{n.startEndTensor.dispose(),n.startPoint.dispose(),n.endPoint.dispose()},WL=n=>({startEndTensor:n,startPoint:Be.slice(n,[0,0],[-1,2]),endPoint:Be.slice(n,[0,2],[-1,2])}),u7=(n,t)=>{const e=Be.mul(n.startPoint,t),i=Be.mul(n.endPoint,t),r=Be.concat2d([e,i],1);return WL(r)};function c7(n,t,e){const i=Be.slice(n,[0,1],[-1,2]),r=Be.add(i,t),a=Be.slice(n,[0,3],[-1,2]),s=Be.div(a,e),o=Be.div(r,e),l=Be.div(s,2),u=Be.sub(o,l),c=Be.add(o,l),h=Be.mul(u,e),d=Be.mul(c,e),p=1;return Be.concat2d([h,d],p)}function UL(n,t){return Be.tidy(()=>{const e=n.box?n.box:n;return u7(e,t).startEndTensor.squeeze()})}class BL{constructor(n,t){this.blazeFaceModel=n,this.width=t.detector.inputSize,this.height=t.detector.inputSize,this.maxFaces=t.detector.maxFaces,this.anchorsData=l7(t.detector.inputSize,t.detector.inputSize,o7),this.anchors=Be.tensor2d(this.anchorsData),this.inputSizeData=[t.detector.inputSize,t.detector.inputSize],this.inputSize=Be.tensor1d([t.detector.inputSize,t.detector.inputSize]),this.iouThreshold=t.detector.iouThreshold,this.scoreThreshold=t.detector.scoreThreshold}async getBoundingBoxes(n){const[t,e,i]=Be.tidy(()=>{const l=n.resizeBilinear([this.width,this.height]),u=Be.mul(Be.sub(l.div(255),.5),2),c=this.blazeFaceModel.predict(u),h=c.squeeze(),d=c7(h,this.anchors,this.inputSize),p=Be.slice(h,[0,0],[-1,1]),f=Be.sigmoid(p).squeeze();return[h,d,f]}),r=await Be.image.nonMaxSuppressionAsync(e,i,this.maxFaces,this.iouThreshold,this.scoreThreshold),a=await r.array();r.dispose();let s=a.map(l=>Be.slice(e,[l,0],[1,-1]));s=await Promise.all(s.map(async l=>{const u=await l.array();return l.dispose(),u}));const o=[];for(let l=0;l{const h=WL(u),d=a[l],p=this.anchorsData[d],f=Be.slice(t,[d,kL-1],[1,-1]).squeeze().reshape([kL,-1]),m=Be.slice(i,[d],[1]);return{box:h,landmarks:f,probability:m,anchor:p}});o.push(c)}return e.dispose(),i.dispose(),t.dispose(),{boxes:o,scaleFactor:[n.shape[2]/this.inputSizeData[0],n.shape[1]/this.inputSizeData[1]]}}async estimateFaces(n,t=!1,e=!0){const i=Be.tidy(()=>(n instanceof Be.Tensor||(n=Be.browser.fromPixels(n)),n.toFloat().expandDims(0))),{boxes:r,scaleFactor:a}=await this.getBoundingBoxes(i);return i.dispose(),t?r.map(s=>{const o=UL(s,a),l={topLeft:o.slice([0],[2]),bottomRight:o.slice([2],[2])};if(e){const{landmarks:u,probability:c,anchor:h}=s,d=u.add(h).mul(a);l.landmarks=d,l.probability=c}return l}):Promise.all(r.map(async s=>{const o=UL(s,a);let l;if(e){const[u,c,h]=await Promise.all([s.landmarks,o,s.probability].map(async g=>g.array())),d=s.anchor,[p,f]=a,m=u.map(g=>[(g[0]+d[0])*p,(g[1]+d[1])*f]);l={topLeft:c.slice(0,2),bottomRight:c.slice(2),landmarks:m,probability:h},FL(s.box),s.landmarks.dispose(),s.probability.dispose()}else{const u=await o.array();l={topLeft:u.slice(0,2),bottomRight:u.slice(2)}}return o.dispose(),l}))}}async function h7(n){const t=await Be.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),e=new BL(t,n);return e}hl.load=h7;hl.BlazeFaceModel=BL;hl.disposeBox=FL});var Cp=Ae(xp=>{xp.MESH_ANNOTATIONS={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]};xp.MESH_TO_IRIS_INDICES_MAP=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]},{key:"EyebrowUpper",indices:[63,64,65,66,67,68,69,70]},{key:"EyebrowLower",indices:[48,49,50,51,52,53]}]});var PL=Ae(Ir=>{const d7=Ut();function p7(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]];return{startPoint:e,endPoint:i}}Ir.scaleBoxCoordinates=p7;function Rp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}Ir.getBoxSize=Rp;function Op(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}Ir.getBoxCenter=Op;function f7(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return d7.image.cropAndResize(t,a,[0],e)}Ir.cutBoxFromImageAndResize=f7;function m7(n,t=1.5){const e=Op(n),i=Rp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}Ir.enlargeBox=m7;function g7(n){const t=Op(n),e=Rp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,landmarks:n.landmarks}}Ir.squarifyBox=g7});var qL=Ae(Nn=>{Nn.IDENTITY_MATRIX=[[1,0,0],[0,1,0],[0,0,1]];function _L(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Nn.normalizeRadians=_L;function v7(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return _L(e)}Nn.computeRotation=v7;function y7(n){return n*180/Math.PI}Nn.radToDegrees=y7;function ML(n,t){return[[1,0,n],[0,1,t],[0,0,1]]}function ba(n,t){let e=0;for(let i=0;i{const Fi=Ut(),Vn=PL(),Wi=Cp(),Ui=qL(),I7=468,A7=.25,T7=13,N7=[T7,Wi.MESH_ANNOTATIONS.midwayBetweenEyes[0]],x7=3,C7=2,R7=[x7,C7],Ep=Wi.MESH_ANNOTATIONS.leftEyeLower0,Dp=[Ep[0],Ep[Ep.length-1]],kp=Wi.MESH_ANNOTATIONS.rightEyeLower0,Fp=[kp[0],kp[kp.length-1]],O7=3,E7=4,D7=71,Wp=76;function dl(n,t,e,i){for(let r=0;r[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=Ui.buildRotationMatrix(e,[0,0]),l=s.map(d=>[...Ui.rotatePoint(d,o),d[2]]),u=Ui.invertTransformMatrix(i),c=[...Vn.getBoxCenter({startPoint:t.startPoint,endPoint:t.endPoint}),1],h=[Ui.dot(c,u[0]),Ui.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}getLeftToRightEyeDepthDifference(n){const t=n[Dp[0]][2],e=n[Fp[0]][2];return t-e}getEyeBox(n,t,e,i,r=!1){const a=Vn.squarifyBox(Vn.enlargeBox(this.calculateLandmarksBoundingBox([n[e],n[i]]),this.irisEnlarge)),s=Vn.getBoxSize(a);let o=Fi.image.cropAndResize(t,[[a.startPoint[1]/this.meshHeight,a.startPoint[0]/this.meshWidth,a.endPoint[1]/this.meshHeight,a.endPoint[0]/this.meshWidth]],[0],[this.irisSize,this.irisSize]);return r&&(o=Fi.image.flipLeftRight(o)),{box:a,boxSize:s,crop:o}}getEyeCoords(n,t,e,i=!1){const r=[];for(let a=0;a{let l=a;return o===2?l=i:o===4&&(l=r),[s[0],s[1],l]})}async predict(n,t){if(this.skipFrames=t.detector.skipFrames,this.maxFaces=t.detector.maxFaces,this.shouldUpdateRegionsOfInterest()){const{boxes:i,scaleFactor:r}=await this.boundingBoxDetector.getBoundingBoxes(n);if(i.length===0)return this.regionsOfInterest=[],null;const a=i.map(s=>{const o={startPoint:s.box.startPoint.squeeze().arraySync(),endPoint:s.box.endPoint.squeeze().arraySync()};s.box.startPoint.dispose(),s.box.endPoint.dispose();const l=Vn.scaleBoxCoordinates(o,r),u=Vn.enlargeBox(l),c=s.landmarks.arraySync();return s.landmarks.dispose(),s.probability.dispose(),{...u,landmarks:c}});this.updateRegionsOfInterest(a),this.runsWithoutFaceDetector=0}else this.runsWithoutFaceDetector++;const e=Fi.tidy(()=>this.regionsOfInterest.map((i,r)=>{let a=0;const s=i.landmarks.length>=I7;let[o,l]=N7;s===!1&&([o,l]=R7),a=Ui.computeRotation(i.landmarks[o],i.landmarks[l]);const u=Vn.getBoxCenter({startPoint:i.startPoint,endPoint:i.endPoint}),c=[u[0]/n.shape[2],u[1]/n.shape[1]];let h=n,d=Ui.IDENTITY_MATRIX;a!==0&&(h=Fi.image.rotateWithOffset(n,a,0,c),d=Ui.buildRotationMatrix(-a,u));const p={startPoint:i.startPoint,endPoint:i.endPoint},f=Vn.cutBoxFromImageAndResize(p,h,[this.meshHeight,this.meshWidth]).div(255),[,m,g]=this.meshDetector.predict(f),v=Fi.reshape(g,[-1,3]);let b=v.arraySync();if(t.iris.enabled){const{box:N,boxSize:C,crop:R}=this.getEyeBox(b,f,Dp[0],Dp[1],!0),{box:D,boxSize:k,crop:W}=this.getEyeBox(b,f,Fp[0],Fp[1]),F=this.irisModel.predict(Fi.concat([R,W])),P=F.dataSync();F.dispose();const H=P.slice(0,Wp*3),{rawCoords:_,iris:K}=this.getEyeCoords(H,N,C,!0),j=P.slice(Wp*3),{rawCoords:q,iris:G}=this.getEyeCoords(j,D,k),Z=this.getLeftToRightEyeDepthDifference(b);Math.abs(Z)<30?(dl(b,_,"left"),dl(b,q,"right")):Z<1?dl(b,_,"left",["EyeUpper0","EyeLower0"]):dl(b,q,"right",["EyeUpper0","EyeLower0"]);const X=this.getAdjustedIrisCoords(b,K,"left"),ee=this.getAdjustedIrisCoords(b,G,"right");b=b.concat(X).concat(ee)}const w=this.transformRawCoords(b,i,a,d);Fi.dispose(b);const S=Vn.enlargeBox(this.calculateLandmarksBoundingBox(w));if(t.mesh.enabled){const N=Fi.tensor2d(w);this.regionsOfInterest[r]={...S,landmarks:N.arraySync()};const C={coords:N,box:S,confidence:m.squeeze(),image:f};return C}const L={coords:null,box:S,confidence:m.squeeze(),image:f};return L}));return e}updateRegionsOfInterest(n){for(let t=0;t=this.skipFrames}calculateLandmarksBoundingBox(n){const t=n.map(a=>a[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}}GL.Pipeline=k7});var jL=Ae(KL=>{KL.UV_COORDS=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]]});var $L=Ae(F7=>{vl(F7,{default:()=>W7});var W7=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255]});var ZL=Ae(As=>{const Ar=Ut(),U7=zL(),XL=Cp(),B7=YL(),z7=jL(),P7=$L().default;class JL{constructor(n,t,e,i){this.pipeline=new B7.Pipeline(n,t,e,i),i&&(this.config=i)}async estimateFaces(n,t){t&&(this.config=t);const e=Ar.tidy(()=>(n instanceof Ar.Tensor||(n=Ar.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.predict(e,t);Ar.dispose(e);const r=[];for(const a of i||[]){const s=a.confidence.arraySync();if(s>=this.config.detector.minConfidence){const o=a.coords?a.coords.arraySync():null,l={};if(o&&o.length>0)for(const u in XL.MESH_ANNOTATIONS)(this.config.iris.enabled||u.includes("Iris")===!1)&&(l[u]=XL.MESH_ANNOTATIONS[u].map(c=>o[c]));r.push({confidence:s||0,box:a.box?[a.box.startPoint[0],a.box.startPoint[1],a.box.endPoint[0]-a.box.startPoint[0],a.box.endPoint[1]-a.box.startPoint[1]]:0,mesh:o,annotations:l,image:a.image?Ar.clone(a.image):null})}a.confidence.dispose(),a.image.dispose()}return r}}async function _7(n){const t=await Promise.all([U7.load(n),Ar.loadGraphModel(n.mesh.modelPath,{fromTFHub:n.mesh.modelPath.includes("tfhub.dev")}),Ar.loadGraphModel(n.iris.modelPath,{fromTFHub:n.iris.modelPath.includes("tfhub.dev")})]),e=new JL(t[0],t[1],t[2],n);return e}As.load=_7;As.MediaPipeFaceMesh=JL;As.uv_coords=z7;As.triangulation=P7});var eI=Ae(pl=>{const Zt=Ut(),wa={};let QL={age:0,gender:""},Up=0;async function M7(n,t){const e=Zt.tidy(()=>{const i=Zt.browser.fromPixels(n),r=Zt.image.resizeBilinear(i,[t,t]),a=Zt.cast(Zt.expandDims(r,0),"float32");return a});return e}async function H7(n){wa.age||(wa.age=await Zt.loadGraphModel(n.face.age.modelPath))}async function V7(n){wa.gender||(wa.gender=await Zt.loadGraphModel(n.face.gender.modelPath))}async function q7(n,t){if(Up+=1,Up>=t.face.age.skipFrames)return Up=0,QL;let e;if(n instanceof Zt.Tensor){const r=Zt.image.resizeBilinear(n,[t.face.age.inputSize,t.face.age.inputSize],!1);e=Zt.mul(r,[255]),Zt.dispose(r)}else e=await M7(n,t.face.age.inputSize);const i={};if(t.face.age.enabled){const r=await wa.age.predict(e),a=await r.data();i.age=Math.trunc(10*a[0])/10,Zt.dispose(r)}if(t.face.gender.enabled){const r=await wa.gender.predict(e),a=await r.data();i.gender=Math.trunc(100*a[0])<50?"female":"male",Zt.dispose(r)}return Zt.dispose(e),QL=i,i}pl.predict=q7;pl.loadAge=H7;pl.loadGender=V7});var iI=Ae(Bp=>{const Bt=Ut(),G7=["angry","discust","fear","happy","sad","surpise","neutral"],zp={};let tI=[],Pp=0;const nI=1.5;function Y7(n,t){const e=Bt.tidy(()=>{const i=Bt.browser.fromPixels(n,1),r=Bt.image.resizeBilinear(i,[t,t]),a=Bt.cast(Bt.expandDims(r,0),"float32");return a});return e}async function K7(n){zp.emotion||(zp.emotion=await Bt.loadGraphModel(n.face.emotion.modelPath))}async function j7(n,t){if(Pp+=1,Pp>=t.face.emotion.skipFrames)return Pp=0,tI;const e=Bt.tidy(()=>{if(n instanceof Bt.Tensor){const r=Bt.image.resizeBilinear(n,[t.face.emotion.inputSize,t.face.emotion.inputSize],!1),[a,s,o]=Bt.split(r,3,3);if(t.face.emotion.useGrayscale){const l=Bt.mul(a,[.2989]),u=Bt.mul(s,[.587]),c=Bt.mul(o,[.114]),h=Bt.addN([l,u,c]);return h}return s}return Y7(n,t.face.emotion.inputSize)}),i=[];if(t.face.emotion.enabled){const r=await zp.emotion.predict(e),a=await r.data();for(let s=0;st.face.emotion.minConfidence&&i.push({score:Math.min(.99,Math.trunc(100*nI*a[s])/100),emotion:G7[s]});i.sort((s,o)=>o.score-s.score),Bt.dispose(r)}return Bt.dispose(e),tI=i,i}Bp.predict=j7;Bp.load=K7});var sI=Ae(rI=>{const aI=Ut();class $7{constructor(n,t){this.model=n,this.outputStride=t;const e=this.model.inputs[0].shape;aI.util.assert(e[1]===-1&&e[2]===-1,()=>`Input shape [${e[1]}, ${e[2]}] must both be equal to or -1`)}predict(n){return aI.tidy(()=>{const t=this.preprocessInput(n.toFloat()),e=t.expandDims(0),i=this.model.predict(e),r=i.map(s=>s.squeeze([0])),a=this.nameOutputResults(r);return{heatmapScores:a.heatmap.sigmoid(),offsets:a.offsets,displacementFwd:a.displacementFwd,displacementBwd:a.displacementBwd}})}dispose(){this.model.dispose()}}rI.BaseModel=$7});var _p=Ae(oI=>{const lI=Ut(),X7=sI();class J7 extends X7.BaseModel{preprocessInput(n){return lI.tidy(()=>lI.div(n,127.5).sub(1))}nameOutputResults(n){const[t,e,i,r]=n;return{offsets:t,heatmap:e,displacementFwd:i,displacementBwd:r}}}oI.MobileNet=J7});var cI=Ae(uI=>{function Mp(n){return Math.floor(n/2)}class Z7{constructor(n,t){this.priorityQueue=new Array(n),this.numberOfElements=-1,this.getElementValue=t}enqueue(n){this.priorityQueue[++this.numberOfElements]=n,this.swim(this.numberOfElements)}dequeue(){const n=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,n}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(n){for(;n>0&&this.less(Mp(n),n);)this.exchange(n,Mp(n)),n=Mp(n)}sink(n){for(;2*n<=this.numberOfElements;){let t=2*n;if(t{const Q7=cI();function eH(n,t,e,i,r,a){const[s,o]=a.shape;let l=!0;const u=Math.max(e-r,0),c=Math.min(e+r+1,s);for(let h=u;ht){l=!1;break}if(!l)break}return l}function tH(n,t,e){const[i,r,a]=e.shape,s=new Q7.MaxHeap(i*r*a,({score:o})=>o);for(let o=0;o{xn.partNames=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"];xn.NUM_KEYPOINTS=xn.partNames.length;xn.partIds=xn.partNames.reduce((n,t,e)=>(n[t]=e,n),{});const nH=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]];xn.poseChain=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];xn.connectedPartIndices=nH.map(([n,t])=>[xn.partIds[n],xn.partIds[t]]);xn.partChannels=["left_face","right_face","right_upper_leg_front","right_lower_leg_back","right_upper_leg_back","left_lower_leg_front","left_upper_leg_front","left_upper_leg_back","left_lower_leg_back","right_feet","right_lower_leg_front","left_feet","torso_front","torso_back","right_upper_arm_front","right_upper_arm_back","right_lower_arm_back","left_lower_arm_front","left_upper_arm_front","left_upper_arm_back","left_lower_arm_back","right_hand","right_lower_arm_front","left_hand"]});var Vp=Ae(Bi=>{const iH=Ts();function pI(n,t,e,i){return{y:i.get(n,t,e),x:i.get(n,t,e+iH.NUM_KEYPOINTS)}}Bi.getOffsetPoint=pI;function rH(n,t,e){const{heatmapY:i,heatmapX:r,id:a}=n,{y:s,x:o}=pI(i,r,a,e);return{x:n.heatmapX*t+o,y:n.heatmapY*t+s}}Bi.getImageCoords=rH;function aH(n,t){const e=new Array(t);for(let i=0;ie?e:n}Bi.clamp=Hp;function sH(n,t,e,i){const r=e-n,a=i-t;return r*r+a*a}Bi.squaredDistance=sH;function oH(n,t){return{x:n.x+t.x,y:n.y+t.y}}Bi.addVectors=oH;function lH(n,t,e){return{y:Hp(n.y,t,e),x:Hp(n.x,t,e)}}Bi.clampVector=lH});var yI=Ae(fI=>{const Ns=Ts(),Sa=Vp(),mI=Ns.poseChain.map(([n,t])=>[Ns.partIds[n],Ns.partIds[t]]),qp=mI.map(([,n])=>n),gI=mI.map(([n])=>n);function uH(n,t,e){const i=e.shape[2]/2;return{y:e.get(t.y,t.x,n),x:e.get(t.y,t.x,i+n)}}function Gp(n,t,e,i){return{y:Sa.clamp(Math.round(n.y/t),0,e-1),x:Sa.clamp(Math.round(n.x/t),0,i-1)}}function vI(n,t,e,i,r,a,s,o=2){const[l,u]=i.shape,c=Gp(t.position,a,l,u),h=uH(n,c,s),d=Sa.addVectors(t.position,h);let p=d;for(let g=0;g=0;--d){const p=qp[d],f=gI[d];l[p]&&!l[f]&&(l[f]=vI(d,l[p],f,t,e,i,a))}for(let d=0;d{const hH=dI(),dH=yI(),wI=Vp();function SI(n,t,{x:e,y:i},r){return n.some(({keypoints:a})=>{const s=a[r].position;return wI.squaredDistance(i,e,s.y,s.x)<=t})}function pH(n,t,e){const i=e.reduce((r,{position:a,score:s},o)=>(SI(n,t,a,o)||(r+=s),r),0);return i/e.length}const fH=1;function mH(n,t,e,i,r,a,s=.5,o=20){const l=[],u=hH.buildPartWithScoreQueue(s,fH,n),c=o*o;for(;l.length{const La=Ut(),gH=Ts();function vH(n,t,e){return n(vH(n[i].score,n[r].score,t)||e.push([n[i],n[r]]),e),[])}pn.getAdjacentKeyPoints=yH;const{NEGATIVE_INFINITY:LI,POSITIVE_INFINITY:II}=Number;function AI(n){return n.reduce(({maxX:t,maxY:e,minX:i,minY:r},{position:{x:a,y:s}})=>({maxX:Math.max(t,a),maxY:Math.max(e,s),minX:Math.min(i,a),minY:Math.min(r,s)}),{maxX:LI,maxY:LI,minX:II,minY:II})}pn.getBoundingBox=AI;function bH(n){const{minX:t,minY:e,maxX:i,maxY:r}=AI(n);return[{x:t,y:e},{x:i,y:e},{x:i,y:r},{x:t,y:r}]}pn.getBoundingBoxPoints=bH;async function wH(n){return Promise.all(n.map(t=>t.buffer()))}pn.toTensorBuffers3D=wH;function TI(n,t,e,i=0,r=0){return{score:n.score,keypoints:n.keypoints.map(({score:a,part:s,position:o})=>({score:a,part:s,position:{x:o.x*e+r,y:o.y*t+i}}))}}pn.scalePose=TI;function NI(n,t,e,i=0,r=0){return e===1&&t===1&&i===0&&r===0?n:n.map(a=>TI(a,t,e,i,r))}pn.scalePoses=NI;function xI(n){return n instanceof La.Tensor?[n.shape[0],n.shape[1]]:[n.height,n.width]}pn.getInputTensorDimensions=xI;function Kp(n){return n instanceof La.Tensor?n:La.browser.fromPixels(n)}pn.toInputTensor=Kp;function SH(n,t,e){return La.tidy(()=>{const i=Kp(n);return i.resizeBilinear([t,e])})}pn.toResizedInputTensor=SH;function LH(n,[t,e]){const[i,r]=xI(n),a=e/t,s=r/i;let[o,l,u,c]=[0,0,0,0];s{let d=Kp(n);return d=La.pad3d(d,[[o,l],[u,c],[0,0]]),d.resizeBilinear([t,e])});return{resized:h,padding:{top:o,left:u,right:c,bottom:l}}}pn.padAndResizeTo=LH;function IH(n,[t,e],[i,r],a){const s=(t+a.top+a.bottom)/i,o=(e+a.left+a.right)/r,l=NI(n,s,o,-a.top,-a.left);return l}pn.scaleAndFlipPoses=IH});var RI=Ae($p=>{const AH=Ut(),TH=_p(),NH=Yp(),fl=jp();class CI{constructor(n){this.baseModel=n}async estimatePoses(n,t){const e=t.outputStride,[i,r]=fl.getInputTensorDimensions(n),{resized:a,padding:s}=fl.padAndResizeTo(n,[t.inputResolution,t.inputResolution]),{heatmapScores:o,offsets:l,displacementFwd:u,displacementBwd:c}=this.baseModel.predict(a),h=await fl.toTensorBuffers3D([o,l,u,c]),d=h[0],p=h[1],f=h[2],m=h[3],g=await NH.decodeMultiplePoses(d,p,f,m,e,t.maxDetections,t.scoreThreshold,t.nmsRadius),v=fl.scaleAndFlipPoses(g,[i,r],[t.inputResolution,t.inputResolution],s);return o.dispose(),l.dispose(),u.dispose(),c.dispose(),a.dispose(),v}dispose(){this.baseModel.dispose()}}$p.PoseNet=CI;async function xH(n){const t=await AH.loadGraphModel(n.modelPath),e=new TH.MobileNet(t,n.outputStride);return new CI(e)}async function CH(n){return xH(n)}$p.load=CH});var EI=Ae(Qt=>{const RH=_p(),OI=RI(),OH=Yp(),ml=Ts(),xs=jp();Qt.load=OI.load;Qt.PoseNet=OI.PoseNet;Qt.MobileNet=RH.MobileNet;Qt.decodeMultiplePoses=OH.decodeMultiplePoses;Qt.partChannels=ml.partChannels;Qt.partIds=ml.partIds;Qt.partNames=ml.partNames;Qt.poseChain=ml.poseChain;Qt.getAdjacentKeyPoints=xs.getAdjacentKeyPoints;Qt.getBoundingBox=xs.getBoundingBox;Qt.getBoundingBoxPoints=xs.getBoundingBoxPoints;Qt.scaleAndFlipPoses=xs.scaleAndFlipPoses;Qt.scalePose=xs.scalePose});var Zp=Ae(zi=>{const EH=Ut();function Xp(n){return[Math.abs(n.endPoint[0]-n.startPoint[0]),Math.abs(n.endPoint[1]-n.startPoint[1])]}zi.getBoxSize=Xp;function Jp(n){return[n.startPoint[0]+(n.endPoint[0]-n.startPoint[0])/2,n.startPoint[1]+(n.endPoint[1]-n.startPoint[1])/2]}zi.getBoxCenter=Jp;function DH(n,t,e){const i=t.shape[1],r=t.shape[2],a=[[n.startPoint[1]/i,n.startPoint[0]/r,n.endPoint[1]/i,n.endPoint[0]/r]];return EH.image.cropAndResize(t,a,[0],e)}zi.cutBoxFromImageAndResize=DH;function kH(n,t){const e=[n.startPoint[0]*t[0],n.startPoint[1]*t[1]],i=[n.endPoint[0]*t[0],n.endPoint[1]*t[1]],r=n.palmLandmarks.map(a=>{const s=[a[0]*t[0],a[1]*t[1]];return s});return{startPoint:e,endPoint:i,palmLandmarks:r}}zi.scaleBoxCoordinates=kH;function FH(n,t=1.5){const e=Jp(n),i=Xp(n),r=[t*i[0]/2,t*i[1]/2],a=[e[0]-r[0],e[1]-r[1]],s=[e[0]+r[0],e[1]+r[1]];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}zi.enlargeBox=FH;function WH(n){const t=Jp(n),e=Xp(n),i=Math.max(...e),r=i/2,a=[t[0]-r,t[1]-r],s=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:s,palmLandmarks:n.palmLandmarks}}zi.squarifyBox=WH;function UH(n,t){const e=[n.endPoint[0]-n.startPoint[0],n.endPoint[1]-n.startPoint[1]],i=[e[0]*t[0],e[1]*t[1]],r=[n.startPoint[0]+i[0],n.startPoint[1]+i[1]],a=[n.endPoint[0]+i[0],n.endPoint[1]+i[1]];return{startPoint:r,endPoint:a,palmLandmarks:n.palmLandmarks}}zi.shiftBox=UH});var kI=Ae(DI=>{const Ve=Ut(),BH=Zp();class zH{constructor(n,t,e){this.model=n,this.width=e.inputSize,this.height=e.inputSize,this.anchors=t.map(i=>[i.x_center,i.y_center]),this.anchorsTensor=Ve.tensor2d(this.anchors),this.inputSizeTensor=Ve.tensor1d([e.inputSize,e.inputSize]),this.doubleInputSizeTensor=Ve.tensor1d([e.inputSize*2,e.inputSize*2])}normalizeBoxes(n){return Ve.tidy(()=>{const t=Ve.slice(n,[0,0],[-1,2]),e=Ve.slice(n,[0,2],[-1,2]),i=Ve.add(Ve.div(t,this.inputSizeTensor),this.anchorsTensor),r=Ve.div(e,this.doubleInputSizeTensor),a=Ve.mul(Ve.sub(i,r),this.inputSizeTensor),s=Ve.mul(Ve.add(i,r),this.inputSizeTensor);return Ve.concat2d([a,s],1)})}normalizeLandmarks(n,t){return Ve.tidy(()=>{const e=Ve.add(Ve.div(n.reshape([-1,7,2]),this.inputSizeTensor),this.anchors[t]);return Ve.mul(e,this.inputSizeTensor)})}async getBoundingBoxes(n){const t=Ve.tidy(()=>Ve.mul(Ve.sub(n,.5),2)),e=this.model.predict(t),i=e.squeeze(),r=Ve.tidy(()=>Ve.sigmoid(Ve.slice(i,[0,0],[-1,1])).squeeze()),a=Ve.slice(i,[0,1],[-1,4]),s=this.normalizeBoxes(a),o=await Ve.image.nonMaxSuppressionAsync(s,r,this.maxHands,this.iouThreshold,this.scoreThreshold),l=await o.array(),u=[t,e,o,i,s,a,r];if(l.length===0)return u.forEach(h=>h.dispose()),null;const c=Ve.tidy(()=>{const h=[];for(const d in l){const p=l[d],f=Ve.slice(s,[p,0],[1,-1]),m=Ve.slice(i,[p,5],[1,14]),g=Ve.tidy(()=>this.normalizeLandmarks(m,p).reshape([-1,2]));h.push({boxes:f,palmLandmarks:g})}return h});return c}async estimateHandBounds(n,t){const e=n.shape[1],i=n.shape[2];this.iouThreshold=t.iouThreshold,this.scoreThreshold=t.scoreThreshold,this.maxHands=t.maxHands;const r=Ve.tidy(()=>n.resizeBilinear([this.width,this.height]).div(255)),a=await this.getBoundingBoxes(r);if(r.dispose(),!a||a.length===0)return null;const s=[];for(const o in a){const l=a[o],u=await l.boxes.array(),c=u[0].slice(0,2),h=u[0].slice(2,4),d=await l.palmLandmarks.array();l.boxes.dispose(),l.palmLandmarks.dispose(),s.push(BH.scaleBoxCoordinates({startPoint:c,endPoint:h,palmLandmarks:d},[i/this.width,e/this.height]))}return s}}DI.HandDetector=zH});var WI=Ae(FI=>{FI.MESH_ANNOTATIONS={thumb:[1,2,3,4],indexFinger:[5,6,7,8],middleFinger:[9,10,11,12],ringFinger:[13,14,15,16],pinky:[17,18,19,20],palmBase:[0]}});var _I=Ae(Pi=>{function UI(n){return n-2*Math.PI*Math.floor((n+Math.PI)/(2*Math.PI))}Pi.normalizeRadians=UI;function PH(n,t){const e=Math.PI/2-Math.atan2(-(t[1]-n[1]),t[0]-n[0]);return UI(e)}Pi.computeRotation=PH;const BI=(n,t)=>[[1,0,n],[0,1,t],[0,0,1]];function Ia(n,t){let e=0;for(let i=0;i{const HI=Ut(),qn=Zp(),_i=_I(),VH=.8,qH=[0,-.4],GH=[0,-.1],YH=1.65,VI=[0,5,9,13,17,1,2],KH=0,jH=2;class $H{constructor(n,t,e){this.regionsOfInterest=[],this.runsWithoutHandDetector=0,this.boundingBoxDetector=n,this.meshDetector=t,this.meshWidth=e.inputSize,this.meshHeight=e.inputSize,this.enlargeFactor=e.enlargeFactor}getBoxForPalmLandmarks(n,t){const e=n.map(r=>{const a=[...r,1];return _i.rotatePoint(a,t)}),i=this.calculateLandmarksBoundingBox(e);return qn.enlargeBox(qn.squarifyBox(qn.shiftBox(i,qH)),this.enlargeFactor)}getBoxForHandLandmarks(n){const t=this.calculateLandmarksBoundingBox(n),e=qn.enlargeBox(qn.squarifyBox(qn.shiftBox(t,GH)),YH),i=[];for(let r=0;r[a[0]*(d[0]-this.meshWidth/2),a[1]*(d[1]-this.meshHeight/2),d[2]]),o=_i.buildRotationMatrix(e,[0,0]),l=s.map(d=>{const p=_i.rotatePoint(d,o);return[...p,d[2]]}),u=_i.invertTransformMatrix(i),c=[...qn.getBoxCenter(t),1],h=[_i.dot(c,u[0]),_i.dot(c,u[1])];return l.map(d=>[d[0]+h[0],d[1]+h[1],d[2]])}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=this.shouldUpdateRegionsOfInterest();if(e===!0){const r=await this.boundingBoxDetector.estimateHandBounds(n,t);this.regionsOfInterest=[];for(const a in r)this.updateRegionsOfInterest(r[a],!0,a);this.runsWithoutHandDetector=0}else this.runsWithoutHandDetector++;const i=[];if(!this.regionsOfInterest)return i;for(const r in this.regionsOfInterest){const a=this.regionsOfInterest[r][0];if(!a)return i;const s=_i.computeRotation(a.palmLandmarks[KH],a.palmLandmarks[jH]),o=qn.getBoxCenter(a),l=[o[0]/n.shape[2],o[1]/n.shape[1]],u=HI.image.rotateWithOffset(n,s,0,l),c=_i.buildRotationMatrix(-s,o),h=e?this.getBoxForPalmLandmarks(a.palmLandmarks,c):a,d=qn.cutBoxFromImageAndResize(h,u,[this.meshWidth,this.meshHeight]),p=d.div(255);d.dispose(),u.dispose();const f=this.meshDetector.predict(p),[m,g]=f;p.dispose();const v=m.dataSync()[0];if(m.dispose(),va[0]),e=n.map(a=>a[1]),i=[Math.min(...t),Math.min(...e)],r=[Math.max(...t),Math.max(...e)];return{startPoint:i,endPoint:r}}updateRegionsOfInterest(n,t,e){if(t)this.regionsOfInterest[e]=[n];else{const i=this.regionsOfInterest[e][0];let r=0;if(i!=null&&i.startPoint!=null){const[a,s]=n.startPoint,[o,l]=n.endPoint,[u,c]=i.startPoint,[h,d]=i.endPoint,p=Math.max(a,u),f=Math.max(s,c),m=Math.min(o,h),g=Math.min(l,d),v=(m-p)*(g-f),b=(o-a)*(l-s),w=(h-u)*(d-s);r=v/(b+w-v)}this.regionsOfInterest[e][0]=r>VH?i:n}}shouldUpdateRegionsOfInterest(){return!this.regionsOfInterest||this.regionsOfInterest.length===0||this.runsWithoutHandDetector>=this.maxContinuousChecks}}MI.HandPipeline=$H});var KI=Ae(Qp=>{const Tr=Ut(),XH=kI(),GI=WI(),JH=qI();class YI{constructor(n){this.pipeline=n}async estimateHands(n,t){this.maxContinuousChecks=t.skipFrames,this.detectionConfidence=t.minConfidence,this.maxHands=t.maxHands;const e=Tr.tidy(()=>(n instanceof Tr.Tensor||(n=Tr.browser.fromPixels(n)),n.toFloat().expandDims(0))),i=await this.pipeline.estimateHands(e,t);e.dispose();const r=[];if(!i)return r;for(const a of i){if(!a)return[];const s={};for(const o of Object.keys(GI.MESH_ANNOTATIONS))s[o]=GI.MESH_ANNOTATIONS[o].map(l=>a.landmarks[l]);r.push({confidence:a.confidence||0,box:a.box?[a.box.topLeft[0],a.box.topLeft[1],a.box.bottomRight[0]-a.box.topLeft[0],a.box.bottomRight[1]-a.box.topLeft[1]]:0,landmarks:a.landmarks,annotations:s})}return r}}Qp.HandPose=YI;async function ZH(n){if(Tr.env().features.IS_NODE){const t=require("fs"),e=await t.readFileSync(n.replace("file://",""));return JSON.parse(e)}return Tr.util.fetch(n).then(t=>t.json())}async function QH(n){const[t,e,i]=await Promise.all([ZH(n.detector.anchors),Tr.loadGraphModel(n.detector.modelPath,{fromTFHub:n.detector.modelPath.includes("tfhub.dev")}),Tr.loadGraphModel(n.skeleton.modelPath,{fromTFHub:n.skeleton.modelPath.includes("tfhub.dev")})]),r=new XH.HandDetector(e,t,n),a=new JH.HandPipeline(r,i,n),s=new YI(a);return s}Qp.load=QH});var jI=Ae(eV=>{vl(eV,{default:()=>tV});var tV={face:{enabled:!0,detector:{modelPath:"../models/blazeface/model.json",inputSize:128,maxFaces:10,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7},mesh:{enabled:!0,modelPath:"../models/facemesh/model.json",inputSize:192},iris:{enabled:!0,modelPath:"../models/iris/model.json",enlargeFactor:2.3,inputSize:64},age:{enabled:!0,modelPath:"../models/ssrnet-age/imdb/model.json",inputSize:64,skipFrames:10},gender:{enabled:!0,modelPath:"../models/ssrnet-gender/imdb/model.json"},emotion:{enabled:!0,inputSize:64,minConfidence:.5,skipFrames:10,useGrayscale:!0,modelPath:"../models/emotion/model.json"}},body:{enabled:!0,modelPath:"../models/posenet/model.json",inputResolution:257,outputStride:16,maxDetections:5,scoreThreshold:.7,nmsRadius:20},hand:{enabled:!0,inputSize:256,skipFrames:10,minConfidence:.5,iouThreshold:.3,scoreThreshold:.7,enlargeFactor:1.65,maxHands:2,detector:{anchors:"../models/handdetect/anchors.json",modelPath:"../models/handdetect/model.json"},skeleton:{modelPath:"../models/handskeleton/model.json"}}}});const Nr=Ut(),$I=ZL(),gl=eI(),XI=iI(),JI=EI(),ZI=KI(),QI=jI().default,Gn={facemesh:null,blazeface:null,ssrnet:null,iris:null};function e2(...n){const t=e=>e&&typeof e=="object";return n.reduce((e,i)=>(Object.keys(i||{}).forEach(r=>{const a=e[r],s=i[r];Array.isArray(a)&&Array.isArray(s)?e[r]=a.concat(...s):t(a)&&t(s)?e[r]=e2(a,s):e[r]=s}),e),{})}async function nV(n,t){return new Promise(async e=>{const i=e2(QI,t);i.face.age.enabled&&await gl.loadAge(i),i.face.gender.enabled&&await gl.loadGender(i),i.face.emotion.enabled&&await XI.load(i),i.body.enabled&&!Gn.posenet&&(Gn.posenet=await JI.load(i.body)),i.hand.enabled&&!Gn.handpose&&(Gn.handpose=await ZI.load(i.hand)),i.face.enabled&&!Gn.facemesh&&(Gn.facemesh=await $I.load(i.face)),Nr.engine().startScope();let r;Nr.getBackend()==="webgl"&&(r=Nr.env().get("WEBGL_PACK_DEPTHWISECONV"),Nr.env().set("WEBGL_PACK_DEPTHWISECONV",!0));const a={};let s;s=performance.now();let o=[];i.body.enabled&&(o=await Gn.posenet.estimatePoses(n,i.body)),a.body=Math.trunc(performance.now()-s),s=performance.now();let l=[];i.hand.enabled&&(l=await Gn.handpose.estimateHands(n,i.hand)),a.hand=Math.trunc(performance.now()-s);const u=[];if(i.face.enabled){s=performance.now();const c=await Gn.facemesh.estimateFaces(n,i.face);a.face=Math.trunc(performance.now()-s);for(const h of c){s=performance.now();const d=i.face.age.enabled||i.face.gender.enabled?await gl.predict(h.image,i):{};a.agegender=Math.trunc(performance.now()-s),s=performance.now();const p=i.face.emotion.enabled?await XI.predict(h.image,i):{};a.emotion=Math.trunc(performance.now()-s),h.image.dispose();const f=h.annotations.leftEyeIris&&h.annotations.rightEyeIris?Math.max(h.annotations.leftEyeIris[3][0]-h.annotations.leftEyeIris[1][0],h.annotations.rightEyeIris[3][0]-h.annotations.rightEyeIris[1][0]):0;u.push({confidence:h.confidence,box:h.box,mesh:h.mesh,annotations:h.annotations,age:d.age,gender:d.gender,emotion:p,iris:f!==0?Math.trunc(100*11.7/f)/100:0})}}Nr.env().set("WEBGL_PACK_DEPTHWISECONV",r),Nr.engine().endScope(),a.total=Object.values(a).reduce((c,h)=>c+h),e({face:u,body:o,hand:l,performance:a})})}exports.detect=nV;exports.defaults=QI;exports.models=Gn;exports.facemesh=$I;exports.ssrnet=gl;exports.posenet=JI;exports.handpose=ZI;exports.tf=Nr; +/*! ***************************************************************************** +Copyright (c) Microsoft Corporation. All rights reserved. +Licensed under the Apache License, Version 2.0 (the "License"); you may not use +this file except in compliance with the License. You may obtain a copy of the +License at http://www.apache.org/licenses/LICENSE-2.0 + +THIS CODE IS PROVIDED ON AN *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +MERCHANTABLITY OR NON-INFRINGEMENT. + +See the Apache Version 2.0 License for specific language governing permissions +and limitations under the License. +***************************************************************************** */ +/** + * @license + * Copyright 2017 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * ============================================================================= + */ +/** + * @license + * Copyright 2018 Google LLC + * + * Use of this source code is governed by an MIT-style + * license that can be found in the LICENSE file or at + * https://opensource.org/licenses/MIT. + * ============================================================================= + */ +/** + * @license + * Copyright 2018 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * ============================================================================= + */ +/** + * @license + * Copyright 2018 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * ============================================================================= + */ +/** + * @license + * Copyright 2019 Google LLC + * + * Use of this source code is governed by an MIT-style + * license that can be found in the LICENSE file or at + * https://opensource.org/licenses/MIT. + * ============================================================================= + */ +/** + * @license + * Copyright 2019 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * ============================================================================= + */ +/** + * @license + * Copyright 2019 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * ============================================================================= + */ +/** + * @license + * Copyright 2020 Google Inc. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * ============================================================================= + */ +/** + * @license + * Copyright 2020 Google LLC + * + * Use of this source code is governed by an MIT-style + * license that can be found in the LICENSE file or at + * https://opensource.org/licenses/MIT. + * ============================================================================= + */ +/** + * @license + * Copyright 2020 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * ============================================================================= + */ +/** + * @license + * Copyright 2020 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the License); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * ============================================================================= + */ +/** +* @license +* Copyright 2018 Google LLC. All Rights Reserved. +* Licensed under the Apache License, Version 2.0 (the "License"); +* you may not use this file except in compliance with the License. +* You may obtain a copy of the License at +* +* http://www.apache.org/licenses/LICENSE-2.0 +* +* Unless required by applicable law or agreed to in writing, software +* distributed under the License is distributed on an "AS IS" BASIS, +* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +* See the License for the specific language governing permissions and +* limitations under the License. +* ============================================================================= +*/ +/** @license See the LICENSE file. */ +//# sourceMappingURL=human.cjs.map diff --git a/dist/human.cjs.map b/dist/human.cjs.map new file mode 100644 index 00000000..0ec19762 --- /dev/null +++ b/dist/human.cjs.map @@ -0,0 +1,7 @@ +{ + "version": 3, + "sources": ["../node_modules/node-fetch/lib/index.mjs", "../node_modules/@tensorflow/tfjs-core/src/backends/backend.ts", "../node_modules/@tensorflow/tfjs-core/src/environment.ts", "../node_modules/@tensorflow/tfjs-core/src/global_util.ts", "../node_modules/@tensorflow/tfjs-core/src/kernel_names.ts", "../node_modules/@tensorflow/tfjs-core/src/kernel_registry.ts", "../node_modules/@tensorflow/tfjs-core/src/util.ts", "../node_modules/@tensorflow/tfjs-core/src/profiler.ts", "../node_modules/@tensorflow/tfjs-core/src/tape.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_format.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor.ts", "../node_modules/@tensorflow/tfjs-core/src/types.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_util.ts", "../node_modules/@tensorflow/tfjs-core/src/engine.ts", "../node_modules/@tensorflow/tfjs-core/src/device_util.ts", "../node_modules/@tensorflow/tfjs-core/src/flags.ts", "../node_modules/@tensorflow/tfjs-core/src/tensor_util_env.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/operation.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/complex.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor.ts", "../node_modules/@tensorflow/tfjs-core/src/io/types.ts", "../node_modules/@tensorflow/tfjs-core/src/io/io_utils.ts", "../node_modules/@tensorflow/tfjs-core/src/io/router_registry.ts", "../node_modules/@tensorflow/tfjs-core/src/io/indexed_db.ts", "../node_modules/@tensorflow/tfjs-core/src/io/local_storage.ts", "../node_modules/@tensorflow/tfjs-core/src/io/model_management.ts", "../node_modules/@tensorflow/tfjs-core/src/platforms/platform_browser.ts", "../node_modules/@tensorflow/tfjs-core/src/platforms/platform_node.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/buffer.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cast.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/clone.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/print.ts", "../node_modules/@tensorflow/tfjs-core/src/base_side_effects.ts", "../node_modules/@tensorflow/tfjs-core/src/io/browser_files.ts", "../node_modules/@tensorflow/tfjs-core/src/io/progress.ts", "../node_modules/@tensorflow/tfjs-core/src/io/weights_loader.ts", "../node_modules/@tensorflow/tfjs-core/src/io/http.ts", "../node_modules/@tensorflow/tfjs-core/src/io/passthrough.ts", "../node_modules/@tensorflow/tfjs-core/src/io/io.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reshape.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/one_hot.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/confusion_matrix.ts", "../node_modules/@tensorflow/tfjs-core/src/math.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/browser.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather_nd_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scatter_nd_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice_util.ts", "../node_modules/@tensorflow/tfjs-core/src/serialization.ts", "../node_modules/@tensorflow/tfjs-core/src/test_util.ts", "../node_modules/@tensorflow/tfjs-core/src/version.ts", "../node_modules/@tensorflow/tfjs-core/src/globals.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/add.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/floorDiv.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/div.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/abs.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/acos.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/acosh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/add_n.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/axis_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/all.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/any.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/arg_max.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/arg_min.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/asin.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/asinh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atan2.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/atanh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tanh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/basic_lstm_cell.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batch_to_space_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/batchnorm4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/broadcast_to.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ceil.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/clip_by_value.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/concat_4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cos.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cosh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/cumsum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depth_to_space.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/diag.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dilation2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/broadcast_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/where.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/zeros_like.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/div_no_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dot.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/elu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/erf.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/exp.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/expand_dims.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/expm1.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tile.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/eye.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fill.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/floor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reduce_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/segment_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/greater.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/greater_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/imag.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_finite.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_inf.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/is_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/maximum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scalar.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/leaky_relu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/less.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/less_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linspace.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/local_response_normalization.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log1p.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/neg.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/softplus.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sub.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/log_sum_exp.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_and.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_not.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_or.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/logical_xor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_with_argmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/zeros.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ones.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mean.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/min.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/minimum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/mod.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/square.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/moments.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/multi_rnn_cell.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/multinomial.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/not_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/real.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ones_like.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/outer_product.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pad4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/space_to_batch_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pool.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/pow.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/prelu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/prod.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rand.ts", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/alea.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xor128.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xorwow.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xorshift7.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/xor4096.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/lib/tychei.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/seedrandom.js", "../node_modules/@tensorflow/tfjs-core/node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-core/src/ops/rand_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_gamma.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_normal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/random_uniform.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/range.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reciprocal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/relu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/relu6.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/reverse_4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/round.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rsqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/selu.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/separable_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/setdiff1d_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sign.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sin.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sinh.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice1d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice3d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/slice4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/fft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/ifft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/irfft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/split_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/split.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/spectral/rfft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/squared_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/squeeze.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/stack.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/step.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/strided_slice.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tan.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor4d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor5d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/tensor6d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/topk.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/truncated_normal.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unique.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unsorted_segment_sum.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/unstack.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/variable.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/where_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/where_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/boolean_mask.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/compare.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/binary_ops.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/norm.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/moving_average.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/scatter_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sparse_to_dense_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/sparse_to_dense.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/gather_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dropout_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/dropout.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/in_top_k.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv2d_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d_native_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/depthwise_conv2d_native_backprop_input.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/fused_ops.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/hamming_window.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/hann_window.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/frame.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/signal/stft.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/crop_and_resize.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/flip_left_right.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/rotate_with_offset.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/nonmax_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/array_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/non_max_suppression_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_with_score.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_with_score_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_padded.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/non_max_suppression_padded_async.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/resize_bilinear.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/image/resize_nearest_neighbor.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/band_part.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/gram_schmidt.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/linalg/qr.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/loss_ops_utils.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/compute_weighted_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/absolute_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/cosine_distance.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/hinge_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/huber_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/log_loss.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/mean_squared_error.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/sigmoid_cross_entropy.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/losses/softmax_cross_entropy.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/ops.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adadelta_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adagrad_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adam_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/adamax_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/sgd_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/momentum_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/rmsprop_optimizer.ts", "../node_modules/@tensorflow/tfjs-core/src/optimizers/optimizer_constructors.ts", "../node_modules/@tensorflow/tfjs-core/src/train.ts", "../node_modules/@tensorflow/tfjs-core/src/browser_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/rotate_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/array_ops_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/selu_util.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/erf_util.ts", "../node_modules/@tensorflow/tfjs-core/src/log.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/complex_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/backend_util.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/split_shared.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/tile_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/topk_impl.ts", "../node_modules/@tensorflow/tfjs-core/src/backends/kernel_impls.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Abs_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Acos_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Acosh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Add_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AddN_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ArgMax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ArgMin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Asin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Asinh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atan2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Atanh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_3d_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AvgPool3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/avg_pool_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/AvgPool_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BatchMatMul_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BatchToSpaceND_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/BroadcastTo_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cast_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Ceil_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ClipByValue_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Concat_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv2D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv2DBackpropInput_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/conv3d_backprop_filter.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Conv3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cos_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cosh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Cumsum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/DepthwiseConv2dNative_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Dilation2D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Div_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Elu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Erf_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Exp_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Expm1_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Floor_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/FloorDiv_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/FusedBatchNorm_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/GatherV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/GreaterEqual_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Identity_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsFinite_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsInf_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/IsNan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Log1p_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Log_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/LogSoftmax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/local_response_normalization_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/LRN_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/min_max_grad_util.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Max_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Maximum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_3d_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/MaxPool3D_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/ops/max_pool_backprop.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/MaxPool_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Min_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Minimum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Mod_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Multiply_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Negate_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/OneHot_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/OnesLike_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/PadV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Pow_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Prelu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reciprocal_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Relu6_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Relu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reshape_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ResizeBilinear_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ResizeNearestNeighbor_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Reverse_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Round_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Rsqrt_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SelectV2_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Selu_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sigmoid_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sign_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sin_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sinh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Slice_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Softmax_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Softplus_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SpaceToBatchND_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SplitV_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sqrt_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Square_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/SquaredDifference_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Step_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sub_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Sum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tan_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tanh_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Tile_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Transpose_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/Unpack_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/UnsortedSegmentSum_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/gradients/ZerosLike_grad.ts", "../node_modules/@tensorflow/tfjs-core/src/register_all_gradients.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/abs.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/acos.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/acosh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/add_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/add.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/all.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/any.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/arg_max.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/arg_min.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as_scalar.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as_type.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as1d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as3d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as4d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/as5d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/asin.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/asinh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atan2.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/atanh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/avg_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/batch_to_space_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/batchnorm.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/broadcast_to.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cast.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ceil.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/clip_by_value.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/concat.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv1d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv2d_transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cos.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cosh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/cumsum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depth_to_space.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depthwise_conv2D_deprecated.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/depthwise_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/dilation2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div_no_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/div.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/dot.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/elu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/erf.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/exp.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/expand_dims.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/expm1.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/fft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/flatten.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/floor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/floorDiv.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/gather.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/greater.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ifft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/irfft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_finite.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_inf.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/is_nan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/leaky_relu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/less.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/local_response_normalization.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log_sum_exp.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/log1p.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_and.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_not.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_or.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/logical_xor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mat_mul.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/max_pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/max.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/maximum_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/maximum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mean.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/min.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/minimum_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/minimum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mod_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mod.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mul_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/mul.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/neg.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/norm.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/not_equal_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/not_equal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/one_hot.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/ones_like.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pad.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pow_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/pow.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/prelu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/prod.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reciprocal.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/relu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/relu6.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reshape_as.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reshape.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/resize_bilinear.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/resize_nearest_neighbor.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/reverse.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/rfft.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/round.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/rsqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/selu.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/separable_conv2d.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sigmoid.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sign.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sin.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sinh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/slice.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/softmax.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/softplus.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/space_to_batch_nd.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/split.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sqrt.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/square.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squared_difference.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squared_difference_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/squeeze.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/stack.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/step.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/strided_slice.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sub_strict.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sub.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/sum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tan.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tanh.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/tile.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_bool.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_float.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/to_int.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/topk.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/transpose.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unique.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unsorted_segment_sum.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/unstack.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/where.ts", "../node_modules/@tensorflow/tfjs-core/src/public/chained_ops/zeros_like.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/errors.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/generic_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/constraints.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_constraints.ts", "../node_modules/@tensorflow/tfjs-layers/src/keras_format/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/common.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/math_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/tfjs_backend.ts", "../node_modules/@tensorflow/tfjs-layers/src/keras_format/initializer_config.ts", "../node_modules/@tensorflow/tfjs-layers/src/initializers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_initializers.ts", "../node_modules/@tensorflow/tfjs-layers/src/backend/state.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/types_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/variable_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/variables.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/topology.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/input_layer.ts", "../node_modules/@tensorflow/tfjs-layers/src/logs.ts", "../node_modules/@tensorflow/tfjs-layers/src/base_callbacks.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/serialization.ts", "../node_modules/@tensorflow/tfjs-layers/src/losses.ts", "../node_modules/@tensorflow/tfjs-layers/src/metrics.ts", "../node_modules/@tensorflow/tfjs-layers/src/optimizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/user_defined_metadata.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/layer_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/serialization_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/version.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/executor.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/container.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_dataset.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training_tensors.ts", "../node_modules/@tensorflow/tfjs-layers/src/engine/training.ts", "../node_modules/@tensorflow/tfjs-layers/src/models.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports.ts", "../node_modules/@tensorflow/tfjs-layers/src/activations.ts", "../node_modules/@tensorflow/tfjs-layers/src/regularizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/advanced_activations.ts", "../node_modules/@tensorflow/tfjs-layers/src/utils/conv_utils.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional_depthwise.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/recurrent.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/convolutional_recurrent.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/core.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/embeddings.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/merge.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/noise.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/normalization.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/padding.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/pooling.ts", "../node_modules/@tensorflow/tfjs-layers/src/layers/wrappers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_layers.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_metrics.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_models.ts", "../node_modules/@tensorflow/tfjs-layers/src/exports_regularizers.ts", "../node_modules/@tensorflow/tfjs-layers/src/callbacks.ts", "../node_modules/@tensorflow/tfjs-converter/src/data/compiled_api.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/custom_op/register.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/utils.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/arithmetic.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/basic_math.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/control.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/convolution.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/creation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/dynamic.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/evaluation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/graph.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/image.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/logical.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/matrices.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/normalization.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/reduction.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/slice_join.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/spectral.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/op_list/transformation.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/operation_mapper.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/custom_op/node_value_impl.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/arithmetic_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/basic_math_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_utils.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_array.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/tensor_list.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/control_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/convolution_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/creation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/dynamic_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/evaluation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/graph_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/image_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/logical_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/matrices_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/normalization_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/reduction_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/slice_join_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/spectral_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/executors/transformation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/operations/operation_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/execution_context.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/model_analysis.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/graph_executor.ts", "../node_modules/@tensorflow/tfjs-converter/src/executor/graph_model.ts", "../node_modules/@tensorflow/tfjs-converter/src/version.ts", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/alea.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xor128.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xorwow.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xorshift7.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/xor4096.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/lib/tychei.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/seedrandom.js", "../node_modules/@tensorflow/tfjs-data/node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-data/src/util/deep_map.ts", "../node_modules/@tensorflow/tfjs-data/src/util/deep_clone.ts", "../node_modules/@tensorflow/tfjs-data/src/util/ring_buffer.ts", "../node_modules/@tensorflow/tfjs-data/src/util/growing_ring_buffer.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/lazy_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/datasets/text_line_dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/datasets/csv_dataset.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/microphone_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/webcam_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/datasource.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/string_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/byte_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/file_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/iterators/url_chunk_iterator.ts", "../node_modules/@tensorflow/tfjs-data/src/util/source_util.ts", "../node_modules/@tensorflow/tfjs-data/src/sources/file_data_source.ts", "../node_modules/@tensorflow/tfjs-data/src/sources/url_data_source.ts", "../node_modules/@tensorflow/tfjs-data/src/readers.ts", "../node_modules/@tensorflow/tfjs-data/src/version.ts", "../node_modules/seedrandom/lib/alea.js", "../node_modules/seedrandom/lib/xor128.js", "../node_modules/seedrandom/lib/xorwow.js", "../node_modules/seedrandom/lib/xorshift7.js", "../node_modules/seedrandom/lib/xor4096.js", "../node_modules/seedrandom/lib/tychei.js", "../node_modules/seedrandom/seedrandom.js", "../node_modules/seedrandom/index.js", "../node_modules/@tensorflow/tfjs-backend-cpu/src/cpu_util.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/backend_cpu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Abs.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/binary_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Complex.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Identity.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Real.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cast.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/kernel_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Add.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/unary_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/unary_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Ceil.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Exp.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Expm1.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Floor.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Log.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Max_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Multiply.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Rsqrt.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Slice.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sub.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Transpose_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Unique_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/shared.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/version.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/base.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Acos.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Acosh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Asin.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Asinh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Atan.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Atanh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/pool_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/AvgPool.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/AvgPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/BatchNorm.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Clip.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Imag.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Reshape.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Concat.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cos.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Cosh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2D.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2DBackpropFilter.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Dilation2DBackpropInput.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Div.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Elu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Erf.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/utils/fft_utils.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/FFT.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/FlipLeftRight.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IFFT.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsFinite.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsInf.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/IsNaN.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Log1p.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/LogicalNot.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Max.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPool.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolWithArgmax_impl.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/MaxPoolWithArgmax.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NonMaxSuppressionV4.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NonMaxSuppressionV5.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/NotEqual.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/PadV2.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Reciprocal.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/RotateWithOffset.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Round.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Selu.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sigmoid.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sign.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sin.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sinh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Softplus.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Transpose.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/SpaceToBatchND.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Sqrt.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Square.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/SquaredDifference.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Step.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Tan.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Tanh.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/kernels/Unique.ts", "../node_modules/@tensorflow/tfjs-backend-cpu/src/register_all_kernels.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/canvas_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/tex_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/webgl_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/flags_webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Abs.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/utils/binary_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Add.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/utils/unary_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Ceil.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Exp.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Expm1.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Floor.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Log.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Max_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Multiply.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Rsqrt.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Slice.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Sub.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Transpose_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/node_modules/@tensorflow/tfjs-backend-cpu/dist/kernels/Unique_impl.js", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/shared.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/addn_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/addn_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/argminmax_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/packing_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/glsl_version.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/shader_compiler_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/shader_compiler.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/argminmax_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/avg_pool_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_complex_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/binaryop_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/clip_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/clip_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/complex_abs_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/concat_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/concat_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_backprop_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/conv_packed_gpu_depthwise.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/crop_and_resize_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/cumsum_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/decode_matrix_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/decode_matrix_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/depth_to_space_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/diag_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_float_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_float_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_matrix_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/encode_matrix_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/fft_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/fill_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gather_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gather_nd_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_util.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_context.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/gpgpu_math.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/im2col_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_grad_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/lrn_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/max_pool_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/mulmat_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/multinomial_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/onehot_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pack_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pad_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pad_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/pool_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reduce_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reshape_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_bilinear_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_nearest_neighbor_backprop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/resize_nearest_neighbor_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reverse_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/reverse_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/scatter_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/segment_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/select_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/slice_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/slice_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/strided_slice_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/texture_manager.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/tile_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unaryop_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unaryop_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/unpack_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/backend_webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/version.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/webgl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/base.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/kernel_funcs_utils.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Atan2.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Identity.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/AvgPool.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/AvgPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/batchnorm_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/batchnorm_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/BatchNorm.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Cos.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Div.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/flip_left_right_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FlipLeftRight.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels_utils/from_pixels_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels_utils/from_pixels_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/FromPixels.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/reduce.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernel_utils/reshape.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Reshape.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Max_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/transpose_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/transpose_packed_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Transpose_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Max.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPool.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolBackprop.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolWithArgmax_impl.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/MaxPoolWithArgmax.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV3.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV4.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/NonMaxSuppressionV5.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/rotate_gpu.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/RotateWithOffset.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Sin.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Square.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/SquaredDifference.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Tan.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Transpose.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/kernels/Unique.ts", "../node_modules/@tensorflow/tfjs-backend-webgl/src/register_all_kernels.ts", "../node_modules/@tensorflow/tfjs/src/version.ts", "../node_modules/@tensorflow/tfjs/src/index.ts", "../src/facemesh/blazeface.js", "../src/facemesh/keypoints.js", "../src/facemesh/box.js", "../src/facemesh/util.js", "../src/facemesh/pipeline.js", "../src/facemesh/uvcoords.js", "../src/facemesh/triangulation.js", "../src/facemesh/facemesh.js", "../src/ssrnet/ssrnet.js", "../src/emotion/emotion.js", "../src/posenet/modelBase.js", "../src/posenet/modelMobileNet.js", "../src/posenet/heapSort.js", "../src/posenet/buildParts.js", "../src/posenet/keypoints.js", "../src/posenet/vectors.js", "../src/posenet/decodePose.js", "../src/posenet/decodeMultiple.js", "../src/posenet/util.js", "../src/posenet/modelPoseNet.js", "../src/posenet/posenet.js", "../src/handpose/box.js", "../src/handpose/handdetector.js", "../src/handpose/keypoints.js", "../src/handpose/util.js", "../src/handpose/pipeline.js", "../src/handpose/handpose.js", "../src/config.js", "../src/index.js"], + "sourcesContent": ["import Stream from 'stream';\nimport http from 'http';\nimport Url from 'url';\nimport https from 'https';\nimport zlib from 'zlib';\n\n// Based on https://github.com/tmpvar/jsdom/blob/aa85b2abf07766ff7bf5c1f6daafb3726f2f2db5/lib/jsdom/living/blob.js\n\n// fix for \"Readable\" isn't a named export issue\nconst Readable = Stream.Readable;\n\nconst BUFFER = Symbol('buffer');\nconst TYPE = Symbol('type');\n\nclass Blob {\n\tconstructor() {\n\t\tthis[TYPE] = '';\n\n\t\tconst blobParts = arguments[0];\n\t\tconst options = arguments[1];\n\n\t\tconst buffers = [];\n\t\tlet size = 0;\n\n\t\tif (blobParts) {\n\t\t\tconst a = blobParts;\n\t\t\tconst length = Number(a.length);\n\t\t\tfor (let i = 0; i < length; i++) {\n\t\t\t\tconst element = a[i];\n\t\t\t\tlet buffer;\n\t\t\t\tif (element instanceof Buffer) {\n\t\t\t\t\tbuffer = element;\n\t\t\t\t} else if (ArrayBuffer.isView(element)) {\n\t\t\t\t\tbuffer = Buffer.from(element.buffer, element.byteOffset, element.byteLength);\n\t\t\t\t} else if (element instanceof ArrayBuffer) {\n\t\t\t\t\tbuffer = Buffer.from(element);\n\t\t\t\t} else if (element instanceof Blob) {\n\t\t\t\t\tbuffer = element[BUFFER];\n\t\t\t\t} else {\n\t\t\t\t\tbuffer = Buffer.from(typeof element === 'string' ? element : String(element));\n\t\t\t\t}\n\t\t\t\tsize += buffer.length;\n\t\t\t\tbuffers.push(buffer);\n\t\t\t}\n\t\t}\n\n\t\tthis[BUFFER] = Buffer.concat(buffers);\n\n\t\tlet type = options && options.type !== undefined && String(options.type).toLowerCase();\n\t\tif (type && !/[^\\u0020-\\u007E]/.test(type)) {\n\t\t\tthis[TYPE] = type;\n\t\t}\n\t}\n\tget size() {\n\t\treturn this[BUFFER].length;\n\t}\n\tget type() {\n\t\treturn this[TYPE];\n\t}\n\ttext() {\n\t\treturn Promise.resolve(this[BUFFER].toString());\n\t}\n\tarrayBuffer() {\n\t\tconst buf = this[BUFFER];\n\t\tconst ab = buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n\t\treturn Promise.resolve(ab);\n\t}\n\tstream() {\n\t\tconst readable = new Readable();\n\t\treadable._read = function () {};\n\t\treadable.push(this[BUFFER]);\n\t\treadable.push(null);\n\t\treturn readable;\n\t}\n\ttoString() {\n\t\treturn '[object Blob]';\n\t}\n\tslice() {\n\t\tconst size = this.size;\n\n\t\tconst start = arguments[0];\n\t\tconst end = arguments[1];\n\t\tlet relativeStart, relativeEnd;\n\t\tif (start === undefined) {\n\t\t\trelativeStart = 0;\n\t\t} else if (start < 0) {\n\t\t\trelativeStart = Math.max(size + start, 0);\n\t\t} else {\n\t\t\trelativeStart = Math.min(start, size);\n\t\t}\n\t\tif (end === undefined) {\n\t\t\trelativeEnd = size;\n\t\t} else if (end < 0) {\n\t\t\trelativeEnd = Math.max(size + end, 0);\n\t\t} else {\n\t\t\trelativeEnd = Math.min(end, size);\n\t\t}\n\t\tconst span = Math.max(relativeEnd - relativeStart, 0);\n\n\t\tconst buffer = this[BUFFER];\n\t\tconst slicedBuffer = buffer.slice(relativeStart, relativeStart + span);\n\t\tconst blob = new Blob([], { type: arguments[2] });\n\t\tblob[BUFFER] = slicedBuffer;\n\t\treturn blob;\n\t}\n}\n\nObject.defineProperties(Blob.prototype, {\n\tsize: { enumerable: true },\n\ttype: { enumerable: true },\n\tslice: { enumerable: true }\n});\n\nObject.defineProperty(Blob.prototype, Symbol.toStringTag, {\n\tvalue: 'Blob',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\n/**\n * fetch-error.js\n *\n * FetchError interface for operational errors\n */\n\n/**\n * Create FetchError instance\n *\n * @param String message Error message for human\n * @param String type Error type for machine\n * @param String systemError For Node.js system error\n * @return FetchError\n */\nfunction FetchError(message, type, systemError) {\n Error.call(this, message);\n\n this.message = message;\n this.type = type;\n\n // when err.type is `system`, err.code contains system error code\n if (systemError) {\n this.code = this.errno = systemError.code;\n }\n\n // hide custom error implementation details from end-users\n Error.captureStackTrace(this, this.constructor);\n}\n\nFetchError.prototype = Object.create(Error.prototype);\nFetchError.prototype.constructor = FetchError;\nFetchError.prototype.name = 'FetchError';\n\nlet convert;\ntry {\n\tconvert = require('encoding').convert;\n} catch (e) {}\n\nconst INTERNALS = Symbol('Body internals');\n\n// fix an issue where \"PassThrough\" isn't a named export for node <10\nconst PassThrough = Stream.PassThrough;\n\n/**\n * Body mixin\n *\n * Ref: https://fetch.spec.whatwg.org/#body\n *\n * @param Stream body Readable stream\n * @param Object opts Response options\n * @return Void\n */\nfunction Body(body) {\n\tvar _this = this;\n\n\tvar _ref = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {},\n\t _ref$size = _ref.size;\n\n\tlet size = _ref$size === undefined ? 0 : _ref$size;\n\tvar _ref$timeout = _ref.timeout;\n\tlet timeout = _ref$timeout === undefined ? 0 : _ref$timeout;\n\n\tif (body == null) {\n\t\t// body is undefined or null\n\t\tbody = null;\n\t} else if (isURLSearchParams(body)) {\n\t\t// body is a URLSearchParams\n\t\tbody = Buffer.from(body.toString());\n\t} else if (isBlob(body)) ; else if (Buffer.isBuffer(body)) ; else if (Object.prototype.toString.call(body) === '[object ArrayBuffer]') {\n\t\t// body is ArrayBuffer\n\t\tbody = Buffer.from(body);\n\t} else if (ArrayBuffer.isView(body)) {\n\t\t// body is ArrayBufferView\n\t\tbody = Buffer.from(body.buffer, body.byteOffset, body.byteLength);\n\t} else if (body instanceof Stream) ; else {\n\t\t// none of the above\n\t\t// coerce to string then buffer\n\t\tbody = Buffer.from(String(body));\n\t}\n\tthis[INTERNALS] = {\n\t\tbody,\n\t\tdisturbed: false,\n\t\terror: null\n\t};\n\tthis.size = size;\n\tthis.timeout = timeout;\n\n\tif (body instanceof Stream) {\n\t\tbody.on('error', function (err) {\n\t\t\tconst error = err.name === 'AbortError' ? err : new FetchError(`Invalid response body while trying to fetch ${_this.url}: ${err.message}`, 'system', err);\n\t\t\t_this[INTERNALS].error = error;\n\t\t});\n\t}\n}\n\nBody.prototype = {\n\tget body() {\n\t\treturn this[INTERNALS].body;\n\t},\n\n\tget bodyUsed() {\n\t\treturn this[INTERNALS].disturbed;\n\t},\n\n\t/**\n * Decode response as ArrayBuffer\n *\n * @return Promise\n */\n\tarrayBuffer() {\n\t\treturn consumeBody.call(this).then(function (buf) {\n\t\t\treturn buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n\t\t});\n\t},\n\n\t/**\n * Return raw response as Blob\n *\n * @return Promise\n */\n\tblob() {\n\t\tlet ct = this.headers && this.headers.get('content-type') || '';\n\t\treturn consumeBody.call(this).then(function (buf) {\n\t\t\treturn Object.assign(\n\t\t\t// Prevent copying\n\t\t\tnew Blob([], {\n\t\t\t\ttype: ct.toLowerCase()\n\t\t\t}), {\n\t\t\t\t[BUFFER]: buf\n\t\t\t});\n\t\t});\n\t},\n\n\t/**\n * Decode response as json\n *\n * @return Promise\n */\n\tjson() {\n\t\tvar _this2 = this;\n\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\ttry {\n\t\t\t\treturn JSON.parse(buffer.toString());\n\t\t\t} catch (err) {\n\t\t\t\treturn Body.Promise.reject(new FetchError(`invalid json response body at ${_this2.url} reason: ${err.message}`, 'invalid-json'));\n\t\t\t}\n\t\t});\n\t},\n\n\t/**\n * Decode response as text\n *\n * @return Promise\n */\n\ttext() {\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\treturn buffer.toString();\n\t\t});\n\t},\n\n\t/**\n * Decode response as buffer (non-spec api)\n *\n * @return Promise\n */\n\tbuffer() {\n\t\treturn consumeBody.call(this);\n\t},\n\n\t/**\n * Decode response as text, while automatically detecting the encoding and\n * trying to decode to UTF-8 (non-spec api)\n *\n * @return Promise\n */\n\ttextConverted() {\n\t\tvar _this3 = this;\n\n\t\treturn consumeBody.call(this).then(function (buffer) {\n\t\t\treturn convertBody(buffer, _this3.headers);\n\t\t});\n\t}\n};\n\n// In browsers, all properties are enumerable.\nObject.defineProperties(Body.prototype, {\n\tbody: { enumerable: true },\n\tbodyUsed: { enumerable: true },\n\tarrayBuffer: { enumerable: true },\n\tblob: { enumerable: true },\n\tjson: { enumerable: true },\n\ttext: { enumerable: true }\n});\n\nBody.mixIn = function (proto) {\n\tfor (const name of Object.getOwnPropertyNames(Body.prototype)) {\n\t\t// istanbul ignore else: future proof\n\t\tif (!(name in proto)) {\n\t\t\tconst desc = Object.getOwnPropertyDescriptor(Body.prototype, name);\n\t\t\tObject.defineProperty(proto, name, desc);\n\t\t}\n\t}\n};\n\n/**\n * Consume and convert an entire Body to a Buffer.\n *\n * Ref: https://fetch.spec.whatwg.org/#concept-body-consume-body\n *\n * @return Promise\n */\nfunction consumeBody() {\n\tvar _this4 = this;\n\n\tif (this[INTERNALS].disturbed) {\n\t\treturn Body.Promise.reject(new TypeError(`body used already for: ${this.url}`));\n\t}\n\n\tthis[INTERNALS].disturbed = true;\n\n\tif (this[INTERNALS].error) {\n\t\treturn Body.Promise.reject(this[INTERNALS].error);\n\t}\n\n\tlet body = this.body;\n\n\t// body is null\n\tif (body === null) {\n\t\treturn Body.Promise.resolve(Buffer.alloc(0));\n\t}\n\n\t// body is blob\n\tif (isBlob(body)) {\n\t\tbody = body.stream();\n\t}\n\n\t// body is buffer\n\tif (Buffer.isBuffer(body)) {\n\t\treturn Body.Promise.resolve(body);\n\t}\n\n\t// istanbul ignore if: should never happen\n\tif (!(body instanceof Stream)) {\n\t\treturn Body.Promise.resolve(Buffer.alloc(0));\n\t}\n\n\t// body is stream\n\t// get ready to actually consume the body\n\tlet accum = [];\n\tlet accumBytes = 0;\n\tlet abort = false;\n\n\treturn new Body.Promise(function (resolve, reject) {\n\t\tlet resTimeout;\n\n\t\t// allow timeout on slow response body\n\t\tif (_this4.timeout) {\n\t\t\tresTimeout = setTimeout(function () {\n\t\t\t\tabort = true;\n\t\t\t\treject(new FetchError(`Response timeout while trying to fetch ${_this4.url} (over ${_this4.timeout}ms)`, 'body-timeout'));\n\t\t\t}, _this4.timeout);\n\t\t}\n\n\t\t// handle stream errors\n\t\tbody.on('error', function (err) {\n\t\t\tif (err.name === 'AbortError') {\n\t\t\t\t// if the request was aborted, reject with this Error\n\t\t\t\tabort = true;\n\t\t\t\treject(err);\n\t\t\t} else {\n\t\t\t\t// other errors, such as incorrect content-encoding\n\t\t\t\treject(new FetchError(`Invalid response body while trying to fetch ${_this4.url}: ${err.message}`, 'system', err));\n\t\t\t}\n\t\t});\n\n\t\tbody.on('data', function (chunk) {\n\t\t\tif (abort || chunk === null) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif (_this4.size && accumBytes + chunk.length > _this4.size) {\n\t\t\t\tabort = true;\n\t\t\t\treject(new FetchError(`content size at ${_this4.url} over limit: ${_this4.size}`, 'max-size'));\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\taccumBytes += chunk.length;\n\t\t\taccum.push(chunk);\n\t\t});\n\n\t\tbody.on('end', function () {\n\t\t\tif (abort) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tclearTimeout(resTimeout);\n\n\t\t\ttry {\n\t\t\t\tresolve(Buffer.concat(accum, accumBytes));\n\t\t\t} catch (err) {\n\t\t\t\t// handle streams that have accumulated too much data (issue #414)\n\t\t\t\treject(new FetchError(`Could not create Buffer from response body for ${_this4.url}: ${err.message}`, 'system', err));\n\t\t\t}\n\t\t});\n\t});\n}\n\n/**\n * Detect buffer encoding and convert to target encoding\n * ref: http://www.w3.org/TR/2011/WD-html5-20110113/parsing.html#determining-the-character-encoding\n *\n * @param Buffer buffer Incoming buffer\n * @param String encoding Target encoding\n * @return String\n */\nfunction convertBody(buffer, headers) {\n\tif (typeof convert !== 'function') {\n\t\tthrow new Error('The package `encoding` must be installed to use the textConverted() function');\n\t}\n\n\tconst ct = headers.get('content-type');\n\tlet charset = 'utf-8';\n\tlet res, str;\n\n\t// header\n\tif (ct) {\n\t\tres = /charset=([^;]*)/i.exec(ct);\n\t}\n\n\t// no charset in content type, peek at response body for at most 1024 bytes\n\tstr = buffer.slice(0, 1024).toString();\n\n\t// html5\n\tif (!res && str) {\n\t\tres = / 0 && arguments[0] !== undefined ? arguments[0] : undefined;\n\n\t\tthis[MAP] = Object.create(null);\n\n\t\tif (init instanceof Headers) {\n\t\t\tconst rawHeaders = init.raw();\n\t\t\tconst headerNames = Object.keys(rawHeaders);\n\n\t\t\tfor (const headerName of headerNames) {\n\t\t\t\tfor (const value of rawHeaders[headerName]) {\n\t\t\t\t\tthis.append(headerName, value);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\treturn;\n\t\t}\n\n\t\t// We don't worry about converting prop to ByteString here as append()\n\t\t// will handle it.\n\t\tif (init == null) ; else if (typeof init === 'object') {\n\t\t\tconst method = init[Symbol.iterator];\n\t\t\tif (method != null) {\n\t\t\t\tif (typeof method !== 'function') {\n\t\t\t\t\tthrow new TypeError('Header pairs must be iterable');\n\t\t\t\t}\n\n\t\t\t\t// sequence>\n\t\t\t\t// Note: per spec we have to first exhaust the lists then process them\n\t\t\t\tconst pairs = [];\n\t\t\t\tfor (const pair of init) {\n\t\t\t\t\tif (typeof pair !== 'object' || typeof pair[Symbol.iterator] !== 'function') {\n\t\t\t\t\t\tthrow new TypeError('Each header pair must be iterable');\n\t\t\t\t\t}\n\t\t\t\t\tpairs.push(Array.from(pair));\n\t\t\t\t}\n\n\t\t\t\tfor (const pair of pairs) {\n\t\t\t\t\tif (pair.length !== 2) {\n\t\t\t\t\t\tthrow new TypeError('Each header pair must be a name/value tuple');\n\t\t\t\t\t}\n\t\t\t\t\tthis.append(pair[0], pair[1]);\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\t// record\n\t\t\t\tfor (const key of Object.keys(init)) {\n\t\t\t\t\tconst value = init[key];\n\t\t\t\t\tthis.append(key, value);\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tthrow new TypeError('Provided initializer must be an object');\n\t\t}\n\t}\n\n\t/**\n * Return combined header value given name\n *\n * @param String name Header name\n * @return Mixed\n */\n\tget(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key === undefined) {\n\t\t\treturn null;\n\t\t}\n\n\t\treturn this[MAP][key].join(', ');\n\t}\n\n\t/**\n * Iterate over all headers\n *\n * @param Function callback Executed for each item with parameters (value, name, thisArg)\n * @param Boolean thisArg `this` context for callback function\n * @return Void\n */\n\tforEach(callback) {\n\t\tlet thisArg = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : undefined;\n\n\t\tlet pairs = getHeaders(this);\n\t\tlet i = 0;\n\t\twhile (i < pairs.length) {\n\t\t\tvar _pairs$i = pairs[i];\n\t\t\tconst name = _pairs$i[0],\n\t\t\t value = _pairs$i[1];\n\n\t\t\tcallback.call(thisArg, value, name, this);\n\t\t\tpairs = getHeaders(this);\n\t\t\ti++;\n\t\t}\n\t}\n\n\t/**\n * Overwrite header values given name\n *\n * @param String name Header name\n * @param String value Header value\n * @return Void\n */\n\tset(name, value) {\n\t\tname = `${name}`;\n\t\tvalue = `${value}`;\n\t\tvalidateName(name);\n\t\tvalidateValue(value);\n\t\tconst key = find(this[MAP], name);\n\t\tthis[MAP][key !== undefined ? key : name] = [value];\n\t}\n\n\t/**\n * Append a value onto existing header\n *\n * @param String name Header name\n * @param String value Header value\n * @return Void\n */\n\tappend(name, value) {\n\t\tname = `${name}`;\n\t\tvalue = `${value}`;\n\t\tvalidateName(name);\n\t\tvalidateValue(value);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key !== undefined) {\n\t\t\tthis[MAP][key].push(value);\n\t\t} else {\n\t\t\tthis[MAP][name] = [value];\n\t\t}\n\t}\n\n\t/**\n * Check for header name existence\n *\n * @param String name Header name\n * @return Boolean\n */\n\thas(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\treturn find(this[MAP], name) !== undefined;\n\t}\n\n\t/**\n * Delete all header values given name\n *\n * @param String name Header name\n * @return Void\n */\n\tdelete(name) {\n\t\tname = `${name}`;\n\t\tvalidateName(name);\n\t\tconst key = find(this[MAP], name);\n\t\tif (key !== undefined) {\n\t\t\tdelete this[MAP][key];\n\t\t}\n\t}\n\n\t/**\n * Return raw headers (non-spec api)\n *\n * @return Object\n */\n\traw() {\n\t\treturn this[MAP];\n\t}\n\n\t/**\n * Get an iterator on keys.\n *\n * @return Iterator\n */\n\tkeys() {\n\t\treturn createHeadersIterator(this, 'key');\n\t}\n\n\t/**\n * Get an iterator on values.\n *\n * @return Iterator\n */\n\tvalues() {\n\t\treturn createHeadersIterator(this, 'value');\n\t}\n\n\t/**\n * Get an iterator on entries.\n *\n * This is the default iterator of the Headers object.\n *\n * @return Iterator\n */\n\t[Symbol.iterator]() {\n\t\treturn createHeadersIterator(this, 'key+value');\n\t}\n}\nHeaders.prototype.entries = Headers.prototype[Symbol.iterator];\n\nObject.defineProperty(Headers.prototype, Symbol.toStringTag, {\n\tvalue: 'Headers',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nObject.defineProperties(Headers.prototype, {\n\tget: { enumerable: true },\n\tforEach: { enumerable: true },\n\tset: { enumerable: true },\n\tappend: { enumerable: true },\n\thas: { enumerable: true },\n\tdelete: { enumerable: true },\n\tkeys: { enumerable: true },\n\tvalues: { enumerable: true },\n\tentries: { enumerable: true }\n});\n\nfunction getHeaders(headers) {\n\tlet kind = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : 'key+value';\n\n\tconst keys = Object.keys(headers[MAP]).sort();\n\treturn keys.map(kind === 'key' ? function (k) {\n\t\treturn k.toLowerCase();\n\t} : kind === 'value' ? function (k) {\n\t\treturn headers[MAP][k].join(', ');\n\t} : function (k) {\n\t\treturn [k.toLowerCase(), headers[MAP][k].join(', ')];\n\t});\n}\n\nconst INTERNAL = Symbol('internal');\n\nfunction createHeadersIterator(target, kind) {\n\tconst iterator = Object.create(HeadersIteratorPrototype);\n\titerator[INTERNAL] = {\n\t\ttarget,\n\t\tkind,\n\t\tindex: 0\n\t};\n\treturn iterator;\n}\n\nconst HeadersIteratorPrototype = Object.setPrototypeOf({\n\tnext() {\n\t\t// istanbul ignore if\n\t\tif (!this || Object.getPrototypeOf(this) !== HeadersIteratorPrototype) {\n\t\t\tthrow new TypeError('Value of `this` is not a HeadersIterator');\n\t\t}\n\n\t\tvar _INTERNAL = this[INTERNAL];\n\t\tconst target = _INTERNAL.target,\n\t\t kind = _INTERNAL.kind,\n\t\t index = _INTERNAL.index;\n\n\t\tconst values = getHeaders(target, kind);\n\t\tconst len = values.length;\n\t\tif (index >= len) {\n\t\t\treturn {\n\t\t\t\tvalue: undefined,\n\t\t\t\tdone: true\n\t\t\t};\n\t\t}\n\n\t\tthis[INTERNAL].index = index + 1;\n\n\t\treturn {\n\t\t\tvalue: values[index],\n\t\t\tdone: false\n\t\t};\n\t}\n}, Object.getPrototypeOf(Object.getPrototypeOf([][Symbol.iterator]())));\n\nObject.defineProperty(HeadersIteratorPrototype, Symbol.toStringTag, {\n\tvalue: 'HeadersIterator',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\n/**\n * Export the Headers object in a form that Node.js can consume.\n *\n * @param Headers headers\n * @return Object\n */\nfunction exportNodeCompatibleHeaders(headers) {\n\tconst obj = Object.assign({ __proto__: null }, headers[MAP]);\n\n\t// http.request() only supports string as Host header. This hack makes\n\t// specifying custom Host header possible.\n\tconst hostHeaderKey = find(headers[MAP], 'Host');\n\tif (hostHeaderKey !== undefined) {\n\t\tobj[hostHeaderKey] = obj[hostHeaderKey][0];\n\t}\n\n\treturn obj;\n}\n\n/**\n * Create a Headers object from an object of headers, ignoring those that do\n * not conform to HTTP grammar productions.\n *\n * @param Object obj Object of headers\n * @return Headers\n */\nfunction createHeadersLenient(obj) {\n\tconst headers = new Headers();\n\tfor (const name of Object.keys(obj)) {\n\t\tif (invalidTokenRegex.test(name)) {\n\t\t\tcontinue;\n\t\t}\n\t\tif (Array.isArray(obj[name])) {\n\t\t\tfor (const val of obj[name]) {\n\t\t\t\tif (invalidHeaderCharRegex.test(val)) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tif (headers[MAP][name] === undefined) {\n\t\t\t\t\theaders[MAP][name] = [val];\n\t\t\t\t} else {\n\t\t\t\t\theaders[MAP][name].push(val);\n\t\t\t\t}\n\t\t\t}\n\t\t} else if (!invalidHeaderCharRegex.test(obj[name])) {\n\t\t\theaders[MAP][name] = [obj[name]];\n\t\t}\n\t}\n\treturn headers;\n}\n\nconst INTERNALS$1 = Symbol('Response internals');\n\n// fix an issue where \"STATUS_CODES\" aren't a named export for node <10\nconst STATUS_CODES = http.STATUS_CODES;\n\n/**\n * Response class\n *\n * @param Stream body Readable stream\n * @param Object opts Response options\n * @return Void\n */\nclass Response {\n\tconstructor() {\n\t\tlet body = arguments.length > 0 && arguments[0] !== undefined ? arguments[0] : null;\n\t\tlet opts = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};\n\n\t\tBody.call(this, body, opts);\n\n\t\tconst status = opts.status || 200;\n\t\tconst headers = new Headers(opts.headers);\n\n\t\tif (body != null && !headers.has('Content-Type')) {\n\t\t\tconst contentType = extractContentType(body);\n\t\t\tif (contentType) {\n\t\t\t\theaders.append('Content-Type', contentType);\n\t\t\t}\n\t\t}\n\n\t\tthis[INTERNALS$1] = {\n\t\t\turl: opts.url,\n\t\t\tstatus,\n\t\t\tstatusText: opts.statusText || STATUS_CODES[status],\n\t\t\theaders,\n\t\t\tcounter: opts.counter\n\t\t};\n\t}\n\n\tget url() {\n\t\treturn this[INTERNALS$1].url || '';\n\t}\n\n\tget status() {\n\t\treturn this[INTERNALS$1].status;\n\t}\n\n\t/**\n * Convenience property representing if the request ended normally\n */\n\tget ok() {\n\t\treturn this[INTERNALS$1].status >= 200 && this[INTERNALS$1].status < 300;\n\t}\n\n\tget redirected() {\n\t\treturn this[INTERNALS$1].counter > 0;\n\t}\n\n\tget statusText() {\n\t\treturn this[INTERNALS$1].statusText;\n\t}\n\n\tget headers() {\n\t\treturn this[INTERNALS$1].headers;\n\t}\n\n\t/**\n * Clone this response\n *\n * @return Response\n */\n\tclone() {\n\t\treturn new Response(clone(this), {\n\t\t\turl: this.url,\n\t\t\tstatus: this.status,\n\t\t\tstatusText: this.statusText,\n\t\t\theaders: this.headers,\n\t\t\tok: this.ok,\n\t\t\tredirected: this.redirected\n\t\t});\n\t}\n}\n\nBody.mixIn(Response.prototype);\n\nObject.defineProperties(Response.prototype, {\n\turl: { enumerable: true },\n\tstatus: { enumerable: true },\n\tok: { enumerable: true },\n\tredirected: { enumerable: true },\n\tstatusText: { enumerable: true },\n\theaders: { enumerable: true },\n\tclone: { enumerable: true }\n});\n\nObject.defineProperty(Response.prototype, Symbol.toStringTag, {\n\tvalue: 'Response',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nconst INTERNALS$2 = Symbol('Request internals');\n\n// fix an issue where \"format\", \"parse\" aren't a named export for node <10\nconst parse_url = Url.parse;\nconst format_url = Url.format;\n\nconst streamDestructionSupported = 'destroy' in Stream.Readable.prototype;\n\n/**\n * Check if a value is an instance of Request.\n *\n * @param Mixed input\n * @return Boolean\n */\nfunction isRequest(input) {\n\treturn typeof input === 'object' && typeof input[INTERNALS$2] === 'object';\n}\n\nfunction isAbortSignal(signal) {\n\tconst proto = signal && typeof signal === 'object' && Object.getPrototypeOf(signal);\n\treturn !!(proto && proto.constructor.name === 'AbortSignal');\n}\n\n/**\n * Request class\n *\n * @param Mixed input Url or Request instance\n * @param Object init Custom options\n * @return Void\n */\nclass Request {\n\tconstructor(input) {\n\t\tlet init = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};\n\n\t\tlet parsedURL;\n\n\t\t// normalize input\n\t\tif (!isRequest(input)) {\n\t\t\tif (input && input.href) {\n\t\t\t\t// in order to support Node.js' Url objects; though WHATWG's URL objects\n\t\t\t\t// will fall into this branch also (since their `toString()` will return\n\t\t\t\t// `href` property anyway)\n\t\t\t\tparsedURL = parse_url(input.href);\n\t\t\t} else {\n\t\t\t\t// coerce input to a string before attempting to parse\n\t\t\t\tparsedURL = parse_url(`${input}`);\n\t\t\t}\n\t\t\tinput = {};\n\t\t} else {\n\t\t\tparsedURL = parse_url(input.url);\n\t\t}\n\n\t\tlet method = init.method || input.method || 'GET';\n\t\tmethod = method.toUpperCase();\n\n\t\tif ((init.body != null || isRequest(input) && input.body !== null) && (method === 'GET' || method === 'HEAD')) {\n\t\t\tthrow new TypeError('Request with GET/HEAD method cannot have body');\n\t\t}\n\n\t\tlet inputBody = init.body != null ? init.body : isRequest(input) && input.body !== null ? clone(input) : null;\n\n\t\tBody.call(this, inputBody, {\n\t\t\ttimeout: init.timeout || input.timeout || 0,\n\t\t\tsize: init.size || input.size || 0\n\t\t});\n\n\t\tconst headers = new Headers(init.headers || input.headers || {});\n\n\t\tif (inputBody != null && !headers.has('Content-Type')) {\n\t\t\tconst contentType = extractContentType(inputBody);\n\t\t\tif (contentType) {\n\t\t\t\theaders.append('Content-Type', contentType);\n\t\t\t}\n\t\t}\n\n\t\tlet signal = isRequest(input) ? input.signal : null;\n\t\tif ('signal' in init) signal = init.signal;\n\n\t\tif (signal != null && !isAbortSignal(signal)) {\n\t\t\tthrow new TypeError('Expected signal to be an instanceof AbortSignal');\n\t\t}\n\n\t\tthis[INTERNALS$2] = {\n\t\t\tmethod,\n\t\t\tredirect: init.redirect || input.redirect || 'follow',\n\t\t\theaders,\n\t\t\tparsedURL,\n\t\t\tsignal\n\t\t};\n\n\t\t// node-fetch-only options\n\t\tthis.follow = init.follow !== undefined ? init.follow : input.follow !== undefined ? input.follow : 20;\n\t\tthis.compress = init.compress !== undefined ? init.compress : input.compress !== undefined ? input.compress : true;\n\t\tthis.counter = init.counter || input.counter || 0;\n\t\tthis.agent = init.agent || input.agent;\n\t}\n\n\tget method() {\n\t\treturn this[INTERNALS$2].method;\n\t}\n\n\tget url() {\n\t\treturn format_url(this[INTERNALS$2].parsedURL);\n\t}\n\n\tget headers() {\n\t\treturn this[INTERNALS$2].headers;\n\t}\n\n\tget redirect() {\n\t\treturn this[INTERNALS$2].redirect;\n\t}\n\n\tget signal() {\n\t\treturn this[INTERNALS$2].signal;\n\t}\n\n\t/**\n * Clone this request\n *\n * @return Request\n */\n\tclone() {\n\t\treturn new Request(this);\n\t}\n}\n\nBody.mixIn(Request.prototype);\n\nObject.defineProperty(Request.prototype, Symbol.toStringTag, {\n\tvalue: 'Request',\n\twritable: false,\n\tenumerable: false,\n\tconfigurable: true\n});\n\nObject.defineProperties(Request.prototype, {\n\tmethod: { enumerable: true },\n\turl: { enumerable: true },\n\theaders: { enumerable: true },\n\tredirect: { enumerable: true },\n\tclone: { enumerable: true },\n\tsignal: { enumerable: true }\n});\n\n/**\n * Convert a Request to Node.js http request options.\n *\n * @param Request A Request instance\n * @return Object The options object to be passed to http.request\n */\nfunction getNodeRequestOptions(request) {\n\tconst parsedURL = request[INTERNALS$2].parsedURL;\n\tconst headers = new Headers(request[INTERNALS$2].headers);\n\n\t// fetch step 1.3\n\tif (!headers.has('Accept')) {\n\t\theaders.set('Accept', '*/*');\n\t}\n\n\t// Basic fetch\n\tif (!parsedURL.protocol || !parsedURL.hostname) {\n\t\tthrow new TypeError('Only absolute URLs are supported');\n\t}\n\n\tif (!/^https?:$/.test(parsedURL.protocol)) {\n\t\tthrow new TypeError('Only HTTP(S) protocols are supported');\n\t}\n\n\tif (request.signal && request.body instanceof Stream.Readable && !streamDestructionSupported) {\n\t\tthrow new Error('Cancellation of streamed requests with AbortSignal is not supported in node < 8');\n\t}\n\n\t// HTTP-network-or-cache fetch steps 2.4-2.7\n\tlet contentLengthValue = null;\n\tif (request.body == null && /^(POST|PUT)$/i.test(request.method)) {\n\t\tcontentLengthValue = '0';\n\t}\n\tif (request.body != null) {\n\t\tconst totalBytes = getTotalBytes(request);\n\t\tif (typeof totalBytes === 'number') {\n\t\t\tcontentLengthValue = String(totalBytes);\n\t\t}\n\t}\n\tif (contentLengthValue) {\n\t\theaders.set('Content-Length', contentLengthValue);\n\t}\n\n\t// HTTP-network-or-cache fetch step 2.11\n\tif (!headers.has('User-Agent')) {\n\t\theaders.set('User-Agent', 'node-fetch/1.0 (+https://github.com/bitinn/node-fetch)');\n\t}\n\n\t// HTTP-network-or-cache fetch step 2.15\n\tif (request.compress && !headers.has('Accept-Encoding')) {\n\t\theaders.set('Accept-Encoding', 'gzip,deflate');\n\t}\n\n\tlet agent = request.agent;\n\tif (typeof agent === 'function') {\n\t\tagent = agent(parsedURL);\n\t}\n\n\tif (!headers.has('Connection') && !agent) {\n\t\theaders.set('Connection', 'close');\n\t}\n\n\t// HTTP-network fetch step 4.2\n\t// chunked encoding is handled by Node.js\n\n\treturn Object.assign({}, parsedURL, {\n\t\tmethod: request.method,\n\t\theaders: exportNodeCompatibleHeaders(headers),\n\t\tagent\n\t});\n}\n\n/**\n * abort-error.js\n *\n * AbortError interface for cancelled requests\n */\n\n/**\n * Create AbortError instance\n *\n * @param String message Error message for human\n * @return AbortError\n */\nfunction AbortError(message) {\n Error.call(this, message);\n\n this.type = 'aborted';\n this.message = message;\n\n // hide custom error implementation details from end-users\n Error.captureStackTrace(this, this.constructor);\n}\n\nAbortError.prototype = Object.create(Error.prototype);\nAbortError.prototype.constructor = AbortError;\nAbortError.prototype.name = 'AbortError';\n\n// fix an issue where \"PassThrough\", \"resolve\" aren't a named export for node <10\nconst PassThrough$1 = Stream.PassThrough;\nconst resolve_url = Url.resolve;\n\n/**\n * Fetch function\n *\n * @param Mixed url Absolute url or Request instance\n * @param Object opts Fetch options\n * @return Promise\n */\nfunction fetch(url, opts) {\n\n\t// allow custom promise\n\tif (!fetch.Promise) {\n\t\tthrow new Error('native promise missing, set fetch.Promise to your favorite alternative');\n\t}\n\n\tBody.Promise = fetch.Promise;\n\n\t// wrap http.request into fetch\n\treturn new fetch.Promise(function (resolve, reject) {\n\t\t// build request object\n\t\tconst request = new Request(url, opts);\n\t\tconst options = getNodeRequestOptions(request);\n\n\t\tconst send = (options.protocol === 'https:' ? https : http).request;\n\t\tconst signal = request.signal;\n\n\t\tlet response = null;\n\n\t\tconst abort = function abort() {\n\t\t\tlet error = new AbortError('The user aborted a request.');\n\t\t\treject(error);\n\t\t\tif (request.body && request.body instanceof Stream.Readable) {\n\t\t\t\trequest.body.destroy(error);\n\t\t\t}\n\t\t\tif (!response || !response.body) return;\n\t\t\tresponse.body.emit('error', error);\n\t\t};\n\n\t\tif (signal && signal.aborted) {\n\t\t\tabort();\n\t\t\treturn;\n\t\t}\n\n\t\tconst abortAndFinalize = function abortAndFinalize() {\n\t\t\tabort();\n\t\t\tfinalize();\n\t\t};\n\n\t\t// send request\n\t\tconst req = send(options);\n\t\tlet reqTimeout;\n\n\t\tif (signal) {\n\t\t\tsignal.addEventListener('abort', abortAndFinalize);\n\t\t}\n\n\t\tfunction finalize() {\n\t\t\treq.abort();\n\t\t\tif (signal) signal.removeEventListener('abort', abortAndFinalize);\n\t\t\tclearTimeout(reqTimeout);\n\t\t}\n\n\t\tif (request.timeout) {\n\t\t\treq.once('socket', function (socket) {\n\t\t\t\treqTimeout = setTimeout(function () {\n\t\t\t\t\treject(new FetchError(`network timeout at: ${request.url}`, 'request-timeout'));\n\t\t\t\t\tfinalize();\n\t\t\t\t}, request.timeout);\n\t\t\t});\n\t\t}\n\n\t\treq.on('error', function (err) {\n\t\t\treject(new FetchError(`request to ${request.url} failed, reason: ${err.message}`, 'system', err));\n\t\t\tfinalize();\n\t\t});\n\n\t\treq.on('response', function (res) {\n\t\t\tclearTimeout(reqTimeout);\n\n\t\t\tconst headers = createHeadersLenient(res.headers);\n\n\t\t\t// HTTP fetch step 5\n\t\t\tif (fetch.isRedirect(res.statusCode)) {\n\t\t\t\t// HTTP fetch step 5.2\n\t\t\t\tconst location = headers.get('Location');\n\n\t\t\t\t// HTTP fetch step 5.3\n\t\t\t\tconst locationURL = location === null ? null : resolve_url(request.url, location);\n\n\t\t\t\t// HTTP fetch step 5.5\n\t\t\t\tswitch (request.redirect) {\n\t\t\t\t\tcase 'error':\n\t\t\t\t\t\treject(new FetchError(`uri requested responds with a redirect, redirect mode is set to error: ${request.url}`, 'no-redirect'));\n\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\treturn;\n\t\t\t\t\tcase 'manual':\n\t\t\t\t\t\t// node-fetch-specific step: make manual redirect a bit easier to use by setting the Location header value to the resolved URL.\n\t\t\t\t\t\tif (locationURL !== null) {\n\t\t\t\t\t\t\t// handle corrupted header\n\t\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\t\theaders.set('Location', locationURL);\n\t\t\t\t\t\t\t} catch (err) {\n\t\t\t\t\t\t\t\t// istanbul ignore next: nodejs server prevent invalid response headers, we can't test this through normal request\n\t\t\t\t\t\t\t\treject(err);\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\tbreak;\n\t\t\t\t\tcase 'follow':\n\t\t\t\t\t\t// HTTP-redirect fetch step 2\n\t\t\t\t\t\tif (locationURL === null) {\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 5\n\t\t\t\t\t\tif (request.counter >= request.follow) {\n\t\t\t\t\t\t\treject(new FetchError(`maximum redirect reached at: ${request.url}`, 'max-redirect'));\n\t\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 6 (counter increment)\n\t\t\t\t\t\t// Create a new Request object.\n\t\t\t\t\t\tconst requestOpts = {\n\t\t\t\t\t\t\theaders: new Headers(request.headers),\n\t\t\t\t\t\t\tfollow: request.follow,\n\t\t\t\t\t\t\tcounter: request.counter + 1,\n\t\t\t\t\t\t\tagent: request.agent,\n\t\t\t\t\t\t\tcompress: request.compress,\n\t\t\t\t\t\t\tmethod: request.method,\n\t\t\t\t\t\t\tbody: request.body,\n\t\t\t\t\t\t\tsignal: request.signal,\n\t\t\t\t\t\t\ttimeout: request.timeout,\n\t\t\t\t\t\t\tsize: request.size\n\t\t\t\t\t\t};\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 9\n\t\t\t\t\t\tif (res.statusCode !== 303 && request.body && getTotalBytes(request) === null) {\n\t\t\t\t\t\t\treject(new FetchError('Cannot follow redirect with body being a readable stream', 'unsupported-redirect'));\n\t\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 11\n\t\t\t\t\t\tif (res.statusCode === 303 || (res.statusCode === 301 || res.statusCode === 302) && request.method === 'POST') {\n\t\t\t\t\t\t\trequestOpts.method = 'GET';\n\t\t\t\t\t\t\trequestOpts.body = undefined;\n\t\t\t\t\t\t\trequestOpts.headers.delete('content-length');\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// HTTP-redirect fetch step 15\n\t\t\t\t\t\tresolve(fetch(new Request(locationURL, requestOpts)));\n\t\t\t\t\t\tfinalize();\n\t\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// prepare response\n\t\t\tres.once('end', function () {\n\t\t\t\tif (signal) signal.removeEventListener('abort', abortAndFinalize);\n\t\t\t});\n\t\t\tlet body = res.pipe(new PassThrough$1());\n\n\t\t\tconst response_options = {\n\t\t\t\turl: request.url,\n\t\t\t\tstatus: res.statusCode,\n\t\t\t\tstatusText: res.statusMessage,\n\t\t\t\theaders: headers,\n\t\t\t\tsize: request.size,\n\t\t\t\ttimeout: request.timeout,\n\t\t\t\tcounter: request.counter\n\t\t\t};\n\n\t\t\t// HTTP-network fetch step 12.1.1.3\n\t\t\tconst codings = headers.get('Content-Encoding');\n\n\t\t\t// HTTP-network fetch step 12.1.1.4: handle content codings\n\n\t\t\t// in following scenarios we ignore compression support\n\t\t\t// 1. compression support is disabled\n\t\t\t// 2. HEAD request\n\t\t\t// 3. no Content-Encoding header\n\t\t\t// 4. no content response (204)\n\t\t\t// 5. content not modified response (304)\n\t\t\tif (!request.compress || request.method === 'HEAD' || codings === null || res.statusCode === 204 || res.statusCode === 304) {\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// For Node v6+\n\t\t\t// Be less strict when decoding compressed responses, since sometimes\n\t\t\t// servers send slightly invalid responses that are still accepted\n\t\t\t// by common browsers.\n\t\t\t// Always using Z_SYNC_FLUSH is what cURL does.\n\t\t\tconst zlibOptions = {\n\t\t\t\tflush: zlib.Z_SYNC_FLUSH,\n\t\t\t\tfinishFlush: zlib.Z_SYNC_FLUSH\n\t\t\t};\n\n\t\t\t// for gzip\n\t\t\tif (codings == 'gzip' || codings == 'x-gzip') {\n\t\t\t\tbody = body.pipe(zlib.createGunzip(zlibOptions));\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// for deflate\n\t\t\tif (codings == 'deflate' || codings == 'x-deflate') {\n\t\t\t\t// handle the infamous raw deflate response from old servers\n\t\t\t\t// a hack for old IIS and Apache servers\n\t\t\t\tconst raw = res.pipe(new PassThrough$1());\n\t\t\t\traw.once('data', function (chunk) {\n\t\t\t\t\t// see http://stackoverflow.com/questions/37519828\n\t\t\t\t\tif ((chunk[0] & 0x0F) === 0x08) {\n\t\t\t\t\t\tbody = body.pipe(zlib.createInflate());\n\t\t\t\t\t} else {\n\t\t\t\t\t\tbody = body.pipe(zlib.createInflateRaw());\n\t\t\t\t\t}\n\t\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\t\tresolve(response);\n\t\t\t\t});\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// for br\n\t\t\tif (codings == 'br' && typeof zlib.createBrotliDecompress === 'function') {\n\t\t\t\tbody = body.pipe(zlib.createBrotliDecompress());\n\t\t\t\tresponse = new Response(body, response_options);\n\t\t\t\tresolve(response);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// otherwise, use response as-is\n\t\t\tresponse = new Response(body, response_options);\n\t\t\tresolve(response);\n\t\t});\n\n\t\twriteToStream(req, request);\n\t});\n}\n/**\n * Redirect code matching\n *\n * @param Number code Status code\n * @return Boolean\n */\nfetch.isRedirect = function (code) {\n\treturn code === 301 || code === 302 || code === 303 || code === 307 || code === 308;\n};\n\n// expose Promise\nfetch.Promise = global.Promise;\n\nexport default fetch;\nexport { Headers, Request, Response, FetchError };\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Conv2DInfo, Conv3DInfo} from '../ops/conv_util';\nimport {FusedBatchMatMulConfig, FusedConv2DConfig} from '../ops/fused_types';\nimport {Backend, DataId, Scalar, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D} from '../tensor';\nimport {BackendValues, DataType, Rank, ShapeMap} from '../types';\n\nexport const EPSILON_FLOAT32 = 1e-7;\nexport const EPSILON_FLOAT16 = 1e-4;\n\n// Required information for all backends.\nexport interface BackendTimingInfo {\n kernelMs: number|{error: string};\n getExtraProfileInfo?(): string; // a field for additional timing information\n // e.g. packing / unpacking for WebGL backend\n}\n\nexport interface TensorStorage {\n read(dataId: DataId): Promise;\n readSync(dataId: DataId): BackendValues;\n disposeData(dataId: DataId): void;\n write(values: BackendValues, shape: number[], dtype: DataType): DataId;\n move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n void;\n memory(): {unreliable: boolean;}; // Backend-specific information.\n /** Returns number of data ids currently in the storage. */\n numDataIds(): number;\n}\n\n/** Convenient class for storing tensor-related data. */\nexport class DataStorage {\n private data = new WeakMap();\n private dataIdsCount = 0;\n\n constructor(private backend: KernelBackend, private dataMover: DataMover) {}\n\n get(dataId: DataId) {\n if (!this.data.has(dataId)) {\n this.dataMover.moveData(this.backend, dataId);\n }\n return this.data.get(dataId);\n }\n\n set(dataId: DataId, value: T): void {\n this.dataIdsCount++;\n this.data.set(dataId, value);\n }\n\n has(dataId: DataId): boolean {\n return this.data.has(dataId);\n }\n\n delete(dataId: DataId): boolean {\n this.dataIdsCount--;\n return this.data.delete(dataId);\n }\n\n numDataIds(): number {\n return this.dataIdsCount;\n }\n}\n\nexport interface DataMover {\n /**\n * To be called by backends whenever they see a dataId that they don't own.\n * Upon calling this method, the mover will fetch the tensor from another\n * backend and register it with the current active backend.\n */\n moveData(backend: KernelBackend, dataId: DataId): void;\n}\n\nexport interface BackendTimer {\n time(f: () => void): Promise;\n}\n\n/**\n * The interface that defines the kernels that should be implemented when\n * adding a new backend. New backends don't need to implement every one of the\n * methods, this can be done gradually (throw an error for unimplemented\n * methods).\n */\nexport class KernelBackend implements TensorStorage, Backend, BackendTimer {\n time(f: () => void): Promise {\n return notYetImplemented('time');\n }\n read(dataId: object): Promise {\n return notYetImplemented('read');\n }\n readSync(dataId: object): BackendValues {\n return notYetImplemented('readSync');\n }\n numDataIds(): number {\n return notYetImplemented('numDataIds');\n }\n disposeData(dataId: object): void {\n return notYetImplemented('disposeData');\n }\n write(values: BackendValues, shape: number[], dtype: DataType): DataId {\n return notYetImplemented('write');\n }\n move(dataId: DataId, values: BackendValues, shape: number[], dtype: DataType):\n void {\n return notYetImplemented('move');\n }\n memory(): {unreliable: boolean; reasons?: string[]} {\n return notYetImplemented('memory');\n }\n /** Returns the highest precision for floats in bits (e.g. 16 or 32) */\n floatPrecision(): 16|32 {\n return notYetImplemented('floatPrecision');\n }\n /** Returns the smallest representable number. */\n epsilon(): number {\n return this.floatPrecision() === 32 ? EPSILON_FLOAT32 : EPSILON_FLOAT16;\n }\n\n batchMatMul(\n a: Tensor3D, b: Tensor3D, transposeA: boolean,\n transposeB: boolean): Tensor3D {\n return notYetImplemented('batchMatMul');\n }\n\n fusedBatchMatMul(\n {a, b, transposeA, transposeB, bias, activation, preluActivationWeights}:\n FusedBatchMatMulConfig): Tensor3D {\n return notYetImplemented('fusedBatchMatMul');\n }\n\n slice(x: T, begin: number[], size: number[]): T {\n return notYetImplemented('slice');\n }\n stridedSlice(\n x: T, begin: number[], end: number[], strides: number[]): T {\n return notYetImplemented('stridedSlice');\n }\n unstack(x: Tensor, axis: number): Tensor[] {\n return notYetImplemented('unstack');\n }\n reverse(a: T, axis: number[]): T {\n return notYetImplemented('reverse');\n }\n\n concat(tensors: Tensor[], axis: number): Tensor {\n return notYetImplemented('concat');\n }\n\n neg(a: T): T {\n return notYetImplemented('neg');\n }\n\n add(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('add');\n }\n addN(tensors: T[]): T {\n return notYetImplemented('addN');\n }\n subtract(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('subtract');\n }\n multiply(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('multiply');\n }\n realDivide(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('realDivide');\n }\n floorDiv(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('floorDiv');\n }\n\n sum(x: Tensor, axes: number[]): Tensor {\n return notYetImplemented('sum');\n }\n prod(x: Tensor, axes: number[]): Tensor {\n return notYetImplemented('prod');\n }\n\n unsortedSegmentSum(\n x: T, segmentIds: Tensor1D, numSegments: number): Tensor {\n return notYetImplemented('unsortedSegmentSum');\n }\n\n argMin(x: Tensor, axis: number): Tensor {\n return notYetImplemented('argMin');\n }\n argMax(x: Tensor, axis: number): Tensor {\n return notYetImplemented('argMax');\n }\n\n equal(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('equal');\n }\n notEqual(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('notEqual');\n }\n\n less(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('less');\n }\n lessEqual(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('lessEqual');\n }\n\n greater(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('greater');\n }\n greaterEqual(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('greaterEqual');\n }\n\n logicalNot(a: T): T {\n return notYetImplemented('logicalNot');\n }\n logicalAnd(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('logicalAnd');\n }\n logicalOr(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('logicalOr');\n }\n\n where(condition: Tensor): Tensor2D {\n return notYetImplemented('where');\n }\n select(condition: Tensor, a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('select');\n }\n\n topk(x: T, k: number, sorted: boolean): [T, T] {\n return notYetImplemented('topk');\n }\n\n min(x: Tensor, axes: number[]): Tensor {\n return notYetImplemented('min');\n }\n minimum(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('minimum');\n }\n\n mod(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('mod');\n }\n\n max(x: Tensor, axes: number[]): Tensor {\n return notYetImplemented('max');\n }\n maximum(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('maximum');\n }\n\n all(x: Tensor, axes: number[]): Tensor {\n return notYetImplemented('all');\n }\n any(x: Tensor, axes: number[]): Tensor {\n return notYetImplemented('any');\n }\n\n squaredDifference(a: Tensor, b: Tensor): Tensor {\n return notYetImplemented('squaredDifference');\n }\n\n ceil(x: T): T {\n return notYetImplemented('ceil');\n }\n floor(x: T): T {\n return notYetImplemented('floor');\n }\n round(x: T): T {\n return notYetImplemented('round');\n }\n\n sign(x: T): T {\n return notYetImplemented('sign');\n }\n\n isNaN(x: T): T {\n return notYetImplemented('isNaN');\n }\n isInf(x: T): T {\n return notYetImplemented('isInf');\n }\n isFinite(x: T): T {\n return notYetImplemented('isFinite');\n }\n\n pow(a: T, b: Tensor): T {\n return notYetImplemented('pow');\n }\n exp(x: T): T {\n return notYetImplemented('exp');\n }\n expm1(x: T): T {\n return notYetImplemented('expm1');\n }\n softmax(x: T, dim: number): T {\n return notYetImplemented('softmax');\n }\n log(x: T): T {\n return notYetImplemented('log');\n }\n log1p(x: T): T {\n return notYetImplemented('log1p');\n }\n sqrt(x: T): T {\n return notYetImplemented('sqrt');\n }\n rsqrt(x: T): T {\n return notYetImplemented('rsqrt');\n }\n square(x: T): T {\n return notYetImplemented('square');\n }\n reciprocal(x: T): T {\n return notYetImplemented('reciprocal');\n }\n relu(x: T): T {\n return notYetImplemented('relu');\n }\n relu6(x: T): T {\n return notYetImplemented('relu6');\n }\n prelu(x: T, a: T): T {\n return notYetImplemented('prelu');\n }\n elu(x: T): T {\n return notYetImplemented('elu');\n }\n eluDer(dy: T, y: T): T {\n return notYetImplemented('eluDer');\n }\n selu(x: T): T {\n return notYetImplemented('selu');\n }\n int(x: T): T {\n return notYetImplemented('int');\n }\n\n clip(x: T, min: number, max: number): T {\n return notYetImplemented('clip');\n }\n\n abs(x: T): T {\n return notYetImplemented('abs');\n }\n complexAbs(x: T): T {\n return notYetImplemented('complexAbs');\n }\n\n sigmoid(x: T): T {\n return notYetImplemented('sigmoid');\n }\n\n softplus(x: T): T {\n return notYetImplemented('softplus');\n }\n\n sin(x: T): T {\n return notYetImplemented('sin');\n }\n cos(x: T): T {\n return notYetImplemented('cos');\n }\n tan(x: T): T {\n return notYetImplemented('tan');\n }\n\n asin(x: T): T {\n return notYetImplemented('asin');\n }\n acos(x: T): T {\n return notYetImplemented('acos');\n }\n atan(x: T): T {\n return notYetImplemented('atan');\n }\n atan2(a: T, b: T): T {\n return notYetImplemented('atan2');\n }\n\n sinh(x: T): T {\n return notYetImplemented('sinh');\n }\n cosh(x: T): T {\n return notYetImplemented('cosh');\n }\n tanh(x: T): T {\n return notYetImplemented('tanh');\n }\n\n asinh(x: T): T {\n return notYetImplemented('asinh');\n }\n acosh(x: T): T {\n return notYetImplemented('acosh');\n }\n atanh(x: T): T {\n return notYetImplemented('atanh');\n }\n\n erf(x: T): T {\n return notYetImplemented('erf');\n }\n\n step(x: T, alpha: number): T {\n return notYetImplemented('step');\n }\n\n fusedConv2d(\n {input, filter, convInfo, bias, activation, preluActivationWeights}:\n FusedConv2DConfig): Tensor4D {\n return notYetImplemented('fusedConv2d');\n }\n\n conv2d(x: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n return notYetImplemented('conv2d');\n }\n conv2dDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n Tensor4D {\n return notYetImplemented('conv2dDerInput');\n }\n conv2dDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n return notYetImplemented('conv2dDerFilter');\n }\n\n fusedDepthwiseConv2D(\n {input, filter, convInfo, bias, activation, preluActivationWeights}:\n FusedConv2DConfig): Tensor4D {\n return notYetImplemented('fusedDepthwiseConv2D');\n }\n\n depthwiseConv2D(input: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n Tensor4D {\n return notYetImplemented('depthwiseConv2D');\n }\n depthwiseConv2DDerInput(dy: Tensor4D, filter: Tensor4D, convInfo: Conv2DInfo):\n Tensor4D {\n return notYetImplemented('depthwiseConv2DDerInput');\n }\n depthwiseConv2DDerFilter(x: Tensor4D, dY: Tensor4D, convInfo: Conv2DInfo):\n Tensor4D {\n return notYetImplemented('depthwiseConv2DDerFilter');\n }\n conv3d(x: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n return notYetImplemented('conv3d');\n }\n conv3dDerInput(dy: Tensor5D, filter: Tensor5D, convInfo: Conv3DInfo):\n Tensor5D {\n return notYetImplemented('conv3dDerInput');\n }\n conv3dDerFilter(x: Tensor5D, dY: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n return notYetImplemented('conv3dDerFilter');\n }\n maxPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n return notYetImplemented('maxPool');\n }\n maxPoolBackprop(dy: Tensor4D, x: Tensor4D, y: Tensor4D, convInfo: Conv2DInfo):\n Tensor4D {\n return notYetImplemented('maxPoolBackprop');\n }\n avgPool(x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n return notYetImplemented('avgPool');\n }\n avgPoolBackprop(dy: Tensor4D, x: Tensor4D, convInfo: Conv2DInfo): Tensor4D {\n return notYetImplemented('avgPoolBackprop');\n }\n avgPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n return notYetImplemented('avgPool3d');\n }\n avgPool3dBackprop(dy: Tensor5D, x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n return notYetImplemented('avgPool3dBackprop');\n }\n maxPool3d(x: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n return notYetImplemented('maxPool3d');\n }\n maxPool3dBackprop(\n dy: Tensor5D, x: Tensor5D, y: Tensor5D, convInfo: Conv3DInfo): Tensor5D {\n return notYetImplemented('maxPool3dBackprop');\n }\n\n reshape(x: T, shape: ShapeMap[R]):\n Tensor {\n return notYetImplemented('reshape');\n }\n cast(x: T, dtype: DataType): T {\n return notYetImplemented('cast');\n }\n\n tile(x: T, reps: number[]): T {\n return notYetImplemented('tile');\n }\n\n pad(\n x: T, paddings: Array<[number, number]>, constantValue: number): T {\n return notYetImplemented('pad');\n }\n\n transpose(x: T, perm: number[]): T {\n return notYetImplemented('transpose');\n }\n\n gather(x: T, indices: Tensor1D, axis: number): T {\n return notYetImplemented('gather');\n }\n\n gatherND(x: Tensor, indices: Tensor): Tensor {\n return notYetImplemented('gatherND');\n }\n\n scatterND(\n indices: Tensor, updates: Tensor, shape: ShapeMap[R]): Tensor {\n return notYetImplemented('scatterND');\n }\n\n batchToSpaceND(\n x: T, blockShape: number[], crops: number[][]): T {\n return notYetImplemented('batchToSpaceND');\n }\n\n spaceToBatchND(\n x: T, blockShape: number[], paddings: number[][]): T {\n return notYetImplemented('spaceToBatchND');\n }\n\n resizeBilinear(\n x: Tensor4D, newHeight: number, newWidth: number,\n alignCorners: boolean): Tensor4D {\n return notYetImplemented('resizeBilinear');\n }\n\n resizeBilinearBackprop(dy: Tensor4D, x: Tensor4D, alignCorners: boolean):\n Tensor4D {\n return notYetImplemented('resizeBilinearBackprop');\n }\n\n resizeNearestNeighbor(\n x: Tensor4D, newHEight: number, newWidth: number,\n alignCorners: boolean): Tensor4D {\n return notYetImplemented('resizeNearestNeighbor');\n }\n\n resizeNearestNeighborBackprop(\n dy: Tensor4D, x: Tensor4D, alignCorners: boolean): Tensor4D {\n return notYetImplemented('resizeNearestNeighborBackprop');\n }\n\n batchNorm(\n x: Tensor4D, mean: Tensor4D|Tensor1D, variance: Tensor4D|Tensor1D,\n offset?: Tensor4D|Tensor1D, scale?: Tensor4D|Tensor1D,\n varianceEpsilon?: number): Tensor4D {\n return notYetImplemented('batchNorm');\n }\n\n localResponseNormalization4D(\n x: Tensor4D, radius: number, bias: number, alpha: number,\n beta: number): Tensor4D {\n return notYetImplemented('localResponseNormalization4D');\n }\n\n LRNGrad(\n dy: Tensor4D, inputImage: Tensor4D, outputImage: Tensor4D, radius: number,\n bias: number, alpha: number, beta: number): Tensor4D {\n return notYetImplemented('LRNGrad');\n }\n\n multinomial(\n logits: Tensor2D, normalized: boolean, numSamples: number,\n seed: number): Tensor2D {\n return notYetImplemented('multinomial');\n }\n\n oneHot(indices: Tensor1D, depth: number, onValue: number, offValue: number):\n Tensor2D {\n return notYetImplemented('oneHot');\n }\n\n cumsum(x: Tensor, axis: number, exclusive: boolean, reverse: boolean):\n Tensor {\n return notYetImplemented('cumsum');\n }\n\n nonMaxSuppression(\n boxes: Tensor2D, scores: Tensor1D, maxOutputSize: number,\n iouThreshold: number, scoreThreshold?: number): Tensor1D {\n return notYetImplemented('nonMaxSuppression');\n }\n\n fft(x: Tensor2D): Tensor2D {\n return notYetImplemented('fft');\n }\n ifft(x: Tensor2D): Tensor2D {\n return notYetImplemented('ifft');\n }\n complex(real: T, imag: T): T {\n return notYetImplemented('complex');\n }\n real(input: T): T {\n return notYetImplemented('real');\n }\n imag(input: T): T {\n return notYetImplemented('imag');\n }\n\n cropAndResize(\n image: Tensor4D, boxes: Tensor2D, boxIndex: Tensor1D,\n cropSize: [number, number], method: 'bilinear'|'nearest',\n extrapolationValue: number): Tensor4D {\n return notYetImplemented('cropAndResize');\n }\n\n depthToSpace(x: Tensor4D, blockSize: number, dataFormat: string): Tensor4D {\n return notYetImplemented('depthToSpace');\n }\n\n // Aligns with the \"SplitV\" kernel in TensorFlow.\n split(value: T, sizeSplits: number[], axis: number): T[] {\n return notYetImplemented('split');\n }\n\n sparseToDense(\n sparseIndices: Tensor, sparseValues: Tensor, outputShape: ShapeMap[R],\n defaultValue: Scalar): Tensor {\n return notYetImplemented('sparseToDense');\n }\n\n diag(x: Tensor): Tensor {\n return notYetImplemented('diag');\n }\n\n fill(\n shape: ShapeMap[R], value: number|string, dtype?: DataType): Tensor {\n return notYetImplemented('fill');\n }\n\n onesLike(x: Tensor): Tensor {\n return notYetImplemented('onesLike');\n }\n\n zerosLike(x: Tensor): Tensor {\n return notYetImplemented('zerosLike');\n }\n\n linspace(start: number, stop: number, num: number): Tensor1D {\n return notYetImplemented('linspace');\n }\n\n dispose(): void {\n return notYetImplemented('dispose');\n }\n}\n\nfunction notYetImplemented(kernelName: string): never {\n throw new Error(\n `'${kernelName}' not yet implemented or not found in the registry. ` +\n `This kernel may not be supported by the tfjs backend you have chosen`);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Platform} from './platforms/platform';\n\n// Expects flags from URL in the format ?tfjsflags=FLAG1:1,FLAG2:true.\nconst TENSORFLOWJS_FLAGS_PREFIX = 'tfjsflags';\n\ntype FlagValue = number|boolean;\ntype FlagEvaluationFn = (() => FlagValue)|(() => Promise);\nexport type Flags = {\n [featureName: string]: FlagValue\n};\nexport type FlagRegistryEntry = {\n evaluationFn: FlagEvaluationFn;\n setHook?: (value: FlagValue) => void;\n};\n\n/**\n * The environment contains evaluated flags as well as the registered platform.\n * This is always used as a global singleton and can be retrieved with\n * `tf.env()`.\n *\n * @doc {heading: 'Environment'}\n */\nexport class Environment {\n private flags: Flags = {};\n private flagRegistry: {[flagName: string]: FlagRegistryEntry} = {};\n\n private urlFlags: Flags = {};\n\n platformName: string;\n platform: Platform;\n\n // tslint:disable-next-line: no-any\n constructor(public global: any) {\n this.populateURLFlags();\n }\n\n setPlatform(platformName: string, platform: Platform) {\n if (this.platform != null) {\n console.warn(\n `Platform ${this.platformName} has already been set. ` +\n `Overwriting the platform with ${platform}.`);\n }\n this.platformName = platformName;\n this.platform = platform;\n }\n\n registerFlag(\n flagName: string, evaluationFn: FlagEvaluationFn,\n setHook?: (value: FlagValue) => void) {\n this.flagRegistry[flagName] = {evaluationFn, setHook};\n\n // Override the flag value from the URL. This has to happen here because the\n // environment is initialized before flags get registered.\n if (this.urlFlags[flagName] != null) {\n const flagValue = this.urlFlags[flagName];\n console.warn(\n `Setting feature override from URL ${flagName}: ${flagValue}.`);\n this.set(flagName, flagValue);\n }\n }\n\n async getAsync(flagName: string): Promise {\n if (flagName in this.flags) {\n return this.flags[flagName];\n }\n\n this.flags[flagName] = await this.evaluateFlag(flagName);\n return this.flags[flagName];\n }\n\n get(flagName: string): FlagValue {\n if (flagName in this.flags) {\n return this.flags[flagName];\n }\n\n const flagValue = this.evaluateFlag(flagName);\n if (flagValue instanceof Promise) {\n throw new Error(\n `Flag ${flagName} cannot be synchronously evaluated. ` +\n `Please use getAsync() instead.`);\n }\n\n this.flags[flagName] = flagValue;\n\n return this.flags[flagName];\n }\n\n getNumber(flagName: string): number {\n return this.get(flagName) as number;\n }\n\n getBool(flagName: string): boolean {\n return this.get(flagName) as boolean;\n }\n\n getFlags(): Flags {\n return this.flags;\n }\n // For backwards compatibility.\n get features(): Flags {\n return this.flags;\n }\n\n set(flagName: string, value: FlagValue): void {\n if (this.flagRegistry[flagName] == null) {\n throw new Error(\n `Cannot set flag ${flagName} as it has not been registered.`);\n }\n this.flags[flagName] = value;\n if (this.flagRegistry[flagName].setHook != null) {\n this.flagRegistry[flagName].setHook(value);\n }\n }\n\n private evaluateFlag(flagName: string): FlagValue|Promise {\n if (this.flagRegistry[flagName] == null) {\n throw new Error(\n `Cannot evaluate flag '${flagName}': no evaluation function found.`);\n }\n return this.flagRegistry[flagName].evaluationFn();\n }\n\n setFlags(flags: Flags) {\n this.flags = Object.assign({}, flags);\n }\n\n reset() {\n this.flags = {};\n this.urlFlags = {};\n this.populateURLFlags();\n }\n\n private populateURLFlags(): void {\n if (typeof this.global === 'undefined' ||\n typeof this.global.location === 'undefined' ||\n typeof this.global.location.search === 'undefined') {\n return;\n }\n\n const urlParams = getQueryParams(this.global.location.search);\n if (TENSORFLOWJS_FLAGS_PREFIX in urlParams) {\n const keyValues = urlParams[TENSORFLOWJS_FLAGS_PREFIX].split(',');\n keyValues.forEach(keyValue => {\n const [key, value] = keyValue.split(':') as [string, string];\n this.urlFlags[key] = parseValue(key, value);\n });\n }\n }\n}\n\nexport function getQueryParams(queryString: string): {[key: string]: string} {\n const params = {};\n queryString.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g, (s, ...t) => {\n decodeParam(params, t[0], t[1]);\n return t.join('=');\n });\n return params;\n}\n\nfunction decodeParam(\n params: {[key: string]: string}, name: string, value?: string) {\n params[decodeURIComponent(name)] = decodeURIComponent(value || '');\n}\n\nfunction parseValue(flagName: string, value: string): FlagValue {\n value = value.toLowerCase();\n if (value === 'true' || value === 'false') {\n return value === 'true';\n } else if (`${+ value}` === value) {\n return +value;\n }\n throw new Error(\n `Could not parse value flag value ${value} for flag ${flagName}.`);\n}\n\n/**\n * Returns the current environment (a global singleton).\n *\n * The environment object contains the evaluated feature values as well as the\n * active platform.\n *\n * @doc {heading: 'Environment'}\n */\nexport function env() {\n return ENV;\n}\n\nexport let ENV: Environment = null;\nexport function setEnvironmentGlobal(environment: Environment) {\n ENV = environment;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Note that the identifier globalNameSpace is scoped to this module, but will\n// always resolve to the same global object regardless of how the module is\n// resolved.\n// tslint:disable-next-line:no-any\nlet globalNameSpace: {_tfGlobals: Map};\n// tslint:disable-next-line:no-any\nexport function getGlobalNamespace(): {_tfGlobals: Map} {\n if (globalNameSpace == null) {\n // tslint:disable-next-line:no-any\n let ns: any;\n if (typeof (window) !== 'undefined') {\n ns = window;\n } else if (typeof (global) !== 'undefined') {\n ns = global;\n } else if (typeof (process) !== 'undefined') {\n ns = process;\n } else if (typeof (self) !== 'undefined') {\n ns = self;\n } else {\n throw new Error('Could not find a global object');\n }\n globalNameSpace = ns;\n }\n return globalNameSpace;\n}\n\n// tslint:disable-next-line:no-any\nfunction getGlobalMap(): Map {\n const ns = getGlobalNamespace();\n if (ns._tfGlobals == null) {\n ns._tfGlobals = new Map();\n }\n return ns._tfGlobals;\n}\n\n/**\n * Returns a globally accessible 'singleton' object.\n *\n * @param key the name of the object\n * @param init a function to initialize to initialize this object\n * the first time it is fetched.\n */\nexport function getGlobal(key: string, init: () => T): T {\n const globalMap = getGlobalMap();\n if (globalMap.has(key)) {\n return globalMap.get(key);\n } else {\n const singleton = init();\n globalMap.set(key, singleton);\n return globalMap.get(key);\n }\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n// Allow UpperCamelCase variable names\n// tslint:disable: variable-name\n// Unfortunately just enabling PascalCase per file (tslint:enable:\n// allow-pascal-case) doesn't work.\nimport {NamedTensorInfoMap, TensorInfo} from './kernel_registry';\nimport {ExplicitPadding} from './ops/conv_util';\nimport {Activation} from './ops/fused_types';\nimport {DataType, PixelData} from './types';\n\nexport const Abs = 'Abs';\nexport type AbsInputs = UnaryInputs;\n\nexport const Acos = 'Acos';\nexport type AcosInputs = UnaryInputs;\n\nexport const Acosh = 'Acosh';\nexport type AcoshInputs = UnaryInputs;\n\nexport const Add = 'Add';\nexport type AddInputs = BinaryInputs;\n\nexport const AddN = 'AddN';\nexport type AddNInputs = TensorInfo[];\n\nexport const All = 'All';\nexport type AllInputs = Pick;\nexport interface AllAttrs {\n axis: number|number[];\n keepDims: boolean;\n}\n\nexport const Any = 'Any';\nexport type AnyInputs = Pick;\nexport interface AnyAttrs {\n axis: number|number[];\n keepDims: boolean;\n}\n\nexport const ArgMax = 'ArgMax';\nexport type ArgMaxInputs = Pick;\nexport interface ArgMaxAttrs {\n axis: number;\n}\n\nexport const ArgMin = 'ArgMin';\nexport type ArgMinInputs = Pick;\nexport interface ArgMinAttrs {\n axis: number;\n}\n\nexport const Asin = 'Asin';\nexport type AsinInputs = UnaryInputs;\n\nexport const Asinh = 'Asinh';\nexport type AsinhInputs = UnaryInputs;\n\nexport const Atan = 'Atan';\nexport type AtanInputs = UnaryInputs;\n\nexport const Atanh = 'Atanh';\nexport type AtanhInputs = UnaryInputs;\n\nexport const Atan2 = 'Atan2';\nexport type Atan2Inputs = BinaryInputs;\n\nexport const AvgPool = 'AvgPool';\nexport type AvgPoolInputs = Pick;\nexport interface AvgPoolAttrs {\n filterSize: [number, number]|number;\n strides: [number, number]|number;\n pad: 'valid'|'same'|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const AvgPoolBackprop = 'AvgPoolBackprop';\nexport type AvgPoolBackpropInputs = Pick;\nexport interface AvgPoolBackpropAttrs {\n filterSize: [number, number]|number;\n strides: [number, number]|number;\n pad: 'valid'|'same'|number;\n}\n\nexport const AvgPool3D = 'AvgPool3D';\nexport type AvgPool3DInputs = Pick;\nexport interface AvgPool3DAttrs {\n filterSize: [number, number, number]|number;\n strides: [number, number, number]|number;\n pad: 'valid'|'same'|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n dataFormat: 'NDHWC'|'NCDHW';\n dilations?: [number, number, number]|number;\n}\n\nexport const AvgPool3DBackprop = 'AvgPool3DBackprop';\nexport type AvgPool3DBackpropInputs = Pick;\nexport interface AvgPool3DBackpropAttrs {\n filterSize: [number, number, number]|number;\n strides: [number, number, number]|number;\n pad: 'valid'|'same'|number;\n dilations: [number, number, number]|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const BatchMatMul = 'BatchMatMul';\nexport type BatchMatMulInputs = Pick;\nexport interface BatchMatMulAttrs {\n transposeA: boolean;\n transposeB: boolean;\n}\n\nexport const BatchToSpaceND = 'BatchToSpaceND';\nexport type BatchToSpaceNDInputs = Pick;\nexport interface BatchToSpaceNDAttrs {\n blockShape: number[];\n crops: number[][];\n}\n\nexport type BinaryInputs = Pick;\n\nexport const BroadcastTo = 'BroadcastTo';\nexport type BroadcastToInputs = Pick;\nexport interface BroadCastToAttrs {\n shape: number[];\n inputShape: number[]; // for gradient\n}\n\nexport const Cast = 'Cast';\nexport type CastInputs = UnaryInputs;\nexport interface CastAttrs {\n dtype: DataType;\n}\n\nexport const Ceil = 'Ceil';\nexport type CeilInputs = UnaryInputs;\n\nexport const ClipByValue = 'ClipByValue';\nexport type ClipByValueInputs = UnaryInputs;\nexport interface ClipByValueAttrs {\n clipValueMin: number;\n clipValueMax: number;\n}\n\nexport const Complex = 'Complex';\nexport type ComplexInputs = Pick;\n\nexport const Concat = 'Concat';\nexport type ConcatInputs = TensorInfo[];\nexport interface ConcatAttrs {\n axis: number;\n}\n\nexport const Conv2D = 'Conv2D';\nexport type Conv2DInputs = Pick;\nexport interface Conv2DAttrs {\n strides: [number, number]|number;\n pad: 'valid'|'same'|number|ExplicitPadding;\n dataFormat: 'NHWC'|'NCHW';\n dilations: [number, number]|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropFilter = 'Conv2DBackpropFilter';\nexport type Conv2DBackpropFilterInputs = Pick;\nexport interface Conv2DBackpropFilterAttrs {\n strides: [number, number]|number;\n pad: 'valid'|'same'|number|ExplicitPadding;\n dataFormat: 'NHWC'|'NCHW';\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const Conv2DBackpropInput = 'Conv2DBackpropInput';\nexport type Conv2DBackpropInputInputs = Pick;\nexport interface Conv2DBackpropInputAttrs {\n strides: [number, number]|number;\n pad: 'valid'|'same'|number|ExplicitPadding;\n dataFormat: 'NHWC'|'NCHW';\n dimRoundingMode?: 'floor'|'round'|'ceil';\n inputShape: [number, number, number, number];\n}\n\nexport const Conv3D = 'Conv3D';\nexport type Conv3DInputs = Pick;\nexport interface Conv3DAttrs {\n strides: [number, number, number]|number;\n pad: 'valid'|'same';\n dataFormat: 'NDHWC'|'NCDHW';\n dilations: [number, number, number]|number;\n}\n\nexport const Conv3DBackpropFilterV2 = 'Conv3DBackpropFilterV2';\nexport type Conv3DBackpropFilterInputs = Pick;\n\nexport interface Conv3DBackpropFilterAttrs {\n strides: [number, number, number]|number;\n pad: 'valid'|'same';\n}\n\nexport const Conv3DBackpropInputV2 = 'Conv3DBackpropInputV2';\nexport type Conv3DBackpropInputInputs = Pick;\nexport interface Conv3DBackpropInputAttrs {\n pad: 'valid'|'same';\n}\n\nexport const Cos = 'Cos';\nexport type CosInputs = UnaryInputs;\n\nexport const Cosh = 'Cosh';\nexport type CoshInputs = UnaryInputs;\n\nexport const Cumsum = 'Cumsum';\nexport type CumsumInputs = Pick;\nexport interface CumsumAttrs {\n axis: number;\n exclusive: boolean;\n reverse: boolean;\n}\n\nexport const CropAndResize = 'CropAndResize';\nexport type CropAndResizeInputs =\n Pick;\nexport interface CropAndResizeAttrs {\n cropSize: [number, number];\n method: 'bilinear'|'nearest';\n extrapolationValue: number;\n}\n\nexport const DepthToSpace = 'DepthToSpace';\nexport type DepthToSpaceInputs = Pick;\nexport interface DepthToSpaceAttrs {\n blockSize: number;\n dataFormat: 'NHWC'|'NCHW';\n}\n\nexport const DepthwiseConv2dNative = 'DepthwiseConv2dNative';\nexport type DepthwiseConv2dNativeInputs =\n Pick;\nexport interface DepthwiseConv2dNativeAttrs {\n strides: [number, number]|number;\n pad: 'valid'|'same'|number;\n dataFormat: 'NHWC'|'NCHW';\n dilations: [number, number]|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const DepthwiseConv2dNativeBackpropFilter =\n 'DepthwiseConv2dNativeBackpropFilter';\nexport type DepthwiseConv2dNativeBackpropFilterInputs =\n Pick;\n\nexport const DepthwiseConv2dNativeBackpropInput =\n 'DepthwiseConv2dNativeBackpropInput';\nexport type DepthwiseConv2dNativeBackpropInputInputs =\n Pick;\n\nexport const Diag = 'Diag';\nexport type DiagInputs = Pick;\n\nexport const Dilation2D = 'Dilation2D';\nexport type Dilation2DInputs = Pick;\nexport interface Dilation2DAttrs {\n strides: [number, number]|number;\n pad: 'valid'|'same'|number;\n dilations: [number, number]|number;\n}\n\nexport const Dilation2DBackpropInput = 'Dilation2DBackpropInput';\nexport type Dilation2DBackpropInputInputs =\n Pick;\n\nexport const Dilation2DBackpropFilter = 'Dilation2DBackpropFilter';\nexport type Dilation2DBackpropFilterInputs =\n Pick;\n\nexport const Div = 'Div';\nexport type DivInputs = BinaryInputs;\n\nexport const Elu = 'Elu';\nexport type EluInputs = Pick;\n\nexport const EluGrad = 'EluGrad';\nexport type EluGradInputs = Pick;\n\nexport const Erf = 'Erf';\nexport type ErfInputs = UnaryInputs;\n\nexport const Equal = 'Equal';\nexport type EqualInputs = BinaryInputs;\n\nexport const Exp = 'Exp';\nexport type ExpInputs = UnaryInputs;\n\nexport const Expm1 = 'Expm1';\nexport type Expm1Inputs = UnaryInputs;\n\nexport const FFT = 'FFT';\nexport type FFTInputs = Pick;\n\nexport const Fill = 'Fill';\nexport interface FillAttrs {\n shape: number[];\n value: number|string;\n dtype: DataType;\n}\n\nexport const FlipLeftRight = 'FlipLeftRight';\nexport type FlipLeftRightInputs = Pick;\n\nexport const Floor = 'Floor';\nexport type FloorInputs = UnaryInputs;\n\nexport const FloorDiv = 'FloorDiv';\nexport type FloorDivInputs = BinaryInputs;\n\nexport const FusedBatchNorm = 'FusedBatchNorm';\nexport type FusedBatchNormInputs =\n Pick;\nexport interface FusedBatchNormAttrs {\n varianceEpsilon: number;\n}\n\nexport const GatherV2 = 'GatherV2';\nexport type GatherV2Inputs = Pick;\nexport interface GatherV2Attrs {\n axis: number;\n}\n\nexport const GatherNd = 'GatherNd';\nexport type GatherNdInputs = Pick;\n\nexport const Greater = 'Greater';\nexport type GreaterInputs = BinaryInputs;\n\nexport const GreaterEqual = 'GreaterEqual';\nexport type GreaterEqualInputs = BinaryInputs;\n\nexport const Identity = 'Identity';\nexport type IdentityInputs = Pick;\n\nexport const IFFT = 'IFFT';\nexport type IFFTInputs = Pick;\n\nexport const Imag = 'Imag';\nexport type ImagInputs = Pick;\n\nexport const IsFinite = 'IsFinite';\nexport type IsFiniteInputs = UnaryInputs;\n\nexport const IsInf = 'IsInf';\nexport type IsInfInputs = UnaryInputs;\n\nexport const IsNan = 'IsNan';\nexport type IsNanInputs = UnaryInputs;\n\nexport const Less = 'Less';\nexport type LessInputs = BinaryInputs;\n\nexport const LessEqual = 'LessEqual';\nexport type LessEqualInputs = BinaryInputs;\n\nexport const LinSpace = 'LinSpace';\nexport interface LinSpaceAttrs {\n start: number;\n stop: number;\n num: number;\n}\nexport const Log = 'Log';\nexport type LogInputs = UnaryInputs;\n\nexport const Log1p = 'Log1p';\nexport type Log1pInputs = UnaryInputs;\n\nexport const LogicalAnd = 'LogicalAnd';\nexport type LogicalAndInputs = BinaryInputs;\n\nexport const LogicalNot = 'LogicalNot';\nexport type LogicalNotInputs = Pick;\n\nexport const LogicalOr = 'LogicalOr';\nexport type LogicalOrInputs = BinaryInputs;\n\nexport const LogSoftmax = 'LogSoftmax';\nexport type LogSoftmaxInputs = Pick;\nexport interface LogSoftmaxAttrs {\n axis: number;\n}\n\nexport const LRN = 'LRN';\nexport type LRNInputs = Pick;\nexport interface LRNAttrs {\n depthRadius: number;\n bias: number;\n alpha: number;\n beta: number;\n}\n\nexport const LRNBackprop = 'LRNBackprop';\nexport type LRNBackpropInputs = Pick;\nexport interface LRNBackpropAttrs {\n depthRadius: number;\n bias: number;\n alpha: number;\n beta: number;\n}\n\nexport const Max = 'Max';\nexport type MaxInputs = Pick;\nexport interface MaxAttrs {\n reductionIndices: number|number[];\n keepDims: boolean;\n}\n\nexport const Maximum = 'Maximum';\nexport type MaximumInputs = BinaryInputs;\n\nexport const MaxPool = 'MaxPool';\nexport type MaxPoolInputs = Pick;\nexport interface MaxPoolAttrs {\n filterSize: [number, number]|number;\n strides: [number, number]|number;\n pad: 'valid'|'same'|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolBackprop = 'MaxPoolBackprop';\nexport type MaxPoolBackpropInputs =\n Pick;\nexport interface MaxPoolBackpropAttrs {\n filterSize: [number, number]|number;\n strides: [number, number]|number;\n pad: 'valid'|'same'|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3D = 'MaxPool3D';\nexport type MaxPool3DInputs = Pick;\nexport interface MaxPool3DAttrs {\n filterSize: [number, number, number]|number;\n strides: [number, number, number]|number;\n pad: 'valid'|'same'|number;\n dataFormat: 'NDHWC'|'NCDHW';\n dilations?: [number, number, number]|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPool3DBackprop = 'MaxPool3DBackprop';\nexport type MaxPool3DBackpropInputs =\n Pick;\nexport interface MaxPool3DBackpropAttrs {\n filterSize: [number, number, number]|number;\n strides: [number, number, number]|number;\n pad: 'valid'|'same'|number;\n dilations?: [number, number, number]|number;\n dimRoundingMode?: 'floor'|'round'|'ceil';\n}\n\nexport const MaxPoolWithArgmax = 'MaxPoolWithArgmax';\nexport type MaxPoolWithArgmaxInputs = Pick;\nexport interface MaxPoolWithArgmaxAttrs {\n filterSize: [number, number]|number;\n strides: [number, number]|number;\n pad: 'valid'|'same'|number;\n includeBatchInIndex: boolean;\n}\n\nexport const Mean = 'Mean';\nexport type MeanInputs = Pick;\nexport interface MeanAttrs {\n axis: number|number[];\n keepDims: boolean;\n}\n\nexport const Min = 'Min';\nexport type MinInputs = Pick;\nexport interface MinAttrs {\n axis: number|number[];\n keepDims: boolean;\n}\n\nexport const Minimum = 'Minimum';\nexport type MinimumInputs = BinaryInputs;\n\nexport const Mod = 'Mod';\nexport type ModInputs = BinaryInputs;\n\nexport const Multiply = 'Multiply';\nexport type MultiplyInputs = BinaryInputs;\n\nexport const Negate = 'Negate';\nexport type NegateInputs = UnaryInputs;\n\nexport const NotEqual = 'NotEqual';\nexport type NotEqualInputs = BinaryInputs;\n\nexport const NonMaxSuppressionV3 = 'NonMaxSuppressionV3';\nexport type NonMaxSuppressionV3Inputs =\n Pick;\nexport interface NonMaxSuppressionV3Attrs {\n maxOutputSize: number;\n iouThreshold: number;\n scoreThreshold: number;\n}\n\nexport const NonMaxSuppressionV4 = 'NonMaxSuppressionV4';\nexport type NonMaxSuppressionV4Inputs =\n Pick;\nexport interface NonMaxSuppressionV4Attrs {\n maxOutputSize: number;\n iouThreshold: number;\n scoreThreshold: number;\n padToMaxOutputSize: boolean;\n}\n\nexport const NonMaxSuppressionV5 = 'NonMaxSuppressionV5';\nexport type NonMaxSuppressionV5Inputs =\n Pick;\nexport interface NonMaxSuppressionV5Attrs {\n maxOutputSize: number;\n iouThreshold: number;\n scoreThreshold: number;\n softNmsSigma: number;\n}\n\nexport const OnesLike = 'OnesLike';\nexport type OnesLikeInputs = UnaryInputs;\n\nexport const OneHot = 'OneHot';\nexport type OneHotInputs = Pick;\nexport interface OneHotAttrs {\n depth: number;\n onValue: number;\n offValue: number;\n}\n\nexport const PadV2 = 'PadV2';\nexport type PadV2Inputs = Pick;\nexport interface PadV2Attrs {\n paddings: Array<[number, number]>;\n constantValue: number;\n}\n\nexport const Pool = 'Pool';\nexport type PoolInputs = Pick;\n\nexport const Pow = 'Pow';\nexport type PowInputs = BinaryInputs;\n\nexport const Prelu = 'Prelu';\nexport type PreluInputs = Pick;\n\nexport const Prod = 'Prod';\nexport type ProdInputs = Pick;\nexport interface ProdAttrs {\n axis: number|number[];\n keepDims: boolean;\n}\n\nexport const Range = 'Range';\nexport interface RangeAttrs {\n start: number;\n stop: number;\n step: number;\n dtype: 'float32'|'int32';\n}\n\nexport const Real = 'Real';\nexport type RealInputs = Pick;\n\nexport const Reciprocal = 'Reciprocal';\nexport type ReciprocalInputs = UnaryInputs;\n\nexport const Relu = 'Relu';\nexport type ReluInputs = Pick;\n\nexport const Reshape = 'Reshape';\nexport type ReshapeInputs = Pick;\nexport interface ReshapeAttrs {\n shape: number[];\n}\n\nexport const ResizeNearestNeighbor = 'ResizeNearestNeighbor';\nexport type ResizeNearestNeighborInputs = Pick;\nexport interface ResizeNearestNeighborAttrs {\n alignCorners: boolean;\n size: [number, number];\n}\n\nexport const ResizeNearestNeighborGrad = 'ResizeNearestNeighborGrad';\nexport type ResizeNearestNeighborGradInputs =\n Pick;\n\nexport const ResizeBilinear = 'ResizeBilinear';\nexport type ResizeBilinearInputs = Pick;\nexport interface ResizeBilinearAttrs {\n alignCorners: boolean;\n size: [number, number];\n}\n\nexport const ResizeBilinearGrad = 'ResizeBilinearGrad';\nexport type ResizeBilinearGradInputs = Pick;\n\nexport const Relu6 = 'Relu6';\nexport type Relu6Inputs = Pick;\n\nexport const Reverse = 'Reverse';\nexport type ReverseInputs = Pick;\nexport interface ReverseAttrs {\n dims: number|number[];\n}\n\nexport const Round = 'Round';\nexport type RoundInputs = UnaryInputs;\n\nexport const Rsqrt = 'Rsqrt';\nexport type RsqrtInputs = UnaryInputs;\n\nexport const ScatterNd = 'ScatterNd';\nexport type ScatterNdInputs = Pick;\nexport interface ScatterNdAttrs {\n shape: number[];\n}\n\nexport const SelectV2 = 'SelectV2';\nexport type SelectV2Inputs = Pick;\n\nexport const Selu = 'Selu';\nexport type SeluInputs = Pick;\n\nexport const Slice = 'Slice';\nexport type SliceInputs = Pick;\nexport interface SliceAttrs {\n begin: number|number[];\n size: number|number[];\n}\nexport const Sin = 'Sin';\nexport type SinInputs = UnaryInputs;\n\nexport const Sinh = 'Sinh';\nexport type SinhInputs = UnaryInputs;\n\nexport const Sign = 'Sign';\nexport type SignInputs = UnaryInputs;\n\nexport const Sigmoid = 'Sigmoid';\nexport type SigmoidInputs = UnaryInputs;\n\nexport const Softplus = 'Softplus';\nexport type SoftplusInputs = UnaryInputs;\n\nexport const Sqrt = 'Sqrt';\nexport type SqrtInputs = UnaryInputs;\n\nexport const Sum = 'Sum';\nexport type SumInputs = Pick;\nexport interface SumAttrs {\n axis: number|number[];\n keepDims: boolean;\n}\n\nexport const SpaceToBatchND = 'SpaceToBatchND';\nexport type SpaceToBatchNDInputs = Pick;\nexport interface SpaceToBatchNDAttrs {\n blockShape: number[];\n paddings: number[][];\n}\n\nexport const SplitV = 'SplitV';\nexport type SplitVInputs = Pick;\nexport interface SplitVAttrs {\n numOrSizeSplits: number[]|number;\n axis: number;\n}\n\nexport const Softmax = 'Softmax';\nexport type SoftmaxInputs = Pick;\nexport interface SoftmaxAttrs {\n dim: number;\n}\n\nexport const SquaredDifference = 'SquaredDifference';\nexport type SquaredDifferenceInputs = BinaryInputs;\n\nexport const Square = 'Square';\nexport type SquareInputs = Pick;\n\nexport const Sub = 'Sub';\nexport type SubInputs = BinaryInputs;\n\nexport const SparseToDense = 'SparseToDense';\nexport type SparseToDenseInputs =\n Pick;\nexport interface SparseToDenseAttrs {\n outputShape: number[];\n}\n\nexport const StridedSlice = 'StridedSlice';\nexport type StridedSliceInputs = Pick;\nexport interface StridedSliceAttrs {\n begin: number[];\n end: number[];\n strides: number[];\n beginMask: number;\n endMask: number;\n ellipsisMask: number;\n newAxisMask: number;\n shrinkAxisMask: number;\n}\n\nexport const Tan = 'Tan';\nexport type TanInputs = UnaryInputs;\n\nexport const Tanh = 'Tanh';\nexport type TanhInputs = UnaryInputs;\n\nexport const Tile = 'Tile';\nexport type TileInputs = Pick;\nexport interface TileAttrs {\n reps: number[];\n}\n\nexport const TopK = 'TopK';\nexport type TopKInputs = Pick;\nexport interface TopKAttrs {\n k: number;\n sorted: boolean;\n}\n\nexport const Transpose = 'Transpose';\nexport type TransposeInputs = Pick;\nexport interface TransposeAttrs {\n perm: number[];\n}\n\nexport const Unique = 'Unique';\nexport type UniqueInputs = Pick;\nexport interface UniqueAttrs {\n axis: number;\n}\n\nexport type UnaryInputs = Pick;\n\nexport const Unpack = 'Unpack';\nexport type UnpackInputs = Pick;\nexport interface UnpackAttrs {\n axis: number;\n}\n\nexport const UnsortedSegmentSum = 'UnsortedSegmentSum';\nexport type UnsortedSegmentSumInputs =\n Pick;\nexport interface UnsortedSegmentSumAttrs {\n numSegments: number;\n}\n\nexport const ZerosLike = 'ZerosLike';\nexport type ZerosLikeInputs = UnaryInputs;\n\n/**\n * TensorFlow.js-only kernels\n */\nexport const Step = 'Step';\nexport type StepInputs = UnaryInputs;\nexport interface StepAttrs {\n alpha: number;\n}\n\nexport const FromPixels = 'FromPixels';\nexport interface FromPixelsInputs {\n pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n HTMLVideoElement;\n}\nexport interface FromPixelsAttrs {\n numChannels: number;\n}\n\nexport const RotateWithOffset = 'RotateWithOffset';\nexport type RotateWithOffsetInputs = Pick;\nexport interface RotateWithOffsetAttrs {\n radians: number;\n fillValue: number|[number, number, number];\n center: number|[number, number];\n}\n\nexport const _FusedMatMul = '_FusedMatMul';\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulInputs extends NamedTensorInfoMap {\n a: TensorInfo;\n b: TensorInfo;\n bias?: TensorInfo;\n preluActivationWeights?: TensorInfo;\n}\n// tslint:disable-next-line: class-name\nexport interface _FusedMatMulAttrs {\n transposeA: boolean;\n transposeB: boolean;\n activation: Activation;\n}\n\nexport const FusedConv2D = 'FusedConv2D';\nexport interface FusedConv2DInputs extends NamedTensorInfoMap {\n x: TensorInfo;\n filter: TensorInfo;\n bias?: TensorInfo;\n preluActivationWeights?: TensorInfo;\n}\nexport interface FusedConv2DAttrs {\n strides: [number, number]|number;\n pad: 'valid'|'same'|number|ExplicitPadding;\n dataFormat: 'NHWC'|'NCHW';\n dilations: [number, number]|number;\n dimRoundingMode: 'floor'|'round'|'ceil';\n activation: Activation;\n}\n\nexport const FusedDepthwiseConv2D = 'FusedDepthwiseConv2D';\nexport interface FusedDepthwiseConv2DInputs extends NamedTensorInfoMap {\n x: TensorInfo;\n filter: TensorInfo;\n bias?: TensorInfo;\n preluActivationWeights?: TensorInfo;\n}\nexport interface FusedDepthwiseConv2DAttrs {\n strides: [number, number]|number;\n pad: 'valid'|'same'|number;\n dataFormat: 'NHWC'|'NCHW';\n dilations: [number, number]|number;\n dimRoundingMode: 'floor'|'round'|'ceil';\n activation: Activation;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from './environment';\n\nimport {getGlobal} from './global_util';\nimport {NamedGradientMap} from './tape';\nimport {Tensor} from './tensor';\nimport {DataType, RecursiveArray} from './types';\n\nconst kernelRegistry =\n getGlobal('kernelRegistry', () => new Map());\nconst gradRegistry =\n getGlobal('gradRegistry', () => new Map());\n\nexport type DataId = object;\n\ntype AttributeValue =\n number|number[]|boolean|boolean[]|string|string[]|NamedAttrMap;\n\n/** These are extra non-tensor/primitive params passed to kernel functions. */\nexport type Attribute = AttributeValue|RecursiveArray;\n\n/** Specifies the code to run when executing a kernel. */\nexport type KernelFunc = (params: {\n inputs: NamedTensorInfoMap,\n backend: {},\n attrs?: NamedAttrMap,\n}) => TensorInfo|TensorInfo[];\n\n/** The function to run when computing a gradient during backprop. */\nexport type GradFunc =\n (dy: Tensor|Tensor[], saved: Tensor[], attrs: NamedAttrMap) =>\n NamedGradientMap;\n\n/** Function that gets called after the backend initializes. */\nexport type KernelSetupFunc = (backend: {}) => void;\n/** Function that gets called right before the backend is disposed. */\nexport type KernelDisposeFunc = KernelSetupFunc;\n\n/** Config object for registering a kernel in the global registry. */\nexport interface KernelConfig {\n kernelName: string;\n backendName: string;\n kernelFunc: KernelFunc;\n setupFunc?: KernelSetupFunc;\n disposeFunc?: KernelDisposeFunc;\n}\n\n/** Config object for registering a gradient in the global registry. */\nexport interface GradConfig {\n kernelName: string;\n inputsToSave?: string[];\n // When saveAllInputs is true, all inputs will be saved. Only use this flag\n // if inputs is an array of Tensors.\n saveAllInputs?: boolean;\n outputsToSave?: boolean[];\n gradFunc: GradFunc;\n}\n\n/** Holds metadata for a given tensor. */\nexport interface TensorInfo {\n dataId: DataId;\n shape: number[];\n dtype: DataType;\n}\n\nexport interface NamedTensorInfoMap {\n [name: string]: TensorInfo;\n}\n\nexport interface NamedAttrMap {\n [name: string]: Attribute;\n}\n\n/**\n * Returns the kernel function (code) associated with the provided names.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n */\nexport function getKernel(\n kernelName: string, backendName: string): KernelConfig {\n const key = makeKey(kernelName, backendName);\n return kernelRegistry.get(key);\n}\n\n/**\n * Returns the registered gradient info associated with the provided kernel.\n * @param kernelName The official TF kernel name.\n */\nexport function getGradient(kernelName: string): GradConfig {\n return gradRegistry.get(kernelName);\n}\n\nexport function getKernelsForBackend(backendName: string): KernelConfig[] {\n const it = kernelRegistry.entries();\n const result: KernelConfig[] = [];\n\n while (true) {\n const {done, value} = it.next();\n if (done) {\n break;\n }\n const [key, config] = value;\n const [backend, ] = key.split('_');\n if (backend === backendName) {\n result.push(config);\n }\n }\n return result;\n}\n\n/**\n * Registers the function (forward pass) for the kernel in a global registry.\n *\n * @param config A config object with the following properties:\n * - `kernelName` The official name of the kernel.\n * - `backendName` The official name of the backend.\n * - `kernelFunc` The function to run during the forward pass of the kernel.\n * - `setupFunc` Optional. Gets called once, after the backend initializes.\n * - `disposeFunc` Optional. Gets called once, right before the backend is\n * disposed.\n */\nexport function registerKernel(config: KernelConfig) {\n const {kernelName, backendName} = config;\n const key = makeKey(kernelName, backendName);\n if (kernelRegistry.has(key)) {\n console.warn(\n `The kernel '${kernelName}' for backend ` +\n `'${backendName}' is already registered`);\n }\n kernelRegistry.set(key, config);\n}\n\n/**\n * Registers a gradient function for a given kernel in the global registry,\n * to be used during the back-propagation of that kernel.\n *\n * @param config An object with the following properties:\n * - `kernelName` The name of the kernel that the gradient function is for.\n * - `gradFunc` The function to run during back-propagation.\n */\nexport function registerGradient(config: GradConfig) {\n const {kernelName} = config;\n\n if (gradRegistry.has(kernelName)) {\n // TODO (yassogba) after 3.0 assess whether we need to keep this gated\n // to debug mode.\n if (env().getBool('DEBUG')) {\n console.warn(`Overriding the gradient for '${kernelName}'`);\n }\n }\n gradRegistry.set(kernelName, config);\n}\n\n/**\n * Removes the kernel function from the registry.\n *\n * @param kernelName The official name of the kernel.\n * @param backendName The official name of the backend.\n *\n */\nexport function unregisterKernel(\n kernelName: string, backendName: string): void {\n const key = makeKey(kernelName, backendName);\n if (!kernelRegistry.has(key)) {\n throw new Error(\n `The kernel '${kernelName}' for backend ` +\n `'${backendName}' is not registered`);\n }\n kernelRegistry.delete(key);\n}\n\n/** Removes the registered gradient from the global registry. */\nexport function unregisterGradient(kernelName: string): void {\n if (!gradRegistry.has(kernelName)) {\n throw new Error(\n `The gradient '${kernelName}' for backend is not registered`);\n }\n gradRegistry.delete(kernelName);\n}\n\n/**\n * Finds kernels that have already been registered to a backend and re-registers\n * them for a new backend. Useful for registering custom backends.\n * @param registeredBackendName Already registered backend.\n * @param newBackendName New backend.\n */\nexport function copyRegisteredKernels(\n registeredBackendName: string, newBackendName: string): void {\n const kernels = getKernelsForBackend(registeredBackendName);\n kernels.forEach(kernelConfig => {\n const newKernelConfig =\n Object.assign({}, kernelConfig, {backendName: newBackendName});\n registerKernel(newKernelConfig);\n });\n}\n\nfunction makeKey(kernelName: string, backendName: string) {\n return `${backendName}_${kernelName}`;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from './environment';\nimport {BackendValues, DataType, DataTypeMap, FlatVector, NumericDataType, RecursiveArray, TensorLike, TypedArray} from './types';\n\n/**\n * Shuffles the array in-place using Fisher-Yates algorithm.\n *\n * ```js\n * const a = [1, 2, 3, 4, 5];\n * tf.util.shuffle(a);\n * console.log(a);\n * ```\n *\n * @param array The array to shuffle in-place.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\n// tslint:disable-next-line:no-any\nexport function shuffle(array: any[]|Uint32Array|Int32Array|\n Float32Array): void {\n let counter = array.length;\n let temp = 0;\n let index = 0;\n // While there are elements in the array\n while (counter > 0) {\n // Pick a random index\n index = (Math.random() * counter) | 0;\n // Decrease counter by 1\n counter--;\n // And swap the last element with it\n temp = array[counter];\n array[counter] = array[index];\n array[index] = temp;\n }\n}\n\n/** Clamps a value to a specified range. */\nexport function clamp(min: number, x: number, max: number): number {\n return Math.max(min, Math.min(x, max));\n}\n\nexport function nearestLargerEven(val: number): number {\n return val % 2 === 0 ? val : val + 1;\n}\n\nexport function sum(arr: number[]): number {\n let sum = 0;\n for (let i = 0; i < arr.length; i++) {\n sum += arr[i];\n }\n return sum;\n}\n\n/**\n * Returns a sample from a uniform [a, b) distribution.\n *\n * @param a The minimum support (inclusive).\n * @param b The maximum support (exclusive).\n * @return A pseudorandom number on the half-open interval [a,b).\n */\nexport function randUniform(a: number, b: number) {\n const r = Math.random();\n return (b * r) + (1 - r) * a;\n}\n\n/** Returns the squared Euclidean distance between two vectors. */\nexport function distSquared(a: FlatVector, b: FlatVector): number {\n let result = 0;\n for (let i = 0; i < a.length; i++) {\n const diff = Number(a[i]) - Number(b[i]);\n result += diff * diff;\n }\n return result;\n}\n\n/**\n * Asserts that the expression is true. Otherwise throws an error with the\n * provided message.\n *\n * ```js\n * const x = 2;\n * tf.util.assert(x === 2, 'x is not 2');\n * ```\n *\n * @param expr The expression to assert (as a boolean).\n * @param msg A function that returns the message to report when throwing an\n * error. We use a function for performance reasons.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function assert(expr: boolean, msg: () => string) {\n if (!expr) {\n throw new Error(typeof msg === 'string' ? msg : msg());\n }\n}\n\nexport function assertShapesMatch(\n shapeA: number[], shapeB: number[], errorMessagePrefix = ''): void {\n assert(\n arraysEqual(shapeA, shapeB),\n () => errorMessagePrefix + ` Shapes ${shapeA} and ${shapeB} must match`);\n}\n\nexport function assertNonNull(a: TensorLike): void {\n assert(\n a != null,\n () => `The input to the tensor constructor must be a non-null value.`);\n}\n\n// NOTE: We explicitly type out what T extends instead of any so that\n// util.flatten on a nested array of number doesn't try to infer T as a\n// number[][], causing us to explicitly type util.flatten().\n/**\n * Flattens an arbitrarily nested array.\n *\n * ```js\n * const a = [[1, 2], [3, 4], [5, [6, [7]]]];\n * const flat = tf.util.flatten(a);\n * console.log(flat);\n * ```\n *\n * @param arr The nested array to flatten.\n * @param result The destination array which holds the elements.\n * @param skipTypedArray If true, avoids flattening the typed arrays. Defaults\n * to false.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function\nflatten|TypedArray>(\n arr: T|RecursiveArray, result: T[] = [], skipTypedArray = false): T[] {\n if (result == null) {\n result = [];\n }\n if (Array.isArray(arr) || isTypedArray(arr) && !skipTypedArray) {\n for (let i = 0; i < arr.length; ++i) {\n flatten(arr[i], result, skipTypedArray);\n }\n } else {\n result.push(arr as T);\n }\n return result;\n}\n\n/**\n * Returns the size (number of elements) of the tensor given its shape.\n *\n * ```js\n * const shape = [3, 4, 2];\n * const size = tf.util.sizeFromShape(shape);\n * console.log(size);\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function sizeFromShape(shape: number[]): number {\n if (shape.length === 0) {\n // Scalar.\n return 1;\n }\n let size = shape[0];\n for (let i = 1; i < shape.length; i++) {\n size *= shape[i];\n }\n return size;\n}\n\nexport function isScalarShape(shape: number[]): boolean {\n return shape.length === 0;\n}\n\nexport function arraysEqual(n1: FlatVector, n2: FlatVector) {\n if (n1 === n2) {\n return true;\n }\n if (n1 == null || n2 == null) {\n return false;\n }\n\n if (n1.length !== n2.length) {\n return false;\n }\n for (let i = 0; i < n1.length; i++) {\n if (n1[i] !== n2[i]) {\n return false;\n }\n }\n return true;\n}\n\nexport function isInt(a: number): boolean {\n return a % 1 === 0;\n}\n\nexport function tanh(x: number): number {\n // tslint:disable-next-line:no-any\n if ((Math as any).tanh != null) {\n // tslint:disable-next-line:no-any\n return (Math as any).tanh(x);\n }\n if (x === Infinity) {\n return 1;\n } else if (x === -Infinity) {\n return -1;\n } else {\n const e2x = Math.exp(2 * x);\n return (e2x - 1) / (e2x + 1);\n }\n}\n\nexport function sizeToSquarishShape(size: number): [number, number] {\n const width = Math.ceil(Math.sqrt(size));\n return [width, Math.ceil(size / width)];\n}\n\n/**\n * Creates a new array with randomized indicies to a given quantity.\n *\n * ```js\n * const randomTen = tf.util.createShuffledIndices(10);\n * console.log(randomTen);\n * ```\n *\n * @param number Quantity of how many shuffled indicies to create.\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function createShuffledIndices(n: number): Uint32Array {\n const shuffledIndices = new Uint32Array(n);\n for (let i = 0; i < n; ++i) {\n shuffledIndices[i] = i;\n }\n shuffle(shuffledIndices);\n return shuffledIndices;\n}\n\nexport function rightPad(a: string, size: number): string {\n if (size <= a.length) {\n return a;\n }\n return a + ' '.repeat(size - a.length);\n}\n\nexport function repeatedTry(\n checkFn: () => boolean, delayFn = (counter: number) => 0,\n maxCounter?: number): Promise {\n return new Promise((resolve, reject) => {\n let tryCount = 0;\n\n const tryFn = () => {\n if (checkFn()) {\n resolve();\n return;\n }\n\n tryCount++;\n\n const nextBackoff = delayFn(tryCount);\n\n if (maxCounter != null && tryCount >= maxCounter) {\n reject();\n return;\n }\n setTimeout(tryFn, nextBackoff);\n };\n\n tryFn();\n });\n}\n\n/**\n * Given the full size of the array and a shape that may contain -1 as the\n * implicit dimension, returns the inferred shape where -1 is replaced.\n * E.g. For shape=[2, -1, 3] and size=24, it will return [2, 4, 3].\n *\n * @param shape The shape, which may contain -1 in some dimension.\n * @param size The full size (number of elements) of the array.\n * @return The inferred shape where -1 is replaced with the inferred size.\n */\nexport function inferFromImplicitShape(\n shape: number[], size: number): number[] {\n let shapeProd = 1;\n let implicitIdx = -1;\n\n for (let i = 0; i < shape.length; ++i) {\n if (shape[i] >= 0) {\n shapeProd *= shape[i];\n } else if (shape[i] === -1) {\n if (implicitIdx !== -1) {\n throw Error(\n `Shapes can only have 1 implicit size. ` +\n `Found -1 at dim ${implicitIdx} and dim ${i}`);\n }\n implicitIdx = i;\n } else if (shape[i] < 0) {\n throw Error(`Shapes can not be < 0. Found ${shape[i]} at dim ${i}`);\n }\n }\n\n if (implicitIdx === -1) {\n if (size > 0 && size !== shapeProd) {\n throw Error(`Size(${size}) must match the product of shape ${shape}`);\n }\n return shape;\n }\n\n if (shapeProd === 0) {\n throw Error(\n `Cannot infer the missing size in [${shape}] when ` +\n `there are 0 elements`);\n }\n if (size % shapeProd !== 0) {\n throw Error(\n `The implicit shape can't be a fractional number. ` +\n `Got ${size} / ${shapeProd}`);\n }\n\n const newShape = shape.slice();\n newShape[implicitIdx] = size / shapeProd;\n return newShape;\n}\n\nexport function parseAxisParam(\n axis: number|number[], shape: number[]): number[] {\n const rank = shape.length;\n\n // Normalize input\n axis = axis == null ? shape.map((s, i) => i) : [].concat(axis);\n\n // Check for valid range\n assert(\n axis.every(ax => ax >= -rank && ax < rank),\n () =>\n `All values in axis param must be in range [-${rank}, ${rank}) but ` +\n `got axis ${axis}`);\n\n // Check for only integers\n assert(\n axis.every(ax => isInt(ax)),\n () => `All values in axis param must be integers but ` +\n `got axis ${axis}`);\n\n // Handle negative axis.\n return axis.map(a => a < 0 ? rank + a : a);\n}\n\n/** Reduces the shape by removing all dimensions of shape 1. */\nexport function squeezeShape(shape: number[], axis?: number[]):\n {newShape: number[], keptDims: number[]} {\n const newShape: number[] = [];\n const keptDims: number[] = [];\n const isEmptyArray = axis != null && Array.isArray(axis) && axis.length === 0;\n const axes = (axis == null || isEmptyArray) ?\n null :\n parseAxisParam(axis, shape).sort();\n let j = 0;\n for (let i = 0; i < shape.length; ++i) {\n if (axes != null) {\n if (axes[j] === i && shape[i] !== 1) {\n throw new Error(\n `Can't squeeze axis ${i} since its dim '${shape[i]}' is not 1`);\n }\n if ((axes[j] == null || axes[j] > i) && shape[i] === 1) {\n newShape.push(shape[i]);\n keptDims.push(i);\n }\n if (axes[j] <= i) {\n j++;\n }\n }\n if (shape[i] !== 1) {\n newShape.push(shape[i]);\n keptDims.push(i);\n }\n }\n return {newShape, keptDims};\n}\n\nexport function getTypedArrayFromDType(\n dtype: D, size: number): DataTypeMap[D] {\n let values = null;\n if (dtype == null || dtype === 'float32') {\n values = new Float32Array(size);\n } else if (dtype === 'int32') {\n values = new Int32Array(size);\n } else if (dtype === 'bool') {\n values = new Uint8Array(size);\n } else {\n throw new Error(`Unknown data type ${dtype}`);\n }\n return values as DataTypeMap[D];\n}\n\nexport function getArrayFromDType(\n dtype: D, size: number): DataTypeMap[D] {\n let values = null;\n if (dtype == null || dtype === 'float32') {\n values = new Float32Array(size);\n } else if (dtype === 'int32') {\n values = new Int32Array(size);\n } else if (dtype === 'bool') {\n values = new Uint8Array(size);\n } else if (dtype === 'string') {\n values = new Array<'string'>(size);\n } else {\n throw new Error(`Unknown data type ${dtype}`);\n }\n return values as DataTypeMap[D];\n}\n\nexport function checkConversionForErrors(\n vals: DataTypeMap[D]|number[], dtype: D): void {\n for (let i = 0; i < vals.length; i++) {\n const num = vals[i] as number;\n if (isNaN(num) || !isFinite(num)) {\n throw Error(`A tensor of type ${dtype} being uploaded contains ${num}.`);\n }\n }\n}\n\n/** Returns true if the dtype is valid. */\nexport function isValidDtype(dtype: DataType): boolean {\n return dtype === 'bool' || dtype === 'complex64' || dtype === 'float32' ||\n dtype === 'int32' || dtype === 'string';\n}\n\n/**\n * Returns true if the new type can't encode the old type without loss of\n * precision.\n */\nexport function hasEncodingLoss(oldType: DataType, newType: DataType): boolean {\n if (newType === 'complex64') {\n return false;\n }\n if (newType === 'float32' && oldType !== 'complex64') {\n return false;\n }\n if (newType === 'int32' && oldType !== 'float32' && oldType !== 'complex64') {\n return false;\n }\n if (newType === 'bool' && oldType === 'bool') {\n return false;\n }\n return true;\n}\n\nexport function isTypedArray(a: {}): a is Float32Array|Int32Array|Uint8Array {\n return a instanceof Float32Array || a instanceof Int32Array ||\n a instanceof Uint8Array;\n}\n\nexport function bytesPerElement(dtype: DataType): number {\n if (dtype === 'float32' || dtype === 'int32') {\n return 4;\n } else if (dtype === 'complex64') {\n return 8;\n } else if (dtype === 'bool') {\n return 1;\n } else {\n throw new Error(`Unknown dtype ${dtype}`);\n }\n}\n\n/**\n * Returns the approximate number of bytes allocated in the string array - 2\n * bytes per character. Computing the exact bytes for a native string in JS is\n * not possible since it depends on the encoding of the html page that serves\n * the website.\n */\nexport function bytesFromStringArray(arr: Uint8Array[]): number {\n if (arr == null) {\n return 0;\n }\n let bytes = 0;\n arr.forEach(x => bytes += x.length);\n return bytes;\n}\n\n/** Returns true if the value is a string. */\nexport function isString(value: {}): value is string {\n return typeof value === 'string' || value instanceof String;\n}\n\nexport function isBoolean(value: {}): boolean {\n return typeof value === 'boolean';\n}\n\nexport function isNumber(value: {}): boolean {\n return typeof value === 'number';\n}\n\nexport function inferDtype(values: TensorLike): DataType {\n if (Array.isArray(values)) {\n return inferDtype(values[0]);\n }\n if (values instanceof Float32Array) {\n return 'float32';\n } else if (values instanceof Int32Array || values instanceof Uint8Array) {\n return 'int32';\n } else if (isNumber(values)) {\n return 'float32';\n } else if (isString(values)) {\n return 'string';\n } else if (isBoolean(values)) {\n return 'bool';\n }\n return 'float32';\n}\n\nexport function isFunction(f: Function) {\n return !!(f && f.constructor && f.call && f.apply);\n}\n\nexport function nearestDivisor(size: number, start: number): number {\n for (let i = start; i < size; ++i) {\n if (size % i === 0) {\n return i;\n }\n }\n return size;\n}\n\nexport function computeStrides(shape: number[]): number[] {\n const rank = shape.length;\n if (rank < 2) {\n return [];\n }\n\n // Last dimension has implicit stride of 1, thus having D-1 (instead of D)\n // strides.\n const strides = new Array(rank - 1);\n strides[rank - 2] = shape[rank - 1];\n for (let i = rank - 3; i >= 0; --i) {\n strides[i] = strides[i + 1] * shape[i + 1];\n }\n return strides;\n}\n\n/**\n * Create typed array for scalar value. Used for storing in `DataStorage`.\n */\nexport function createScalarValue(\n value: DataType, dtype: DataType): BackendValues {\n if (dtype === 'string') {\n return encodeString(value);\n }\n\n return toTypedArray([value], dtype);\n}\n\nexport function toTypedArray(a: TensorLike, dtype: DataType): TypedArray {\n if (dtype === 'string') {\n throw new Error('Cannot convert a string[] to a TypedArray');\n }\n if (Array.isArray(a)) {\n a = flatten(a);\n }\n\n if (env().getBool('DEBUG')) {\n checkConversionForErrors(a as number[], dtype);\n }\n if (noConversionNeeded(a, dtype)) {\n return a as TypedArray;\n }\n if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n return new Float32Array(a as number[]);\n } else if (dtype === 'int32') {\n return new Int32Array(a as number[]);\n } else if (dtype === 'bool') {\n const bool = new Uint8Array((a as number[]).length);\n for (let i = 0; i < bool.length; ++i) {\n if (Math.round((a as number[])[i]) !== 0) {\n bool[i] = 1;\n }\n }\n return bool;\n } else {\n throw new Error(`Unknown data type ${dtype}`);\n }\n}\n\nfunction createNestedArray(offset: number, shape: number[], a: TypedArray) {\n const ret = new Array();\n if (shape.length === 1) {\n const d = shape[0];\n for (let i = 0; i < d; i++) {\n ret[i] = a[offset + i];\n }\n } else {\n const d = shape[0];\n const rest = shape.slice(1);\n const len = rest.reduce((acc, c) => acc * c);\n for (let i = 0; i < d; i++) {\n ret[i] = createNestedArray(offset + i * len, rest, a);\n }\n }\n return ret;\n}\n\n// Provide a nested array of TypedArray in given shape.\nexport function toNestedArray(shape: number[], a: TypedArray) {\n if (shape.length === 0) {\n // Scalar type should return a single number.\n return a[0];\n }\n const size = shape.reduce((acc, c) => acc * c);\n if (size === 0) {\n // A tensor with shape zero should be turned into empty list.\n return [];\n }\n if (size !== a.length) {\n throw new Error(`[${shape}] does not match the input size ${a.length}.`);\n }\n\n return createNestedArray(0, shape, a);\n}\n\nfunction noConversionNeeded(a: TensorLike, dtype: DataType): boolean {\n return (a instanceof Float32Array && dtype === 'float32') ||\n (a instanceof Int32Array && dtype === 'int32') ||\n (a instanceof Uint8Array && dtype === 'bool');\n}\n\nexport function makeOnesTypedArray(\n size: number, dtype: D): DataTypeMap[D] {\n const array = makeZerosTypedArray(size, dtype);\n for (let i = 0; i < array.length; i++) {\n array[i] = 1;\n }\n return array;\n}\n\nexport function makeZerosTypedArray(\n size: number, dtype: D): DataTypeMap[D] {\n if (dtype == null || dtype === 'float32' || dtype === 'complex64') {\n return new Float32Array(size) as DataTypeMap[D];\n } else if (dtype === 'int32') {\n return new Int32Array(size) as DataTypeMap[D];\n } else if (dtype === 'bool') {\n return new Uint8Array(size) as DataTypeMap[D];\n } else {\n throw new Error(`Unknown data type ${dtype}`);\n }\n}\n\n/**\n * Make nested `TypedArray` filled with zeros.\n * @param shape The shape information for the nested array.\n * @param dtype dtype of the array element.\n */\nexport function makeZerosNestedTypedArray(\n shape: number[], dtype: D) {\n const size = shape.reduce((prev, curr) => prev * curr, 1);\n if (dtype == null || dtype === 'float32') {\n return toNestedArray(shape, new Float32Array(size));\n } else if (dtype === 'int32') {\n return toNestedArray(shape, new Int32Array(size));\n } else if (dtype === 'bool') {\n return toNestedArray(shape, new Uint8Array(size));\n } else {\n throw new Error(`Unknown data type ${dtype}`);\n }\n}\n\n/**\n * Returns the current high-resolution time in milliseconds relative to an\n * arbitrary time in the past. It works across different platforms (node.js,\n * browsers).\n *\n * ```js\n * console.log(tf.util.now());\n * ```\n *\n * @doc {heading: 'Util', namespace: 'util'}\n */\nexport function now(): number {\n return env().platform.now();\n}\n\nexport function assertNonNegativeIntegerDimensions(shape: number[]) {\n shape.forEach(dimSize => {\n assert(\n Number.isInteger(dimSize) && dimSize >= 0,\n () =>\n `Tensor must have a shape comprised of positive integers but got ` +\n `shape [${shape}].`);\n });\n}\n\n/**\n * Returns a platform-specific implementation of\n * [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n *\n * If `fetch` is defined on the global object (`window`, `process`, etc.),\n * `tf.util.fetch` returns that function.\n *\n * If not, `tf.util.fetch` returns a platform-specific solution.\n *\n * ```js\n * const resource = await tf.util.fetch('https://unpkg.com/@tensorflow/tfjs');\n * // handle response\n * ```\n *\n * @doc {heading: 'Util'}\n */\nexport function fetch(\n path: string, requestInits?: RequestInit): Promise {\n return env().platform.fetch(path, requestInits);\n}\n\n/**\n * Encodes the provided string into bytes using the provided encoding scheme.\n *\n * @param s The string to encode.\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function encodeString(s: string, encoding = 'utf-8'): Uint8Array {\n encoding = encoding || 'utf-8';\n return env().platform.encode(s, encoding);\n}\n\n/**\n * Decodes the provided bytes into a string using the provided encoding scheme.\n * @param bytes The bytes to decode.\n *\n * @param encoding The encoding scheme. Defaults to utf-8.\n *\n * @doc {heading: 'Util'}\n */\nexport function decodeString(bytes: Uint8Array, encoding = 'utf-8'): string {\n encoding = encoding || 'utf-8';\n return env().platform.decode(bytes, encoding);\n}\n\n/**\n * Computes flat index for a given location (multidimentionsal index) in a\n * Tensor/multidimensional array.\n *\n * @param locs Location in the tensor.\n * @param rank Rank of the tensor.\n * @param strides Tensor strides.\n */\nexport function locToIndex(\n locs: number[], rank: number, strides: number[]): number {\n if (rank === 0) {\n return 0;\n } else if (rank === 1) {\n return locs[0];\n }\n let index = locs[locs.length - 1];\n for (let i = 0; i < locs.length - 1; ++i) {\n index += strides[i] * locs[i];\n }\n return index;\n}\n\n/**\n * Computes the location (multidimensional index) in a tensor/multidimentional\n * array for a given flat index.\n *\n * @param index Index in flat array.\n * @param rank Rank of tensor.\n * @param strides Strides of tensor.\n */\nexport function indexToLoc(\n index: number, rank: number, strides: number[]): number[] {\n if (rank === 0) {\n return [];\n } else if (rank === 1) {\n return [index];\n }\n const locs: number[] = new Array(rank);\n for (let i = 0; i < locs.length - 1; ++i) {\n locs[i] = Math.floor(index / strides[i]);\n index -= locs[i] * strides[i];\n }\n locs[locs.length - 1] = index;\n return locs;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimer} from './backends/backend';\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport {DataType, DataTypeMap, TypedArray} from './types';\nimport * as util from './util';\n\nexport type KernelProfile = {\n kernelName: string,\n outputs: Tensor[],\n inputs: NamedTensorMap,\n timeMs: Promise,\n extraInfo: Promise\n};\n\nexport class Profiler {\n constructor(private backendTimer: BackendTimer, private logger?: Logger) {\n if (logger == null) {\n this.logger = new Logger();\n }\n }\n\n profileKernel(kernelName: string, inputs: NamedTensorMap, f: () => Tensor[]):\n KernelProfile {\n let outputs: Tensor[];\n const holdResultWrapperFn = () => {\n outputs = f();\n };\n const timer = this.backendTimer.time(holdResultWrapperFn);\n\n for (let i = 0; i < outputs.length; i++) {\n const output = outputs[i];\n // Dangling promise here because we don't want to propagate up\n // asynchronicity.\n output.data().then(tensorVals => {\n checkComputationForErrors(tensorVals, output.dtype, kernelName);\n });\n }\n\n const kernelProfile = {\n kernelName,\n outputs,\n inputs,\n timeMs: timer.then(timing => timing.kernelMs),\n extraInfo: timer.then(\n timing => timing.getExtraProfileInfo != null ?\n timing.getExtraProfileInfo() :\n '')\n };\n return kernelProfile;\n }\n\n logKernelProfile(kernelProfile: KernelProfile): void {\n const {kernelName, outputs, timeMs, inputs, extraInfo} = kernelProfile;\n\n outputs.forEach(result => {\n Promise.all([result.data(), timeMs, extraInfo]).then(valueContainer => {\n this.logger.logKernelProfile(\n kernelName, result, valueContainer[0], valueContainer[1], inputs,\n valueContainer[2]);\n });\n });\n }\n}\n\nexport function checkComputationForErrors(\n vals: DataTypeMap[D], dtype: D, kernelName: string): boolean {\n if (dtype !== 'float32') {\n // Only floating point computations will generate NaN values\n return false;\n }\n for (let i = 0; i < vals.length; i++) {\n const num = vals[i] as number;\n if (isNaN(num) || !isFinite(num)) {\n // Throwing custom exception so behavior is testable.\n console.warn(`Found ${num} in the result of '${kernelName}'`);\n return true;\n }\n }\n return false;\n}\n\nexport class Logger {\n logKernelProfile(\n name: string, result: Tensor, vals: TypedArray,\n timeMs: number|{error: string}, inputs: NamedTensorMap,\n extraInfo?: string) {\n const time = typeof timeMs === 'number' ? util.rightPad(`${timeMs}ms`, 9) :\n timeMs['error'];\n const paddedName = util.rightPad(name, 25);\n const rank = result.rank;\n const size = result.size;\n const shape = util.rightPad(result.shape.toString(), 14);\n let inputShapesDescription = '';\n\n for (const name in inputs) {\n const input = inputs[name];\n if (input != null) {\n // The input might be a non-tensor (e.g HTMLImageElement), in which case\n // we claim the output shape as input shape.\n const inputShape = input.shape || result.shape;\n const inputRank = inputShape.length;\n inputShapesDescription +=\n `${name}: ${inputRank}D ${inputRank > 0 ? inputShape : ''} `;\n }\n }\n\n console.log(\n `%c${paddedName}\\t%c${time}\\t%c${rank}D ${shape}\\t%c${size}\\t%c${\n inputShapesDescription}\\t%c${extraInfo}`,\n 'font-weight:bold', 'color:red', 'color:blue', 'color: orange',\n 'color: green', 'color: steelblue');\n }\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {NamedTensorMap} from './tensor_types';\nimport * as util from './util';\n\nexport interface TapeNode {\n id: number;\n kernelName: string;\n outputs: Tensor[];\n inputs: NamedTensorMap;\n // Optional params, defined only for ops with gradient impl.\n gradient?: (dys: Tensor[]) => NamedGradientMap;\n saved?: Tensor[];\n}\n\nexport type NamedGradientMap = {\n [inputName: string]: () => Tensor;\n};\n\n/**\n * Computes a list of TapeNodes that connect x to y, filtering everything else\n * out and preserving the order of the original tape elements.\n *\n * @param tape The tape elements to filter.\n * @param xs The input Tensors.\n * @param y The output Tensor.\n */\nexport function getFilteredNodesXToY(\n tape: TapeNode[], xs: Tensor[], y: Tensor): TapeNode[] {\n // Forward pass to compute all the nodes and Tensors that are transitively a\n // function of x.\n const tensorsFromX: {[tensorId: number]: boolean} = {};\n const nodesFromX: {[nodeId: number]: boolean} = {};\n for (let i = 0; i < xs.length; i++) {\n tensorsFromX[xs[i].id] = true;\n }\n\n for (let i = 0; i < tape.length; i++) {\n const node = tape[i];\n const nodeInputs = node.inputs;\n for (const inputName in nodeInputs) {\n const input = nodeInputs[inputName];\n\n let anyInputFromX = false;\n for (let j = 0; j < xs.length; j++) {\n if (tensorsFromX[input.id]) {\n node.outputs.forEach(output => tensorsFromX[output.id] = true);\n anyInputFromX = true;\n nodesFromX[node.id] = true;\n break;\n }\n }\n\n if (anyInputFromX) {\n break;\n }\n }\n }\n\n // Backward pass to find all of the nodes and Tensors that lead to y.\n const tensorsLeadToY: {[tensorId: number]: boolean} = {};\n tensorsLeadToY[y.id] = true;\n const nodesToY: {[nodeId: number]: boolean} = {};\n\n for (let i = tape.length - 1; i >= 0; i--) {\n const node = tape[i];\n const nodeInputs = node.inputs;\n\n // If any of the outputs lead to y, mark all of the inputs as leading to y.\n for (let j = 0; j < node.outputs.length; j++) {\n if (tensorsLeadToY[node.outputs[j].id]) {\n for (const inputName in nodeInputs) {\n tensorsLeadToY[nodeInputs[inputName].id] = true;\n nodesToY[node.id] = true;\n }\n break;\n }\n }\n }\n\n // Return the paths that come from x and lead to y.\n const filteredTape: TapeNode[] = [];\n for (let i = 0; i < tape.length; i++) {\n const node = tape[i];\n\n if (nodesFromX[node.id] && nodesToY[node.id]) {\n // Prune the inputs from the node that aren't a function of x.\n const prunedInputs: {[inputName: string]: Tensor} = {};\n for (const inputName in node.inputs) {\n const nodeInput = node.inputs[inputName];\n if (tensorsFromX[nodeInput.id]) {\n prunedInputs[inputName] = nodeInput;\n }\n }\n\n // Copy the node and overwrite inputsAndArgs to the pruned version.\n const prunedNode = Object.assign({}, node);\n prunedNode.inputs = prunedInputs;\n prunedNode.outputs = node.outputs;\n\n filteredTape.push(prunedNode);\n }\n }\n\n return filteredTape;\n}\n\n/**\n * Backpropagate gradients through the filtered TapeNodes.\n *\n * @param tensorAccumulatedGradientMap A map of Tensor to its gradient. This map\n * is mutated by this method.\n * @param filteredTape The filtered TapeNodes to backprop through.\n */\nexport function backpropagateGradients(\n tensorAccumulatedGradientMap: {[tensorId: number]: Tensor},\n filteredTape: TapeNode[], tidy: (f: Function) => Tensor,\n add: (a: Tensor, b: Tensor) => Tensor) {\n // Walk the tape backward and keep a map of Tensor to its gradient.\n for (let i = filteredTape.length - 1; i >= 0; i--) {\n const node = filteredTape[i];\n\n const dys: Tensor[] = [];\n node.outputs.forEach(o => {\n const gradTensor = tensorAccumulatedGradientMap[o.id];\n if (gradTensor != null) {\n dys.push(gradTensor);\n } else {\n // This particular output is not in the back-propagation subgraph, so it\n // does not affect the final output, thus we put null for its dy.\n dys.push(null);\n }\n });\n\n if (node.gradient == null) {\n throw new Error(\n `Cannot compute gradient: gradient function not found ` +\n `for ${node.kernelName}.`);\n }\n\n // Backprop dy through this node and accumulate gradients over the inputs.\n const inputGradients = node.gradient(dys);\n\n for (const inputName in node.inputs) {\n if (!(inputName in inputGradients)) {\n throw new Error(\n `Cannot backprop through input ${inputName}. ` +\n `Available gradients found: ${Object.keys(inputGradients)}.`);\n }\n\n // Call the gradient function.\n const dx = tidy(() => inputGradients[inputName]());\n if (dx.dtype !== 'float32') {\n throw new Error(\n `Error in gradient for op ${\n node.kernelName}. The gradient of input ` +\n `${inputName} must have 'float32' dtype, but has '${dx.dtype}'`);\n }\n const x = node.inputs[inputName];\n if (!util.arraysEqual(dx.shape, x.shape)) {\n throw new Error(\n `Error in gradient for op ${\n node.kernelName}. The gradient of input ` +\n `'${inputName}' has shape '${dx.shape}', which does not match ` +\n `the shape of the input '${x.shape}'`);\n }\n\n if (tensorAccumulatedGradientMap[x.id] == null) {\n tensorAccumulatedGradientMap[x.id] = dx;\n } else {\n const curGradient = tensorAccumulatedGradientMap[x.id];\n tensorAccumulatedGradientMap[x.id] = add(curGradient, dx);\n curGradient.dispose();\n }\n }\n }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {DataType, TypedArray} from './types';\nimport {computeStrides, isString, rightPad, sizeFromShape} from './util';\n\n// Maximum number of values before we decide to show ellipsis.\nconst FORMAT_LIMIT_NUM_VALS = 20;\n// Number of first and last values to show when displaying a, b,...,y, z.\nconst FORMAT_NUM_FIRST_LAST_VALS = 3;\n// Number of significant digits to show.\nconst FORMAT_NUM_SIG_DIGITS = 7;\n\nexport function tensorToString(\n vals: TypedArray|string[], shape: number[], dtype: DataType,\n verbose: boolean) {\n const strides = computeStrides(shape);\n const padPerCol = computeMaxSizePerColumn(vals, shape, dtype, strides);\n const rank = shape.length;\n const valsLines = subTensorToString(vals, shape, dtype, strides, padPerCol);\n const lines = ['Tensor'];\n if (verbose) {\n lines.push(` dtype: ${dtype}`);\n lines.push(` rank: ${rank}`);\n lines.push(` shape: [${shape}]`);\n lines.push(` values:`);\n }\n lines.push(valsLines.map(l => ' ' + l).join('\\n'));\n return lines.join('\\n');\n}\n\nfunction computeMaxSizePerColumn(\n vals: TypedArray|string[], shape: number[], dtype: DataType,\n strides: number[]): number[] {\n const n = sizeFromShape(shape);\n const numCols = strides[strides.length - 1];\n const padPerCol = new Array(numCols).fill(0);\n const rank = shape.length;\n const valuesOrTuples =\n dtype === 'complex64' ? createComplexTuples(vals) : vals;\n\n if (rank > 1) {\n for (let row = 0; row < n / numCols; row++) {\n const offset = row * numCols;\n for (let j = 0; j < numCols; j++) {\n padPerCol[j] = Math.max(\n padPerCol[j],\n valToString(valuesOrTuples[offset + j], 0, dtype).length);\n }\n }\n }\n return padPerCol;\n}\n\nfunction valToString(\n val: number|string|[number, number], pad: number, dtype: DataType) {\n let valStr: string;\n if (Array.isArray(val)) {\n valStr = `${parseFloat(val[0].toFixed(FORMAT_NUM_SIG_DIGITS))} + ` +\n `${parseFloat(val[1].toFixed(FORMAT_NUM_SIG_DIGITS))}j`;\n } else if (isString(val)) {\n valStr = `'${val}'`;\n } else if (dtype === 'bool') {\n valStr = boolNumToString(val);\n } else {\n valStr = parseFloat(val.toFixed(FORMAT_NUM_SIG_DIGITS)).toString();\n }\n\n return rightPad(valStr, pad);\n}\n\nfunction boolNumToString(v: number): string {\n return v === 0 ? 'false' : 'true';\n}\n\nfunction subTensorToString(\n vals: TypedArray|string[], shape: number[], dtype: DataType,\n strides: number[], padPerCol: number[], isLast = true): string[] {\n const storagePerElement = dtype === 'complex64' ? 2 : 1;\n\n const size = shape[0];\n const rank = shape.length;\n if (rank === 0) {\n if (dtype === 'complex64') {\n const complexTuple = createComplexTuples(vals);\n return [valToString(complexTuple[0], 0, dtype)];\n }\n if (dtype === 'bool') {\n return [boolNumToString(vals[0] as number)];\n }\n return [vals[0].toString()];\n }\n\n if (rank === 1) {\n if (size > FORMAT_LIMIT_NUM_VALS) {\n const firstValsSize = FORMAT_NUM_FIRST_LAST_VALS * storagePerElement;\n\n let firstVals = Array.from(\n vals.slice(0, firstValsSize));\n let lastVals = Array.from(vals.slice(\n (size - FORMAT_NUM_FIRST_LAST_VALS) * storagePerElement,\n size * storagePerElement));\n if (dtype === 'complex64') {\n firstVals = createComplexTuples(firstVals);\n lastVals = createComplexTuples(lastVals);\n }\n return [\n '[' +\n firstVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n .join(', ') +\n ', ..., ' +\n lastVals\n .map(\n (x, i) => valToString(\n x, padPerCol[size - FORMAT_NUM_FIRST_LAST_VALS + i], dtype))\n .join(', ') +\n ']'\n ];\n }\n const displayVals: Array =\n dtype === 'complex64' ? createComplexTuples(vals) :\n Array.from(vals);\n\n return [\n '[' +\n displayVals.map((x, i) => valToString(x, padPerCol[i], dtype))\n .join(', ') +\n ']'\n ];\n }\n\n // The array is rank 2 or more.\n const subshape = shape.slice(1);\n const substrides = strides.slice(1);\n const stride = strides[0] * storagePerElement;\n const lines: string[] = [];\n if (size > FORMAT_LIMIT_NUM_VALS) {\n for (let i = 0; i < FORMAT_NUM_FIRST_LAST_VALS; i++) {\n const start = i * stride;\n const end = start + stride;\n lines.push(...subTensorToString(\n vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n false /* isLast */));\n }\n lines.push('...');\n for (let i = size - FORMAT_NUM_FIRST_LAST_VALS; i < size; i++) {\n const start = i * stride;\n const end = start + stride;\n lines.push(...subTensorToString(\n vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n i === size - 1 /* isLast */));\n }\n } else {\n for (let i = 0; i < size; i++) {\n const start = i * stride;\n const end = start + stride;\n lines.push(...subTensorToString(\n vals.slice(start, end), subshape, dtype, substrides, padPerCol,\n i === size - 1 /* isLast */));\n }\n }\n const sep = rank === 2 ? ',' : '';\n lines[0] = '[' + lines[0] + sep;\n for (let i = 1; i < lines.length - 1; i++) {\n lines[i] = ' ' + lines[i] + sep;\n }\n let newLineSep = ',\\n';\n for (let i = 2; i < rank; i++) {\n newLineSep += '\\n';\n }\n lines[lines.length - 1] =\n ' ' + lines[lines.length - 1] + ']' + (isLast ? '' : newLineSep);\n return lines;\n}\n\nfunction createComplexTuples(vals: Array<{}>|\n TypedArray): Array<[number, number]> {\n const complexTuples: Array<[number, number]> = [];\n for (let i = 0; i < vals.length; i += 2) {\n complexTuples.push([vals[i], vals[i + 1]] as [number, number]);\n }\n return complexTuples;\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {tensorToString} from './tensor_format';\nimport {ArrayMap, BackendValues, DataType, DataTypeMap, DataValues, NumericDataType, Rank, ShapeMap, SingleValueMap, TypedArray} from './types';\nimport * as util from './util';\nimport {computeStrides, toNestedArray} from './util';\n\nexport interface TensorData {\n dataId?: DataId;\n values?: DataTypeMap[D];\n}\n\n// This interface mimics KernelBackend (in backend.ts), which would create a\n// circular dependency if imported.\nexport interface Backend {}\n\n/**\n * A mutable object, similar to `tf.Tensor`, that allows users to set values\n * at locations before converting to an immutable `tf.Tensor`.\n *\n * See `tf.buffer` for creating a tensor buffer.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class TensorBuffer {\n size: number;\n shape: ShapeMap[R];\n strides: number[];\n values: DataTypeMap[D];\n\n constructor(shape: ShapeMap[R], public dtype: D, values?: DataTypeMap[D]) {\n this.shape = shape.slice() as ShapeMap[R];\n this.size = util.sizeFromShape(shape);\n\n if (values != null) {\n const n = values.length;\n util.assert(\n n === this.size,\n () => `Length of values '${n}' does not match the size ` +\n `inferred by the shape '${this.size}'.`);\n }\n if (dtype === 'complex64') {\n throw new Error(\n `complex64 dtype TensorBuffers are not supported. Please create ` +\n `a TensorBuffer for the real and imaginary parts separately and ` +\n `call tf.complex(real, imag).`);\n }\n this.values = values || util.getArrayFromDType(dtype, this.size);\n this.strides = computeStrides(shape);\n }\n\n /**\n * Sets a value in the buffer at a given location.\n *\n * @param value The value to set.\n * @param locs The location indices.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\n set(value: SingleValueMap[D], ...locs: number[]): void {\n if (locs.length === 0) {\n locs = [0];\n }\n util.assert(\n locs.length === this.rank,\n () => `The number of provided coordinates (${locs.length}) must ` +\n `match the rank (${this.rank})`);\n\n const index = this.locToIndex(locs);\n this.values[index] = value as number;\n }\n\n /**\n * Returns the value in the buffer at the provided location.\n *\n * @param locs The location indices.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\n get(...locs: number[]): SingleValueMap[D] {\n if (locs.length === 0) {\n locs = [0];\n }\n let i = 0;\n for (const loc of locs) {\n if (loc < 0 || loc >= this.shape[i]) {\n const msg = `Requested out of range element at ${locs}. ` +\n ` Buffer shape=${this.shape}`;\n throw new Error(msg);\n }\n i++;\n }\n let index = locs[locs.length - 1];\n for (let i = 0; i < locs.length - 1; ++i) {\n index += this.strides[i] * locs[i];\n }\n return this.values[index] as SingleValueMap[D];\n }\n\n locToIndex(locs: number[]): number {\n if (this.rank === 0) {\n return 0;\n } else if (this.rank === 1) {\n return locs[0];\n }\n let index = locs[locs.length - 1];\n for (let i = 0; i < locs.length - 1; ++i) {\n index += this.strides[i] * locs[i];\n }\n return index;\n }\n\n indexToLoc(index: number): number[] {\n if (this.rank === 0) {\n return [];\n } else if (this.rank === 1) {\n return [index];\n }\n const locs: number[] = new Array(this.shape.length);\n for (let i = 0; i < locs.length - 1; ++i) {\n locs[i] = Math.floor(index / this.strides[i]);\n index -= locs[i] * this.strides[i];\n }\n locs[locs.length - 1] = index;\n return locs;\n }\n\n get rank() {\n return this.shape.length;\n }\n\n /**\n * Creates an immutable `tf.Tensor` object from the buffer.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\n toTensor(): Tensor {\n return trackerFn().makeTensor(this.values, this.shape, this.dtype) as\n Tensor;\n }\n}\n\nexport interface TensorTracker {\n makeTensor(\n values: DataValues, shape: number[], dtype: DataType,\n backend?: Backend): Tensor;\n makeVariable(\n initialValue: Tensor, trainable?: boolean, name?: string,\n dtype?: DataType): Variable;\n incRef(a: Tensor, backend: Backend): void;\n disposeTensor(t: Tensor): void;\n disposeVariable(v: Variable): void;\n read(dataId: DataId): Promise;\n readSync(dataId: DataId): BackendValues;\n}\n\n/**\n * The Tensor class calls into this handler to delegate chaining operations.\n */\nexport interface OpHandler {\n cast(x: T, dtype: DataType): T;\n buffer(\n shape: ShapeMap[R], dtype: D,\n values?: DataTypeMap[D]): TensorBuffer;\n print(x: T, verbose: boolean): void;\n clone(x: T): T;\n // TODO(yassogba) bring reshape back?\n}\n\n// For tracking tensor creation and disposal.\nlet trackerFn: () => TensorTracker = null;\n// Used by chaining methods to call into ops.\nlet opHandler: OpHandler = null;\n// Used to warn about deprecated methods.\nlet deprecationWarningFn: (msg: string) => void = null;\n// This here so that we can use this method on dev branches and keep the\n// functionality at master.\n// tslint:disable-next-line:no-unused-expression\n[deprecationWarningFn];\n\n/**\n * An external consumer can register itself as the tensor tracker. This way\n * the Tensor class can notify the tracker for every tensor created and\n * disposed.\n */\nexport function setTensorTracker(fn: () => TensorTracker) {\n trackerFn = fn;\n}\n\n/**\n * An external consumer can register itself as the op handler. This way the\n * Tensor class can have chaining methods that call into ops via the op\n * handler.\n */\nexport function setOpHandler(handler: OpHandler) {\n opHandler = handler;\n}\n\n/**\n * Sets the deprecation warning function to be used by this file. This way the\n * Tensor class can be a leaf but still use the environment.\n */\nexport function setDeprecationWarningFn(fn: (msg: string) => void) {\n deprecationWarningFn = fn;\n}\n\n/**\n * We wrap data id since we use weak map to avoid memory leaks.\n * Since we have our own memory management, we have a reference counter\n * mapping a tensor to its data, so there is always a pointer (even if that\n * data is otherwise garbage collectable).\n * See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/\n * Global_Objects/WeakMap\n */\nexport type DataId = object; // object instead of {} to force non-primitive.\n\n// Declare this namespace to make Tensor class augmentation work in google3.\nexport declare namespace Tensor {}\n/**\n * A `tf.Tensor` object represents an immutable, multidimensional array of\n * numbers that has a shape and a data type.\n *\n * See `tf.tensor` for details on how to create a `tf.Tensor`.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Tensor {\n /** Unique id of this tensor. */\n readonly id: number;\n /**\n * Id of the bucket holding the data for this tensor. Multiple arrays can\n * point to the same bucket (e.g. when calling array.reshape()).\n */\n dataId: DataId;\n /** The shape of the tensor. */\n readonly shape: ShapeMap[R];\n /** Number of elements in the tensor. */\n readonly size: number;\n /** The data type for the array. */\n readonly dtype: DataType;\n /** The rank type for the array (see `Rank` enum). */\n readonly rankType: R;\n\n /** Whether this tensor has been globally kept. */\n kept = false;\n /** The id of the scope this tensor is being tracked in. */\n scopeId: number;\n\n /**\n * Number of elements to skip in each dimension when indexing. See\n * https://docs.scipy.org/doc/numpy/reference/generated/\\\n * numpy.ndarray.strides.html\n */\n readonly strides: number[];\n\n constructor(shape: ShapeMap[R], dtype: DataType, dataId: DataId, id: number) {\n this.shape = shape.slice() as ShapeMap[R];\n this.dtype = dtype || 'float32';\n this.size = util.sizeFromShape(shape);\n this.strides = computeStrides(shape);\n this.dataId = dataId;\n this.id = id;\n this.rankType = (this.rank < 5 ? this.rank.toString() : 'higher') as R;\n }\n\n get rank(): number {\n return this.shape.length;\n }\n\n /**\n * Returns a promise of `tf.TensorBuffer` that holds the underlying data.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n async buffer(): Promise> {\n const vals = await this.data();\n return opHandler.buffer(this.shape, this.dtype as D, vals);\n }\n\n /**\n * Returns a `tf.TensorBuffer` that holds the underlying data.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n bufferSync(): TensorBuffer {\n return opHandler.buffer(this.shape, this.dtype as D, this.dataSync());\n }\n\n /**\n * Returns the tensor data as a nested array. The transfer of data is done\n * asynchronously.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n async array(): Promise {\n const vals = await this.data();\n return toNestedArray(this.shape, vals) as ArrayMap[R];\n }\n\n /**\n * Returns the tensor data as a nested array. The transfer of data is done\n * synchronously.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n arraySync(): ArrayMap[R] {\n return toNestedArray(this.shape, this.dataSync()) as ArrayMap[R];\n }\n\n /**\n * Asynchronously downloads the values from the `tf.Tensor`. Returns a\n * promise of `TypedArray` that resolves when the computation has finished.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n async data(): Promise {\n this.throwIfDisposed();\n const data = trackerFn().read(this.dataId);\n if (this.dtype === 'string') {\n const bytes = await data as Uint8Array[];\n try {\n return bytes.map(b => util.decodeString(b)) as DataTypeMap[D];\n } catch {\n throw new Error(\n 'Failed to decode the string bytes into utf-8. ' +\n 'To get the original bytes, call tensor.bytes().');\n }\n }\n return data as Promise;\n }\n\n /**\n * Synchronously downloads the values from the `tf.Tensor`. This blocks the\n * UI thread until the values are ready, which can cause performance issues.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n dataSync(): DataTypeMap[D] {\n this.throwIfDisposed();\n const data = trackerFn().readSync(this.dataId);\n if (this.dtype === 'string') {\n try {\n return (data as Uint8Array[]).map(b => util.decodeString(b)) as\n DataTypeMap[D];\n } catch {\n throw new Error(\n 'Failed to decode the string bytes into utf-8. ' +\n 'To get the original bytes, call tensor.bytes().');\n }\n }\n return data as DataTypeMap[D];\n }\n\n /** Returns the underlying bytes of the tensor's data. */\n async bytes(): Promise {\n this.throwIfDisposed();\n const data = await trackerFn().read(this.dataId);\n if (this.dtype === 'string') {\n return data as Uint8Array[];\n } else {\n return new Uint8Array((data as TypedArray).buffer);\n }\n }\n\n /**\n * Disposes `tf.Tensor` from memory.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n dispose(): void {\n if (this.isDisposed) {\n return;\n }\n trackerFn().disposeTensor(this);\n this.isDisposedInternal = true;\n }\n\n protected isDisposedInternal = false;\n get isDisposed(): boolean {\n return this.isDisposedInternal;\n }\n\n throwIfDisposed() {\n if (this.isDisposed) {\n throw new Error(`Tensor is disposed.`);\n }\n }\n\n /**\n * Prints the `tf.Tensor`. See `tf.print` for details.\n *\n * @param verbose Whether to print verbose information about the tensor,\n * including dtype and size.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n print(verbose = false): void {\n return opHandler.print(this, verbose);\n }\n\n /**\n * Returns a copy of the tensor. See `tf.clone` for details.\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n clone(this: T): T {\n this.throwIfDisposed();\n return opHandler.clone(this);\n }\n\n /**\n * Returns a human-readable description of the tensor. Useful for logging.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n toString(verbose = false): string {\n const vals = this.dataSync();\n return tensorToString(vals, this.shape, this.dtype, verbose);\n }\n\n cast(dtype: DataType): T {\n this.throwIfDisposed();\n return opHandler.cast(this as T, dtype);\n }\n variable(trainable = true, name?: string, dtype?: DataType): Variable {\n this.throwIfDisposed();\n return trackerFn().makeVariable(this, trainable, name, dtype) as\n Variable;\n }\n}\nObject.defineProperty(Tensor, Symbol.hasInstance, {\n value: (instance: Tensor) => {\n // Implementation note: we should use properties of the object that will be\n // defined before the constructor body has finished executing (methods).\n // This is because when this code is transpiled by babel, babel will call\n // classCallCheck before the constructor body is run.\n // See https://github.com/tensorflow/tfjs/issues/3384 for backstory.\n return !!instance && instance.data != null && instance.dataSync != null &&\n instance.throwIfDisposed != null;\n }\n});\n\nexport interface NumericTensor extends Tensor {\n dtype: NumericDataType;\n dataSync(): DataTypeMap[D];\n data(): Promise;\n}\n\nexport interface StringTensor extends Tensor {\n dtype: 'string';\n dataSync(): DataTypeMap[D];\n data(): Promise;\n}\n\n/** @doclink Tensor */\nexport type Scalar = Tensor;\n/** @doclink Tensor */\nexport type Tensor1D = Tensor;\n/** @doclink Tensor */\nexport type Tensor2D = Tensor;\n/** @doclink Tensor */\nexport type Tensor3D = Tensor;\n/** @doclink Tensor */\nexport type Tensor4D = Tensor;\n/** @doclink Tensor */\nexport type Tensor5D = Tensor;\n/** @doclink Tensor */\nexport type Tensor6D = Tensor;\n\n/**\n * A mutable `tf.Tensor`, useful for persisting state, e.g. for training.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\nexport class Variable extends Tensor {\n name: string;\n\n constructor(\n initialValue: Tensor, public trainable: boolean, name: string,\n tensorId: number) {\n super(\n initialValue.shape, initialValue.dtype, initialValue.dataId, tensorId);\n this.name = name;\n }\n\n /**\n * Assign a new `tf.Tensor` to this variable. The new `tf.Tensor` must have\n * the same shape and dtype as the old `tf.Tensor`.\n *\n * @param newValue New tensor to be assigned to this variable.\n *\n * @doc {heading: 'Tensors', subheading: 'Classes'}\n */\n assign(newValue: Tensor): void {\n if (newValue.dtype !== this.dtype) {\n throw new Error(\n `dtype of the new value (${newValue.dtype}) and ` +\n `previous value (${this.dtype}) must match`);\n }\n if (!util.arraysEqual(newValue.shape, this.shape)) {\n throw new Error(\n `shape of the new value (${newValue.shape}) and ` +\n `previous value (${this.shape}) must match`);\n }\n trackerFn().disposeTensor(this);\n this.dataId = newValue.dataId;\n trackerFn().incRef(this, null /* backend */);\n }\n\n dispose(): void {\n trackerFn().disposeVariable(this);\n this.isDisposedInternal = true;\n }\n}\n\nObject.defineProperty(Variable, Symbol.hasInstance, {\n value: (instance: Variable) => {\n return instance instanceof Tensor && instance.assign != null &&\n instance.assign instanceof Function;\n }\n});\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/** @docalias number[] */\nexport interface ShapeMap {\n R0: number[];\n R1: [number];\n R2: [number, number];\n R3: [number, number, number];\n R4: [number, number, number, number];\n R5: [number, number, number, number, number];\n R6: [number, number, number, number, number, number];\n}\n\n/** @docalias number[] */\nexport interface ArrayMap {\n R0: number;\n R1: number[];\n R2: number[][];\n R3: number[][][];\n R4: number[][][][];\n R5: number[][][][][];\n R6: number[][][][][][];\n}\n\nexport interface DataTypeMap {\n float32: Float32Array;\n int32: Int32Array;\n bool: Uint8Array;\n complex64: Float32Array;\n string: string[];\n}\n\nexport interface SingleValueMap {\n bool: boolean;\n int32: number;\n float32: number;\n complex64: number;\n string: string;\n}\n\n/** @docalias 'float32'|'int32'|'bool'|'complex64'|'string' */\nexport type DataType = keyof DataTypeMap;\nexport type NumericDataType = 'float32'|'int32'|'bool'|'complex64';\nexport type TypedArray = Float32Array|Int32Array|Uint8Array;\n/** Tensor data used in tensor creation and user-facing API. */\nexport type DataValues = DataTypeMap[DataType];\n/** The underlying tensor data that gets stored in a backend. */\nexport type BackendValues = Float32Array|Int32Array|Uint8Array|Uint8Array[];\n\nexport enum Rank {\n R0 = 'R0',\n R1 = 'R1',\n R2 = 'R2',\n R3 = 'R3',\n R4 = 'R4',\n R5 = 'R5',\n R6 = 'R6'\n}\n\nexport type FlatVector = boolean[]|number[]|TypedArray;\nexport type RegularArray =\n T[]|T[][]|T[][][]|T[][][][]|T[][][][][]|T[][][][][][];\n\n// tslint:disable-next-line:no-any\nexport interface RecursiveArray {\n [index: number]: T|RecursiveArray;\n}\n\n// Looks for upcasting types. Used, for example, in operations with mixed dtype\n// inputs.\nenum UpcastInt32AndMap {\n 'float32' = 'float32',\n 'int32' = 'int32',\n 'bool' = 'int32',\n 'complex64' = 'complex64'\n}\n\nenum UpcastBoolAndMap {\n 'float32' = 'float32',\n 'int32' = 'int32',\n 'bool' = 'bool',\n 'complex64' = 'complex64'\n}\n\nenum UpcastFloat32AndMap {\n 'float32' = 'float32',\n 'int32' = 'float32',\n 'bool' = 'float32',\n 'complex64' = 'complex64'\n}\n\nenum UpcastComplex64AndMap {\n 'float32' = 'complex64',\n 'int32' = 'complex64',\n 'bool' = 'complex64',\n 'complex64' = 'complex64'\n}\n\nconst upcastTypeMap = {\n 'float32': UpcastFloat32AndMap,\n 'int32': UpcastInt32AndMap,\n 'bool': UpcastBoolAndMap,\n 'complex64': UpcastComplex64AndMap\n};\n\nexport function upcastType(typeA: DataType, typeB: DataType): DataType {\n if (typeA === 'string' || typeB === 'string') {\n if (typeA === 'string' && typeB === 'string') {\n return 'string';\n }\n throw new Error(`Can not upcast ${typeA} with ${typeB}`);\n }\n return upcastTypeMap[typeA][typeB];\n}\n\n/** Returns the output type after summation. */\nexport function sumOutType(type: DataType): DataType {\n return upcastType(type, 'int32');\n}\n\n/** @docalias TypedArray|Array */\nexport type TensorLike =\n TypedArray|number|boolean|string|RecursiveArray|\n RecursiveArray|RecursiveArray|Uint8Array[];\nexport type ScalarLike = number|boolean|string|Uint8Array;\n/** @docalias TypedArray|Array */\nexport type TensorLike1D = TypedArray|number[]|boolean[]|string[]|Uint8Array[];\n/** @docalias TypedArray|Array */\nexport type TensorLike2D = TypedArray|number[]|number[][]|boolean[]|boolean[][]|\n string[]|string[][]|Uint8Array[]|Uint8Array[][];\n/** @docalias TypedArray|Array */\nexport type TensorLike3D = TypedArray|number[]|number[][][]|boolean[]|\n boolean[][][]|string[]|string[][][]|Uint8Array[]|Uint8Array[][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike4D = TypedArray|number[]|number[][][][]|boolean[]|\n boolean[][][][]|string[]|string[][][][]|Uint8Array[]|Uint8Array[][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike5D =\n TypedArray|number[]|number[][][][][]|boolean[]|boolean[][][][][]|string[]|\n string[][][][][]|Uint8Array[]|Uint8Array[][][][][];\n/** @docalias TypedArray|Array */\nexport type TensorLike6D =\n TypedArray|number[]|number[][][][][][]|boolean[]|boolean[][][][][][]|\n string[]|string[][][][][][]|Uint8Array[]|Uint8Array[][][][][];\n\n/** Type for representing image dat in Uint8Array type. */\nexport interface PixelData {\n width: number;\n height: number;\n data: Uint8Array;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from './tensor';\nimport {TensorContainer, TensorContainerArray} from './tensor_types';\nimport {upcastType} from './types';\nimport {assert} from './util';\n\nexport function makeTypesMatch(a: T, b: T): [T, T] {\n if (a.dtype === b.dtype) {\n return [a, b];\n }\n const dtype = upcastType(a.dtype, b.dtype);\n return [a.cast(dtype), b.cast(dtype)];\n}\n\nexport function assertTypesMatch(a: Tensor, b: Tensor): void {\n assert(\n a.dtype === b.dtype,\n () => `The dtypes of the first(${a.dtype}) and` +\n ` second(${b.dtype}) input must match`);\n}\n\nexport function isTensorInList(tensor: Tensor, tensorList: Tensor[]): boolean {\n return tensorList.some(x => x.id === tensor.id);\n}\n\n/**\n * Extracts any `Tensor`s found within the provided object.\n *\n * @param container an object that may be a `Tensor` or may directly contain\n * `Tensor`s, such as a `Tensor[]` or `{key: Tensor, ...}`. In general it\n * is safe to pass any object here, except that `Promise`s are not\n * supported.\n * @returns An array of `Tensors` found within the passed object. If the\n * argument is simply a `Tensor', a list containing that `Tensor` is\n * returned. If the object is not a `Tensor` or does not\n * contain `Tensors`, an empty list is returned.\n */\nexport function getTensorsInContainer(result: TensorContainer): Tensor[] {\n const list: Tensor[] = [];\n const seen = new Set<{}|void>();\n walkTensorContainer(result, list, seen);\n return list;\n}\n\nfunction walkTensorContainer(\n container: TensorContainer, list: Tensor[], seen: Set<{}|void>): void {\n if (container == null) {\n return;\n }\n if (container instanceof Tensor) {\n list.push(container);\n return;\n }\n if (!isIterable(container)) {\n return;\n }\n // Iteration over keys works also for arrays.\n const iterable = container as TensorContainerArray;\n for (const k in iterable) {\n const val = iterable[k];\n if (!seen.has(val)) {\n seen.add(val);\n walkTensorContainer(val, list, seen);\n }\n }\n}\n\n// tslint:disable-next-line:no-any\nfunction isIterable(obj: any): boolean {\n return Array.isArray(obj) || typeof obj === 'object';\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {BackendTimingInfo, DataMover, KernelBackend} from './backends/backend';\nimport {Environment, setEnvironmentGlobal} from './environment';\nimport {getGlobalNamespace} from './global_util';\nimport {Add, Cast} from './kernel_names';\nimport {getGradient, getKernel, getKernelsForBackend, GradFunc, NamedAttrMap, TensorInfo} from './kernel_registry';\nimport {KernelProfile, Profiler} from './profiler';\nimport {backpropagateGradients, getFilteredNodesXToY, TapeNode} from './tape';\nimport {DataId, setTensorTracker, Tensor, TensorTracker, Variable} from './tensor';\nimport {GradSaveFunc, NamedTensorMap, NamedVariableMap, TensorContainer} from './tensor_types';\nimport {getTensorsInContainer} from './tensor_util';\nimport {BackendValues, DataType, DataValues} from './types';\nimport * as util from './util';\nimport {bytesFromStringArray, makeOnesTypedArray, now, sizeFromShape} from './util';\n\n/**\n * A function that computes an output. The save function is for saving tensors\n * computed in the forward pass, that we need in the backward pass.\n */\nexport type ForwardFunc = (backend: KernelBackend, save?: GradSaveFunc) => T;\n\n/**\n * @docalias (a: Tensor, b: Tensor,..., save?: Function) => {\n * value: Tensor,\n * gradFunc: (dy: Tensor, saved?: NamedTensorMap) => Tensor | Tensor[]\n * }\n */\nexport type CustomGradientFunc =\n (...inputs: Array) => {\n value: T;\n gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[];\n };\n\nexport type MemoryInfo = {\n numTensors: number; numDataBuffers: number; numBytes: number;\n unreliable?: boolean; reasons: string[];\n};\n\ntype KernelInfo = {\n name: string; bytesAdded: number; totalBytesSnapshot: number;\n tensorsAdded: number;\n totalTensorsSnapshot: number;\n inputShapes: number[][];\n outputShapes: number[][];\n kernelTimeMs: number | {error: string} | Promise;\n extraInfo: string | Promise;\n};\n\nexport type ProfileInfo = {\n newBytes: number; newTensors: number; peakBytes: number;\n kernels: KernelInfo[];\n result: TensorContainer;\n};\n\nexport interface TimingInfo extends BackendTimingInfo {\n wallMs: number;\n}\n\n/** @docalias Function */\nexport type ScopeFn = () => T;\n\ninterface ScopeState {\n track: Tensor[];\n name: string;\n id: number;\n}\n\nclass EngineState {\n // Public since optimizers will use it.\n registeredVariables: NamedVariableMap = {};\n\n nextTapeNodeId = 0;\n numBytes = 0;\n numTensors = 0;\n numStringTensors = 0;\n numDataBuffers = 0;\n\n activeTape: TapeNode[];\n // Number of nested tf.grad() statements when computing higher-order\n // gradients. E.g. `1` for first-order gradients and `2` for second-order\n // gradients. Used to track if the tape should be removed after a backprop.\n gradientDepth = 0;\n // Number of nested kernel calls. When kernel depth is greater than 1, we turn\n // off the tape.\n kernelDepth = 0;\n\n // Keep Tensors that parallel the tapes.\n activeScope: ScopeState;\n scopeStack: ScopeState[] = [];\n /**\n * Keeps track of the number of data moves during a kernel execution. We\n * maintain a stack since kernels can call other kernels, recursively.\n */\n numDataMovesStack: number[] = [];\n nextScopeId = 0;\n\n tensorInfo = new WeakMap();\n\n profiling = false;\n activeProfile: ProfileInfo =\n {newBytes: 0, newTensors: 0, peakBytes: 0, kernels: [], result: null};\n\n dispose() {\n for (const variableName in this.registeredVariables) {\n this.registeredVariables[variableName].dispose();\n }\n }\n}\n\nexport class Engine implements TensorTracker, DataMover {\n state: EngineState;\n backendName: string;\n registry: {[id: string]: KernelBackend} = {};\n registryFactory: {\n [id: string]: {\n factory: () => KernelBackend | Promise,\n priority: number\n }\n } = {};\n\n private profiler: Profiler;\n private backendInstance: KernelBackend;\n private pendingBackendInit: Promise;\n private pendingBackendInitId = 0;\n\n constructor(public ENV: Environment) {\n this.state = new EngineState();\n }\n\n async ready(): Promise {\n if (this.pendingBackendInit != null) {\n return this.pendingBackendInit.then(() => {});\n }\n if (this.backendInstance != null) {\n return;\n }\n const sortedBackends = this.getSortedBackends();\n\n for (let i = 0; i < sortedBackends.length; i++) {\n const backendName = sortedBackends[i];\n const success = await this.initializeBackend(backendName).success;\n if (success) {\n await this.setBackend(backendName);\n return;\n }\n }\n\n throw new Error(\n `Could not initialize any backends, all backend initializations ` +\n `failed.`);\n }\n\n get backend(): KernelBackend {\n if (this.pendingBackendInit != null) {\n throw new Error(\n `Backend '${this.backendName}' has not yet been initialized. Make ` +\n `sure to await tf.ready() or await tf.setBackend() before calling ` +\n `other methods`);\n }\n if (this.backendInstance == null) {\n const {name, asyncInit} = this.initializeBackendsAndReturnBest();\n if (asyncInit) {\n throw new Error(\n `The highest priority backend '${name}' has not yet been ` +\n `initialized. Make sure to await tf.ready() or ` +\n `await tf.setBackend() before calling other methods`);\n }\n this.setBackend(name);\n }\n return this.backendInstance;\n }\n\n backendNames(): string[] {\n return Object.keys(this.registryFactory);\n }\n\n findBackend(backendName: string): KernelBackend {\n if (!(backendName in this.registry)) {\n // If the backend hasn't been initialized but we have a registry entry for\n // it, initialize it and return it.\n if (backendName in this.registryFactory) {\n const {asyncInit} = this.initializeBackend(backendName);\n if (asyncInit) {\n // Backend is not ready yet.\n return null;\n }\n } else {\n return null;\n }\n }\n return this.registry[backendName];\n }\n\n findBackendFactory(backendName: string):\n () => KernelBackend | Promise {\n if (!(backendName in this.registryFactory)) {\n return null;\n }\n return this.registryFactory[backendName].factory;\n }\n\n registerBackend(\n backendName: string,\n factory: () => KernelBackend | Promise,\n priority = 1): boolean {\n if (backendName in this.registryFactory) {\n console.warn(\n `${backendName} backend was already registered. ` +\n `Reusing existing backend factory.`);\n return false;\n }\n this.registryFactory[backendName] = {factory, priority};\n return true;\n }\n\n async setBackend(backendName: string): Promise {\n if (this.registryFactory[backendName] == null) {\n throw new Error(`Backend name '${backendName}' not found in registry`);\n }\n this.backendName = backendName;\n if (this.registry[backendName] == null) {\n this.backendInstance = null;\n const {success, asyncInit} = this.initializeBackend(backendName);\n const result = asyncInit ? await success : success;\n if (!result) {\n return false;\n }\n }\n this.backendInstance = this.registry[backendName];\n this.setupRegisteredKernels();\n // Reset the profiler.\n this.profiler = new Profiler(this.backendInstance);\n\n return true;\n }\n\n private setupRegisteredKernels(): void {\n const kernels = getKernelsForBackend(this.backendName);\n kernels.forEach(kernel => {\n if (kernel.setupFunc != null) {\n kernel.setupFunc(this.backendInstance);\n }\n });\n }\n\n private disposeRegisteredKernels(backendName: string): void {\n const kernels = getKernelsForBackend(backendName);\n kernels.forEach(kernel => {\n if (kernel.disposeFunc != null) {\n kernel.disposeFunc(this.registry[backendName]);\n }\n });\n }\n\n /**\n * Initializes a backend by looking up the backend name in the factory\n * registry and calling the factory method. Returns a boolean representing\n * whether the initialization of the backend suceeded. Throws an error if\n * there is no backend in the factory registry.\n */\n private initializeBackend(backendName: string):\n {success: boolean|Promise, asyncInit: boolean} {\n const registryFactoryEntry = this.registryFactory[backendName];\n if (registryFactoryEntry == null) {\n throw new Error(\n `Cannot initialize backend ${backendName}, no registration found.`);\n }\n\n try {\n const backend = registryFactoryEntry.factory();\n /* Test if the factory returns a promise.\n Done in a more liberal way than\n previous 'Promise.resolve(backend)===backend'\n as we needed to account for custom Promise\n implementations (e.g. Angular) */\n if (backend && !(backend instanceof KernelBackend)\n && typeof backend.then === 'function') {\n const promiseId = ++this.pendingBackendInitId;\n const success =\n backend\n .then(backendInstance => {\n // Outdated promise. Another backend was set in the meantime.\n if (promiseId < this.pendingBackendInitId) {\n return false;\n }\n this.registry[backendName] = backendInstance;\n this.pendingBackendInit = null;\n return true;\n })\n .catch(err => {\n // Outdated promise. Another backend was set in the meantime.\n if (promiseId < this.pendingBackendInitId) {\n return false;\n }\n this.pendingBackendInit = null;\n console.warn(\n `Initialization of backend ${backendName} failed`);\n console.warn(err.stack || err.message);\n return false;\n });\n this.pendingBackendInit = success;\n return {success, asyncInit: true};\n } else {\n this.registry[backendName] = backend as KernelBackend;\n return {success: true, asyncInit: false};\n }\n } catch (err) {\n console.warn(`Initialization of backend ${backendName} failed`);\n console.warn(err.stack || err.message);\n return {success: false, asyncInit: false};\n }\n }\n\n removeBackend(backendName: string): void {\n if (!(backendName in this.registryFactory)) {\n throw new Error(`${backendName} backend not found in registry`);\n }\n if (this.backendName === backendName && this.pendingBackendInit != null) {\n // There is a pending promise of the backend we want to remove. Make it\n // obsolete.\n this.pendingBackendInitId++;\n }\n\n if (backendName in this.registry) {\n this.disposeRegisteredKernels(backendName);\n this.registry[backendName].dispose();\n delete this.registry[backendName];\n }\n\n delete this.registryFactory[backendName];\n\n // Unset the backend if it is active.\n if (this.backendName === backendName) {\n this.pendingBackendInit = null;\n this.backendName = null;\n this.backendInstance = null;\n }\n }\n\n private getSortedBackends(): string[] {\n if (Object.keys(this.registryFactory).length === 0) {\n throw new Error('No backend found in registry.');\n }\n return Object.keys(this.registryFactory).sort((a: string, b: string) => {\n // Highest priority comes first.\n return this.registryFactory[b].priority -\n this.registryFactory[a].priority;\n });\n }\n\n private initializeBackendsAndReturnBest():\n {name: string, asyncInit: boolean} {\n const sortedBackends = this.getSortedBackends();\n\n for (let i = 0; i < sortedBackends.length; i++) {\n const backendName = sortedBackends[i];\n const {success, asyncInit} = this.initializeBackend(backendName);\n if (asyncInit || success) {\n return {name: backendName, asyncInit};\n }\n }\n throw new Error(\n `Could not initialize any backends, all backend initializations ` +\n `failed.`);\n }\n\n moveData(backend: KernelBackend, dataId: DataId) {\n const info = this.state.tensorInfo.get(dataId);\n const srcBackend = info.backend;\n const values = this.readSync(dataId);\n // Delete the tensor from the old backend and move it to the new\n // backend.\n srcBackend.disposeData(dataId);\n info.backend = backend;\n backend.move(dataId, values, info.shape, info.dtype);\n if (this.shouldCheckForMemLeaks()) {\n // Track the number of moves during a kernel execution to correctly\n // detect memory leaks.\n this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1]++;\n }\n }\n\n tidy(nameOrFn: string|ScopeFn, fn?: ScopeFn):\n T {\n let name: string = null;\n if (fn == null) {\n // Called with only 1 argument.\n if (typeof nameOrFn !== 'function') {\n throw new Error('Please provide a function to tidy()');\n }\n fn = nameOrFn;\n } else {\n // Called with 2 arguments.\n if (typeof nameOrFn !== 'string' && !(nameOrFn instanceof String)) {\n throw new Error(\n 'When calling with two arguments, the first argument ' +\n 'to tidy() must be a string');\n }\n if (typeof fn !== 'function') {\n throw new Error(\n 'When calling with two arguments, the 2nd argument ' +\n 'to tidy() must be a function');\n }\n name = nameOrFn as string;\n // TODO(nsthorat,smilkov): Do operation logging and performance\n // profiling.\n }\n let result: T;\n return this.scopedRun(\n () => this.startScope(name), () => this.endScope(result), () => {\n result = fn();\n if (result instanceof Promise) {\n console.error('Cannot return a Promise inside of tidy.');\n }\n return result;\n });\n }\n\n private scopedRun(start: () => void, end: () => void, f: () => T): T {\n start();\n try {\n const res = f();\n end();\n return res;\n } catch (ex) {\n end();\n throw ex;\n }\n }\n\n private static nextTensorId = 0;\n private nextTensorId(): number {\n return Engine.nextTensorId++;\n }\n\n private static nextVariableId = 0;\n private nextVariableId(): number {\n return Engine.nextVariableId++;\n }\n\n /**\n * This method is called instead of the public-facing tensor.clone() when\n * saving a tensor for backwards pass. It makes sure to add the clone\n * operation to the tape regardless of being called inside a kernel\n * execution.\n *\n * This method will go away once all kernels are modularized since we won't\n * need to turn off the tape inside runKernel().\n */\n private clone(x: Tensor): Tensor {\n const y = this.makeTensorFromDataId(x.dataId, x.shape, x.dtype);\n const inputs = {x};\n const grad = (dy: Tensor) => ({\n x: () => {\n const dtype = 'float32';\n const gradInputs = {x: dy};\n const attrs = {dtype};\n\n return ENGINE.runKernelFunc(\n backend => backend.cast(dy, dtype),\n gradInputs as {} as NamedTensorMap, null /* grad */, Cast,\n attrs as {} as NamedAttrMap);\n }\n });\n const saved: Tensor[] = [];\n this.addTapeNode(this.state.activeScope.name, inputs, [y], grad, saved, {});\n return y;\n }\n\n /**\n * Execute a kernel with the given name and return the output tensor.\n *\n * @param kernelName The name of the kernel to execute.\n * @param inputs A map of input names to tensors.\n * @param attrs A map of attribute names to their values. An attribute is a\n * primitive (non-tensor) input to the kernel.\n * @param inputsToSave A list of tensors, inputs to save for the backprop\n * computation.\n * @param outputsToSave A list of booleans, specifying which output to save\n * for the backprop computation. These are booleans since the output\n * tensors are not visible to the user.\n */\n runKernel(\n kernelName: string, inputs: NamedTensorMap, attrs: NamedAttrMap,\n inputsToSave?: Tensor[], outputsToSave?: boolean[]): Tensor|Tensor[] {\n const forwardFunc: null = null;\n const backwardsFunc: null = null;\n // Call runKernel as a stop-gap until we modularize all kernels.\n // Once we modularize all kernels, we will remove the existing\n // `runKernelFunc`.\n return this.runKernelFunc(\n forwardFunc, inputs, backwardsFunc, kernelName, attrs, inputsToSave,\n outputsToSave);\n }\n\n private shouldCheckForMemLeaks(): boolean {\n return this.ENV.getBool('IS_TEST');\n }\n\n private checkKernelForMemLeak(\n kernelName: string, numDataIdsBefore: number,\n outInfos: TensorInfo[]): void {\n const numDataIdsAfter = this.backend.numDataIds();\n\n // Count the number of data ids associated with the result of the kernel.\n let numOutputDataIds = 0;\n outInfos.forEach(info => {\n // Complex numbers allocate 3 data ids, one for 'real', one for\n // 'imaginary', and one for the container that holds the former two.\n numOutputDataIds += (info.dtype === 'complex64' ? 3 : 1);\n });\n\n // Account for the number of moves during kernel execution. A \"data move\"\n // can happen in the middle of a kernel execution, placing a new (key,value)\n // pair in the data storage. Since data moves have net zero effect (we\n // always remove the data from the old backend), we have to cancel them out\n // when detecting memory leaks.\n const numMoves =\n this.state.numDataMovesStack[this.state.numDataMovesStack.length - 1];\n const dataIdsLeaked =\n numDataIdsAfter - numDataIdsBefore - numOutputDataIds - numMoves;\n if (dataIdsLeaked > 0) {\n throw new Error(\n `Backend '${this.backendName}' has an internal memory leak ` +\n `(${dataIdsLeaked} data ids) after running '${kernelName}'`);\n }\n }\n\n /**\n * @deprecated Use `runKernel` for newly added kernels. Keep using this method\n * only for kernels that are not yet fully modularized.\n */\n runKernelFunc(\n forwardFunc: ForwardFunc, inputs: I,\n backwardsFunc?: (dy: T, saved: Tensor[]) => {[P in keyof I]: () => I[P]},\n kernelName?: string, attrs?: NamedAttrMap, inputsToSave?: Tensor[],\n outputsToSave?: boolean[]): T {\n let outputs: Tensor[];\n let saved: Tensor[] = [];\n const isTapeOn = this.isTapeOn();\n if (kernelName == null) {\n kernelName =\n this.state.activeScope != null ? this.state.activeScope.name : '';\n }\n\n const startingBytecount = this.state.numBytes;\n const startingNumTensors = this.state.numTensors;\n\n if (this.shouldCheckForMemLeaks()) {\n this.state.numDataMovesStack.push(0);\n }\n\n let kernelFunc: () => Tensor[];\n const kernel = getKernel(kernelName, this.backendName);\n let out: TensorInfo|TensorInfo[];\n if (kernel != null) {\n kernelFunc = () => {\n const numDataIdsBefore = this.backend.numDataIds();\n out = kernel.kernelFunc({inputs, attrs, backend: this.backend});\n const outInfos = Array.isArray(out) ? out : [out];\n if (this.shouldCheckForMemLeaks()) {\n this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outInfos);\n }\n const outTensors = outInfos.map(\n ({dataId, shape, dtype}) =>\n this.makeTensorFromDataId(dataId, shape, dtype));\n\n // Save the inputs and outputs.\n // Do not save unless we are recording to the tape. Otherwise it would\n // cause a mem leak since we would never run backprop, which disposes\n // the kept tensors.\n if (isTapeOn) {\n let tensorsToSave =\n this.getTensorsForGradient(kernelName, inputs, outTensors);\n if (tensorsToSave == null) {\n // Fallback for ops that call runKernelFunc and pass in\n // inputsToSave and outputsToSave. Currently this is the set of ops\n // with kernel support in the WASM backend. Once those ops and\n // respective gradients are modularised we can remove this path.\n if (outputsToSave == null) {\n outputsToSave = [];\n }\n const outsToSave = outTensors.filter((_, i) => outputsToSave[i]);\n tensorsToSave = (inputsToSave || []).slice().concat(outsToSave);\n }\n saved = this.saveTensorsForBackwardMode(tensorsToSave);\n }\n return outTensors;\n };\n } else {\n const saveFunc: GradSaveFunc = (tensors) => {\n // Do not save unless we are recording to the tape. Otherwise it would\n // cause a mem leak since we would never run backprop, which disposes\n // the kept tensors.\n if (!isTapeOn) {\n return;\n }\n saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n };\n\n kernelFunc = () => {\n const numDataIdsBefore = this.backend.numDataIds();\n out = this.tidy(() => forwardFunc(this.backend, saveFunc));\n const outs = (Array.isArray(out) ? out : [out]) as Tensor[];\n if (this.shouldCheckForMemLeaks()) {\n this.checkKernelForMemLeak(kernelName, numDataIdsBefore, outs);\n }\n return outs;\n };\n }\n\n // Stop recording to a tape when running a kernel.\n let kernelProfile: KernelProfile;\n this.scopedRun(\n () => this.state.kernelDepth++, () => this.state.kernelDepth--, () => {\n if (!this.ENV.getBool('DEBUG') && !this.state.profiling) {\n outputs = kernelFunc();\n } else {\n kernelProfile = this.profiler.profileKernel(\n kernelName, inputs, () => kernelFunc());\n if (this.ENV.getBool('DEBUG')) {\n this.profiler.logKernelProfile(kernelProfile);\n }\n outputs = kernelProfile.outputs;\n }\n });\n\n if (isTapeOn) {\n this.addTapeNode(\n kernelName, inputs, outputs, backwardsFunc, saved, attrs);\n }\n\n if (this.state.profiling) {\n this.state.activeProfile.kernels.push({\n name: kernelName,\n bytesAdded: this.state.numBytes - startingBytecount,\n totalBytesSnapshot: this.state.numBytes,\n tensorsAdded: this.state.numTensors - startingNumTensors,\n totalTensorsSnapshot: this.state.numTensors,\n inputShapes: Object.keys(inputs).map(\n key => inputs[key] != null ? inputs[key].shape : null),\n outputShapes: outputs.map(item => item.shape),\n kernelTimeMs: kernelProfile.timeMs,\n extraInfo: kernelProfile.extraInfo\n });\n }\n return (Array.isArray(out) ? outputs : outputs[0]) as T;\n }\n\n /**\n * Saves tensors used in forward mode for use in backward mode.\n *\n * @param tensors the list of tensors to save.\n */\n private saveTensorsForBackwardMode(tensors: Tensor[]): Tensor[] {\n const saved = tensors.map(tensor => this.keep(this.clone(tensor)));\n return saved;\n }\n\n /**\n * Returns a list of tensors to save for a given gradient calculation.\n *\n * Returns undefined if their is no registered gradient for this kernel in the\n * gradient registry.\n *\n * @param kernelName name of kernel to look up gradient for.\n * @param inputs a map of input tensors.\n * @param outputs an array of output tensors from forward mode of kernel.\n */\n private getTensorsForGradient(\n kernelName: string, inputs: NamedTensorMap,\n outputs: Tensor[]): Tensor[]|null {\n const gradConfig = getGradient(kernelName);\n if (gradConfig != null) {\n const inputsToSave: string[] = gradConfig.inputsToSave || [];\n const outputsToSave: boolean[] = gradConfig.outputsToSave || [];\n\n // If saveAllInputs is true, all inputs will be saved. Otherwise, inputs\n // specified in inputsToSave will be saved.\n let inputTensorsToSave: Tensor[];\n if (gradConfig.saveAllInputs) {\n util.assert(\n Array.isArray(inputs),\n () => 'saveAllInputs is true, expected inputs to be an array.');\n\n inputTensorsToSave = Object.keys(inputs).map((key) => inputs[key]);\n } else {\n inputTensorsToSave = inputsToSave.map((inputName) => inputs[inputName]);\n }\n\n const outputTensorsToSave: Tensor[] =\n outputs.filter((_, i) => outputsToSave[i]);\n\n return inputTensorsToSave.concat(outputTensorsToSave);\n }\n // TODO(yassogba) throw exception here once all runkernelFunc calls with\n // inputsToSave/outputsToSave are removed\n return null;\n }\n\n /**\n * Internal method used by public APIs for tensor creation. Makes a new\n * tensor with the provided shape, dtype and values. It always\n * creates a new data id and writes the values to the underlying backend.\n */\n makeTensor(\n values: DataValues, shape: number[], dtype: DataType,\n backend?: KernelBackend): Tensor {\n if (values == null) {\n throw new Error('Values passed to engine.makeTensor() are null');\n }\n dtype = dtype || 'float32';\n backend = backend || this.backend;\n let backendVals = values as BackendValues;\n if (dtype === 'string' && util.isString(values[0])) {\n backendVals = (values as string[]).map(d => util.encodeString(d));\n }\n const dataId = backend.write(backendVals, shape, dtype);\n const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n this.incRef(t, backend);\n\n // Count bytes for string tensors.\n if (dtype === 'string') {\n const info = this.state.tensorInfo.get(dataId);\n const newBytes = bytesFromStringArray(backendVals as Uint8Array[]);\n this.state.numBytes += newBytes - info.bytes;\n info.bytes = newBytes;\n }\n return t;\n }\n\n /**\n * Internal method used by backends. Makes a new tensor\n * that is a wrapper around an existing data id. It doesn't create\n * a new data id, only increments the ref count used in memory tracking.\n */\n makeTensorFromDataId(\n dataId: DataId, shape: number[], dtype: DataType,\n backend?: KernelBackend): Tensor {\n dtype = dtype || 'float32';\n const t = new Tensor(shape, dtype, dataId, this.nextTensorId());\n this.incRef(t, backend);\n return t;\n }\n\n makeVariable(\n initialValue: Tensor, trainable = true, name?: string,\n dtype?: DataType): Variable {\n name = name || this.nextVariableId().toString();\n if (dtype != null && dtype !== initialValue.dtype) {\n initialValue = initialValue.cast(dtype);\n }\n const v = new Variable(initialValue, trainable, name, this.nextTensorId());\n if (this.state.registeredVariables[v.name] != null) {\n throw new Error(`Variable with name ${v.name} was already registered`);\n }\n this.state.registeredVariables[v.name] = v;\n this.incRef(v, this.backend);\n return v;\n }\n\n incRef(a: Tensor, backend: KernelBackend): void {\n const refCount = this.state.tensorInfo.has(a.dataId) ?\n this.state.tensorInfo.get(a.dataId).refCount :\n 0;\n this.state.numTensors++;\n if (a.dtype === 'string') {\n this.state.numStringTensors++;\n }\n if (refCount === 0) {\n this.state.numDataBuffers++;\n\n // Bytes for complex numbers are counted by their components. Bytes for\n // string tensors are counted when writing values.\n let bytes = 0;\n if (a.dtype !== 'complex64' && a.dtype !== 'string') {\n bytes = a.size * util.bytesPerElement(a.dtype);\n }\n this.state.tensorInfo.set(a.dataId, {\n backend: backend || this.backend,\n dtype: a.dtype,\n shape: a.shape,\n bytes,\n refCount: 0\n });\n this.state.numBytes += bytes;\n }\n\n this.state.tensorInfo.get(a.dataId).refCount++;\n\n if (!(a instanceof Variable)) {\n this.track(a);\n }\n }\n\n disposeTensor(a: Tensor): void {\n if (!this.state.tensorInfo.has(a.dataId)) {\n return;\n }\n\n this.state.numTensors--;\n if (a.dtype === 'string') {\n this.state.numStringTensors--;\n }\n const info = this.state.tensorInfo.get(a.dataId);\n const refCount = info.refCount;\n\n if (refCount <= 1) {\n // Don't count bytes for complex numbers as they are counted by their\n // components.\n if (a.dtype !== 'complex64') {\n this.state.numBytes -= info.bytes;\n }\n this.state.numDataBuffers--;\n\n info.backend.disposeData(a.dataId);\n this.state.tensorInfo.delete(a.dataId);\n } else {\n this.state.tensorInfo.get(a.dataId).refCount--;\n }\n // TODO(nsthorat): Construct an error and save the stack trace for\n // debugging when in debug mode. Creating a stack trace is too expensive\n // to do unconditionally.\n }\n\n disposeVariables(): void {\n for (const varName in this.state.registeredVariables) {\n const v = this.state.registeredVariables[varName];\n this.disposeVariable(v);\n }\n }\n\n disposeVariable(v: Variable): void {\n this.disposeTensor(v);\n if (this.state.registeredVariables[v.name] != null) {\n delete this.state.registeredVariables[v.name];\n }\n }\n\n memory(): MemoryInfo {\n const info = this.backend.memory() as MemoryInfo;\n info.numTensors = this.state.numTensors;\n info.numDataBuffers = this.state.numDataBuffers;\n info.numBytes = this.state.numBytes;\n if (this.state.numStringTensors > 0) {\n info.unreliable = true;\n if (info.reasons == null) {\n info.reasons = [];\n }\n info.reasons.push(\n 'Memory usage by string tensors is approximate ' +\n '(2 bytes per character)');\n }\n return info;\n }\n\n async profile(query: () => (TensorContainer | Promise)):\n Promise {\n this.state.profiling = true;\n\n const startBytes = this.state.numBytes;\n const startNumTensors = this.state.numTensors;\n\n this.state.activeProfile.kernels = [];\n this.state.activeProfile.result = await query();\n\n this.state.profiling = false;\n\n this.state.activeProfile.peakBytes = Math.max(\n ...this.state.activeProfile.kernels.map(d => d.totalBytesSnapshot));\n this.state.activeProfile.newBytes = this.state.numBytes - startBytes;\n this.state.activeProfile.newTensors =\n this.state.numTensors - startNumTensors;\n for (const kernel of this.state.activeProfile.kernels) {\n kernel.kernelTimeMs = await kernel.kernelTimeMs;\n kernel.extraInfo = await kernel.extraInfo;\n }\n return this.state.activeProfile;\n }\n\n isTapeOn(): boolean {\n return this.state.gradientDepth > 0 && this.state.kernelDepth === 0;\n }\n\n private addTapeNode(\n kernelName: string, inputs: NamedTensorMap, outputs: Tensor[],\n gradientsFunc: GradFunc, saved: Tensor[], attrs: NamedAttrMap): void {\n const tapeNode: TapeNode =\n {id: this.state.nextTapeNodeId++, kernelName, inputs, outputs, saved};\n\n const gradConfig = getGradient(kernelName);\n if (gradConfig != null) {\n gradientsFunc = gradConfig.gradFunc;\n }\n if (gradientsFunc != null) {\n tapeNode.gradient = (dys: Tensor[]) => {\n // TODO(smilkov): To optimize back-prop, pass dys that are not used in\n // the backprop graph to the user as null instead of zeros\n dys = dys.map((dy, i) => {\n if (dy == null) {\n const output = outputs[i];\n const vals = util.makeZerosTypedArray(output.size, output.dtype);\n return this.makeTensor(vals, output.shape, output.dtype);\n }\n return dy;\n });\n // Grad functions of ops with single outputs expect a dy, while ops\n // with multiple outputs expect dys (array of dy).\n return gradientsFunc(dys.length > 1 ? dys : dys[0], saved, attrs);\n };\n }\n this.state.activeTape.push(tapeNode);\n }\n\n keep(result: T): T {\n result.kept = true;\n return result;\n }\n\n private startTape() {\n if (this.state.gradientDepth === 0) {\n this.state.activeTape = [];\n }\n this.state.gradientDepth++;\n }\n\n private endTape() {\n this.state.gradientDepth--;\n }\n\n /**\n * Start a scope. Use this with endScope() to achieve the same functionality\n * as scope() without the need for a function closure.\n */\n startScope(name?: string) {\n const scopeInfo: ScopeState = {\n track: [],\n name: 'unnamed scope',\n id: this.state.nextScopeId++\n };\n if (name) {\n scopeInfo.name = name;\n }\n this.state.scopeStack.push(scopeInfo);\n this.state.activeScope = scopeInfo;\n }\n\n /**\n * End a scope. Use this with startScope() to achieve the same functionality\n * as scope() without the need for a function closure.\n */\n endScope(result?: TensorContainer) {\n const tensorsToTrackInParent = getTensorsInContainer(result);\n const tensorsToTrackInParentSet =\n new Set(tensorsToTrackInParent.map(t => t.id));\n\n // Dispose the arrays tracked in this scope.\n for (let i = 0; i < this.state.activeScope.track.length; i++) {\n const tensor = this.state.activeScope.track[i];\n if (!tensor.kept && !tensorsToTrackInParentSet.has(tensor.id)) {\n tensor.dispose();\n }\n }\n\n const oldScope = this.state.scopeStack.pop();\n this.state.activeScope = this.state.scopeStack.length === 0 ?\n null :\n this.state.scopeStack[this.state.scopeStack.length - 1];\n\n // Track the current result in the parent scope.\n tensorsToTrackInParent.forEach(tensor => {\n // Only track the tensor if was allocated in the inner scope and is not\n // globally kept.\n if (!tensor.kept && tensor.scopeId === oldScope.id) {\n this.track(tensor);\n }\n });\n }\n\n /**\n * Returns gradients of `f` with respect to each of the `xs`. The gradients\n * returned are of the same length as `xs`, but some might be null if `f`\n * was not a function of that `x`. It also takes optional dy to multiply the\n * gradient, which defaults to `1`.\n */\n gradients(\n f: () => T, xs: Tensor[], dy?: T,\n allowNoGradients = false): {value: T, grads: Tensor[]} {\n util.assert(\n xs.length > 0, () => 'gradients() received an empty list of xs.');\n if (dy != null && dy.dtype !== 'float32') {\n throw new Error(`dy must have 'float32' dtype, but has '${dy.dtype}'`);\n }\n\n const y = this.scopedRun(\n () => this.startTape(), () => this.endTape(),\n () => this.tidy('forward', f));\n\n util.assert(\n y instanceof Tensor,\n () => 'The result y returned by f() must be a tensor.');\n // Filter out the nodes that don't connect x => y.\n const filteredTape = getFilteredNodesXToY(this.state.activeTape, xs, y);\n if (!allowNoGradients && filteredTape.length === 0 && xs.length > 0) {\n throw new Error(\n 'Cannot compute gradient of y=f(x) with respect to x. Make sure ' +\n 'that the f you passed encloses all operations that lead from x ' +\n 'to y.');\n }\n\n return this.tidy('backward', () => {\n const accumulatedGradientMap: {[tensorId: number]: Tensor} = {};\n accumulatedGradientMap[y.id] = (dy == null) ? ones(y.shape) : dy;\n\n // Backprop gradients through the filtered nodes.\n backpropagateGradients(\n accumulatedGradientMap, filteredTape,\n // Pass the tidy function to avoid circular dep with `tape.ts`.\n f => this.tidy(f as ScopeFn),\n // Pass an add function to avoide a circular dep with `tape.ts`.\n add);\n const grads = xs.map(x => accumulatedGradientMap[x.id]);\n\n if (this.state.gradientDepth === 0) {\n // This means that we are not computing higher-order gradients\n // and can clean up the tape.\n this.state.activeTape.forEach(node => {\n for (const tensor of node.saved) {\n tensor.dispose();\n }\n });\n this.state.activeTape = null;\n }\n return {value: y, grads};\n });\n }\n\n customGrad(f: CustomGradientFunc):\n (...args: Array) => T {\n util.assert(\n util.isFunction(f),\n () => 'The f passed in customGrad(f) must be a function.');\n return (...inputs: Tensor[]): T => {\n util.assert(\n inputs.every(t => t instanceof Tensor),\n () => 'The args passed in customGrad(f)(x1, x2,...) must all be ' +\n 'tensors');\n\n let res: {\n value: T,\n gradFunc: (dy: T, saved: Tensor[]) => Tensor | Tensor[],\n };\n const inputMap: NamedTensorMap = {};\n inputs.forEach((input, i) => {\n inputMap[i] = input;\n });\n return this.runKernelFunc(\n (_, save) => {\n res = f(...[...inputs, save]);\n util.assert(\n res.value instanceof Tensor,\n () => 'The function f passed in customGrad(f) must return an ' +\n 'object where `obj.value` is a tensor');\n util.assert(\n util.isFunction(res.gradFunc),\n () => 'The function f passed in customGrad(f) must return an ' +\n 'object where `obj.gradFunc` is a function.');\n return res.value;\n },\n inputMap,\n (dy: T, saved: Tensor[]) => {\n const gradRes = res.gradFunc(dy, saved);\n const grads: Tensor[] =\n Array.isArray(gradRes) ? gradRes : [gradRes];\n util.assert(\n grads.length === inputs.length,\n () => 'The function f passed in customGrad(f) must return an ' +\n 'object where `obj.gradFunc` is a function that returns ' +\n 'the same number of tensors as inputs passed to f(...).');\n util.assert(\n grads.every(t => t instanceof Tensor),\n () => 'The function f passed in customGrad(f) must return an ' +\n 'object where `obj.gradFunc` is a function that returns ' +\n 'a list of only tensors.');\n const gradMap: {[key: string]: () => Tensor} = {};\n grads.forEach((grad, i) => {\n gradMap[i] = () => grad;\n });\n return gradMap;\n });\n };\n }\n\n readSync(dataId: DataId): BackendValues {\n // Route the read to the correct backend.\n const info = this.state.tensorInfo.get(dataId);\n return info.backend.readSync(dataId);\n }\n read(dataId: DataId): Promise {\n // Route the read to the correct backend.\n const info = this.state.tensorInfo.get(dataId);\n return info.backend.read(dataId);\n }\n\n async time(query: () => void): Promise {\n const start = now();\n const timingInfo = await this.backend.time(query) as TimingInfo;\n timingInfo.wallMs = now() - start;\n return timingInfo;\n }\n\n /**\n * Tracks a Tensor in the current scope to be automatically cleaned up\n * when the current scope ends, and returns the value.\n *\n * @param result The Tensor to track in the current scope.\n */\n private track(result: T): T {\n if (this.state.activeScope != null) {\n result.scopeId = this.state.activeScope.id;\n this.state.activeScope.track.push(result);\n }\n\n return result;\n }\n\n get registeredVariables(): NamedVariableMap {\n return this.state.registeredVariables;\n }\n\n /**\n * Resets the engine state. Removes all backends but does not remove\n * registered backend factories.\n */\n reset(): void {\n // Make any pending promise obsolete.\n this.pendingBackendInitId++;\n\n this.state.dispose();\n this.ENV.reset();\n this.state = new EngineState();\n\n for (const backendName in this.registry) {\n this.disposeRegisteredKernels(backendName);\n this.registry[backendName].dispose();\n delete this.registry[backendName];\n }\n this.backendName = null;\n this.backendInstance = null;\n this.pendingBackendInit = null;\n }\n}\n\nfunction ones(shape: number[]): Tensor {\n const values = makeOnesTypedArray(sizeFromShape(shape), 'float32');\n return ENGINE.makeTensor(values, shape, 'float32');\n}\n\nexport function getOrMakeEngine(): Engine {\n const ns = getGlobalNamespace() as {} as {_tfengine: Engine};\n if (ns._tfengine == null) {\n const environment = new Environment(ns);\n ns._tfengine = new Engine(environment);\n }\n setEnvironmentGlobal(ns._tfengine.ENV);\n\n // Tell the current tensor interface that the global engine is responsible\n // for tracking.\n setTensorTracker(() => ns._tfengine);\n return ns._tfengine;\n}\n\nexport const ENGINE = getOrMakeEngine();\n\n/**\n * A implementation of the add op for use within engine and tape.\n *\n * This allows us to avoid a circular dependency between add.ts and engine.\n * It is exported to be available in tape tests.\n */\nexport function add(a: Tensor, b: Tensor): Tensor {\n // We duplicate Add here to avoid a circular dependency with add.ts.\n const inputs = {a, b};\n return ENGINE.runKernelFunc((backend, save) => {\n const res = backend.add(a, b);\n save([a, b]);\n return res;\n }, inputs as {} as NamedTensorMap, null /* gradient */, Add);\n}\n", "/**\n * @license\n * Copyright 2017 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// tslint:disable-next-line:no-any\nfunction _isNavigatorDefined(): boolean {\n return typeof navigator !== 'undefined' && navigator != null;\n}\n\nexport function isMobile(): boolean {\n if (_isNavigatorDefined()) {\n // tslint:disable-next-line:no-any\n const a = navigator.userAgent || navigator.vendor || (window as any).opera;\n // tslint:disable-next-line:max-line-length\n return /(android|bb\\d+|meego).+mobile|avantgo|bada\\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i\n .test(a) ||\n // tslint:disable-next-line:max-line-length\n /1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\\-(n|u)|c55\\/|capi|ccwa|cdm\\-|cell|chtm|cldc|cmd\\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\\-s|devi|dica|dmob|do(c|p)o|ds(12|\\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\\-|_)|g1 u|g560|gene|gf\\-5|g\\-mo|go(\\.w|od)|gr(ad|un)|haie|hcit|hd\\-(m|p|t)|hei\\-|hi(pt|ta)|hp( i|ip)|hs\\-c|ht(c(\\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\\-(20|go|ma)|i230|iac( |\\-|\\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\\/)|klon|kpt |kwc\\-|kyo(c|k)|le(no|xi)|lg( g|\\/(k|l|u)|50|54|\\-[a-w])|libw|lynx|m1\\-w|m3ga|m50\\/|ma(te|ui|xo)|mc(01|21|ca)|m\\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\\-2|po(ck|rt|se)|prox|psio|pt\\-g|qa\\-a|qc(07|12|21|32|60|\\-[2-7]|i\\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\\-|oo|p\\-)|sdk\\/|se(c(\\-|0|1)|47|mc|nd|ri)|sgh\\-|shar|sie(\\-|m)|sk\\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\\-|v\\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\\-|tdg\\-|tel(i|m)|tim\\-|t\\-mo|to(pl|sh)|ts(70|m\\-|m3|m5)|tx\\-9|up(\\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\\-|your|zeto|zte\\-/i\n .test(a.substr(0, 4));\n }\n return false;\n}\n\nexport function isBrowser(): boolean {\n return (typeof window !== 'undefined' && window.document != null) ||\n //@ts-ignore\n (typeof WorkerGlobalScope !== 'undefined');\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport './engine';\n\nimport * as device_util from './device_util';\nimport {env} from './environment';\n\nconst ENV = env();\n\n/**\n * This file contains environment-related flag registrations.\n */\n\n/** Whether to enable debug mode. */\nENV.registerFlag('DEBUG', () => false, debugValue => {\n if (debugValue) {\n console.warn(\n 'Debugging mode is ON. The output of every math call will ' +\n 'be downloaded to CPU and checked for NaNs. ' +\n 'This significantly impacts performance.');\n }\n});\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag('IS_BROWSER', () => device_util.isBrowser());\n\n/** Whether we are in a browser (as versus, say, node.js) environment. */\nENV.registerFlag(\n 'IS_NODE',\n () => (typeof process !== 'undefined') &&\n (typeof process.versions !== 'undefined') &&\n (typeof process.versions.node !== 'undefined'));\n\n/** Whether this browser is Chrome. */\nENV.registerFlag(\n 'IS_CHROME',\n () => typeof navigator !== 'undefined' && navigator != null &&\n navigator.userAgent != null && /Chrome/.test(navigator.userAgent) &&\n /Google Inc/.test(navigator.vendor));\n\n/**\n * True when the environment is \"production\" where we disable safety checks\n * to gain performance.\n */\nENV.registerFlag('PROD', () => false);\n\n/**\n * Whether to do sanity checks when inferring a shape from user-provided\n * values, used when creating a new tensor.\n */\nENV.registerFlag(\n 'TENSORLIKE_CHECK_SHAPE_CONSISTENCY', () => ENV.getBool('DEBUG'));\n\n/** Whether deprecation warnings are enabled. */\nENV.registerFlag('DEPRECATION_WARNINGS_ENABLED', () => true);\n\n/** True if running unit tests. */\nENV.registerFlag('IS_TEST', () => false);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from './engine';\nimport {env} from './environment';\nimport {Tensor} from './tensor';\nimport {DataType, TensorLike} from './types';\nimport {assert, flatten, inferDtype, isTypedArray, toTypedArray} from './util';\n\nexport function inferShape(val: TensorLike, dtype?: DataType): number[] {\n let firstElem: typeof val = val;\n\n if (isTypedArray(val)) {\n return dtype === 'string' ? [] : [val.length];\n }\n if (!Array.isArray(val)) {\n return []; // Scalar.\n }\n const shape: number[] = [];\n\n while (Array.isArray(firstElem) ||\n isTypedArray(firstElem) && dtype !== 'string') {\n shape.push(firstElem.length);\n firstElem = firstElem[0];\n }\n if (Array.isArray(val) &&\n env().getBool('TENSORLIKE_CHECK_SHAPE_CONSISTENCY')) {\n deepAssertShapeConsistency(val, shape, []);\n }\n\n return shape;\n}\n\nfunction deepAssertShapeConsistency(\n val: TensorLike, shape: number[], indices: number[]) {\n indices = indices || [];\n if (!(Array.isArray(val)) && !isTypedArray(val)) {\n assert(\n shape.length === 0,\n () => `Element arr[${indices.join('][')}] is a primitive, ` +\n `but should be an array/TypedArray of ${shape[0]} elements`);\n return;\n }\n assert(\n shape.length > 0,\n () => `Element arr[${indices.join('][')}] should be a primitive, ` +\n `but is an array of ${val.length} elements`);\n assert(\n val.length === shape[0],\n () => `Element arr[${indices.join('][')}] should have ${shape[0]} ` +\n `elements, but has ${val.length} elements`);\n const subShape = shape.slice(1);\n for (let i = 0; i < val.length; ++i) {\n deepAssertShapeConsistency(val[i], subShape, indices.concat(i));\n }\n}\n\nfunction assertDtype(\n expectedDtype: DataType|'numeric', actualDType: DataType, argName: string,\n functionName: string) {\n if (expectedDtype == null) {\n return;\n }\n if (expectedDtype !== 'numeric' && expectedDtype !== actualDType ||\n expectedDtype === 'numeric' && actualDType === 'string') {\n throw new Error(\n `Argument '${argName}' passed to '${functionName}' must ` +\n `be ${expectedDtype} tensor, but got ${actualDType} tensor`);\n }\n}\n\nexport function convertToTensor(\n x: T|TensorLike, argName: string, functionName: string,\n parseAsDtype: DataType|'numeric' = 'numeric'): T {\n if (x instanceof Tensor) {\n assertDtype(parseAsDtype, x.dtype, argName, functionName);\n return x;\n }\n let inferredDtype = inferDtype(x);\n // If the user expects a bool/int/float, use that info to update the\n // inferredDtype when it is not a string.\n if (inferredDtype !== 'string' &&\n ['bool', 'int32', 'float32'].indexOf(parseAsDtype) >= 0) {\n inferredDtype = parseAsDtype as DataType;\n }\n assertDtype(parseAsDtype, inferredDtype, argName, functionName);\n\n if ((x == null) ||\n (!isTypedArray(x) && !Array.isArray(x) && typeof x !== 'number' &&\n typeof x !== 'boolean' && typeof x !== 'string')) {\n const type = x == null ? 'null' : (x as {}).constructor.name;\n throw new Error(\n `Argument '${argName}' passed to '${functionName}' must be a ` +\n `Tensor or TensorLike, but got '${type}'`);\n }\n const inferredShape = inferShape(x, inferredDtype);\n if (!isTypedArray(x) && !Array.isArray(x)) {\n x = [x] as number[];\n }\n const skipTypedArray = true;\n const values = inferredDtype !== 'string' ?\n toTypedArray(x, inferredDtype as DataType) :\n flatten(x as string[], [], skipTypedArray) as string[];\n return ENGINE.makeTensor(values, inferredShape, inferredDtype) as T;\n}\n\nexport function convertToTensorArray(\n arg: Array, argName: string, functionName: string,\n parseAsDtype: DataType|'numeric' = 'numeric'): T[] {\n if (!Array.isArray(arg)) {\n throw new Error(\n `Argument ${argName} passed to ${functionName} must be a ` +\n '`Tensor[]` or `TensorLike[]`');\n }\n const tensors = arg as T[];\n return tensors.map(\n (t, i) => convertToTensor(t, `${argName}[${i}]`, functionName),\n parseAsDtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\n\nexport const OP_SCOPE_SUFFIX = '__op';\n\n/**\n * Used for wrapping functions that perform math operations on\n * Tensors. The function will be wrapped in a named scope that cleans all\n * memory usage after the function is done.\n */\nexport function op(f: {[name: string]: T}): T {\n const keys = Object.keys(f);\n if (keys.length !== 1) {\n throw new Error(\n `Please provide an object with a single key ` +\n `(operation name) mapping to a function. Got an object with ` +\n `${keys.length} keys.`);\n }\n\n let opName = keys[0];\n const fn = f[opName];\n\n // Strip the underscore from the end of the function name.\n if (opName.endsWith('_')) {\n opName = opName.substring(0, opName.length - 1);\n }\n\n // add an __op suffix to distinguish ops from kernels in tf.profile\n opName = opName + OP_SCOPE_SUFFIX;\n\n // tslint:disable-next-line:no-any\n const f2 = (...args: any[]) => {\n ENGINE.startScope(opName);\n try {\n const result = fn(...args);\n if (result instanceof Promise) {\n console.error('Cannot return a Promise inside of tidy.');\n }\n ENGINE.endScope(result);\n return result;\n } catch (ex) {\n ENGINE.endScope(null);\n throw ex;\n }\n };\n Object.defineProperty(f2, 'name', {value: opName, configurable: true});\n\n // tslint:disable-next-line:no-any\n return f2 as any as T;\n}\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Complex, ComplexInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Converts two real numbers to a complex number.\n *\n * Given a tensor `real` representing the real part of a complex number, and a\n * tensor `imag` representing the imaginary part of a complex number, this\n * operation returns complex numbers elementwise of the form [r0, i0, r1, i1],\n * where r represents the real part and i represents the imag part.\n *\n * The input tensors real and imag must have the same shape.\n *\n * ```js\n * const real = tf.tensor1d([2.25, 3.25]);\n * const imag = tf.tensor1d([4.75, 5.75]);\n * const complex = tf.complex(real, imag);\n *\n * complex.print();\n * ```\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction complex_(real: T|TensorLike, imag: T|TensorLike): T {\n const $real = convertToTensor(real, 'real', 'complex');\n const $imag = convertToTensor(imag, 'imag', 'complex');\n util.assertShapesMatch(\n $real.shape, $imag.shape,\n `real and imag shapes, ${$real.shape} and ${$imag.shape}, ` +\n `must match in call to tf.complex().`);\n\n const forward: ForwardFunc = (backend) => {\n return backend.complex($real, $imag);\n };\n const inputs: ComplexInputs = {real: $real, imag: $imag};\n return ENGINE.runKernelFunc(\n forward, inputs as {} as NamedTensorMap, null /* gradient */,\n Complex) as T;\n}\n\nexport const complex = op({complex_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Tensor} from '../tensor';\nimport {TensorLike, TypedArray} from '../types';\nimport {DataType} from '../types';\nimport {assert, assertNonNegativeIntegerDimensions, flatten, inferDtype, isTypedArray, sizeFromShape, toTypedArray} from '../util';\n\n/** This is shared code across all tensor creation methods. */\nexport function makeTensor(\n values: TensorLike, shape: number[], inferredShape: number[],\n dtype?: DataType): Tensor {\n if (dtype == null) {\n dtype = inferDtype(values);\n }\n if (dtype === 'complex64') {\n throw new Error(\n `Cannot construct a complex64 tensor directly. ` +\n `Please use tf.complex(real, imag).`);\n }\n if (!isTypedArray(values) && !Array.isArray(values) &&\n typeof values !== 'number' && typeof values !== 'boolean' &&\n typeof values !== 'string') {\n throw new Error(\n 'values passed to tensor(values) must be a number/boolean/string or ' +\n 'an array of numbers/booleans/strings, or a TypedArray');\n }\n if (shape != null) {\n assertNonNegativeIntegerDimensions(shape);\n\n const providedSize = sizeFromShape(shape);\n const inferredSize = sizeFromShape(inferredShape);\n assert(\n providedSize === inferredSize,\n () =>\n `Based on the provided shape, [${shape}], the tensor should have ` +\n `${providedSize} values but has ${inferredSize}`);\n\n for (let i = 0; i < inferredShape.length; ++i) {\n const inferred = inferredShape[i];\n const flatDimsDontMatch = i === inferredShape.length - 1 ?\n inferred !== sizeFromShape(shape.slice(i)) :\n true;\n assert(\n inferredShape[i] === shape[i] || !flatDimsDontMatch,\n () => `Error creating a new Tensor. Inferred shape ` +\n `(${inferredShape}) does not match the provided ` +\n `shape (${shape}). `);\n }\n }\n\n if (!isTypedArray(values) && !Array.isArray(values)) {\n values = [values] as number[];\n }\n\n shape = shape || inferredShape;\n values = dtype !== 'string' ?\n toTypedArray(values, dtype) :\n flatten(values as string[], [], true) as string[];\n return ENGINE.makeTensor(values as TypedArray, shape, dtype);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport {DataType, Rank, ShapeMap} from '../types';\n\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates a `tf.Tensor` with the provided values, shape and dtype.\n *\n * ```js\n * // Pass an array of values to create a vector.\n * tf.tensor([1, 2, 3, 4]).print();\n * ```\n *\n * ```js\n * // Pass a nested array of values to make a matrix or a higher\n * // dimensional tensor.\n * tf.tensor([[1, 2], [3, 4]]).print();\n * ```\n *\n * ```js\n * // Pass a flat array and specify a shape yourself.\n * tf.tensor([1, 2, 3, 4], [2, 2]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n * or a flat array, or a `TypedArray`. If the values are strings,\n * they will be encoded as utf-8 and kept as `Uint8Array[]`.\n * @param shape The shape of the tensor. Optional. If not provided,\n * it is inferred from `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor(\n values: TensorLike, shape?: ShapeMap[R], dtype?: DataType): Tensor {\n const inferredShape = inferShape(values, dtype);\n return makeTensor(values, shape, inferredShape, dtype) as Tensor;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/* Type definitions for exporting and importing of models. */\n\n/**\n * A map from Tensor dtype to number of bytes per element of the Tensor.\n */\nexport const DTYPE_VALUE_SIZE_MAP: {[dtype: string]: number} = {\n 'float32': 4,\n 'float16': 2,\n 'int32': 4,\n 'uint16': 2,\n 'uint8': 1,\n 'bool': 1,\n 'complex64': 8\n};\n\n/**\n * A weight manifest.\n *\n * The weight manifest consists of an ordered list of weight-manifest groups.\n * Each weight-manifest group (\"group\" for short hereafter) consists of a\n * number of weight values stored in a number of paths.\n * See the documentation of `WeightManifestGroupConfig` below for more details.\n */\nexport declare type WeightsManifestConfig = WeightsManifestGroupConfig[];\n\n/**\n * A weight-manifest group.\n *\n * Consists of an ordered list of weight values encoded in binary format,\n * stored in an ordered list of paths.\n */\nexport declare interface WeightsManifestGroupConfig {\n /**\n * An ordered list of paths.\n *\n * Paths are intentionally abstract in order to be general. For example, they\n * can be relative URL paths or relative paths on the file system.\n */\n paths: string[];\n\n /**\n * Specifications of the weights stored in the paths.\n */\n weights: WeightsManifestEntry[];\n}\n\n/**\n * Group to which the weight belongs.\n *\n * - 'optimizer': Weight from a stateful optimizer.\n */\nexport type WeightGroup = 'model'|'optimizer';\n\n/**\n * An entry in the weight manifest.\n *\n * The entry contains specification of a weight.\n */\nexport declare interface WeightsManifestEntry {\n /**\n * Name of the weight, e.g., 'Dense_1/bias'\n */\n name: string;\n\n /**\n * Shape of the weight.\n */\n shape: number[];\n\n /**\n * Data type of the weight.\n */\n dtype: 'float32'|'int32'|'bool'|'string'|'complex64';\n\n /**\n * Type of the weight.\n *\n * Optional.\n *\n * The value 'optimizer' indicates the weight belongs to an optimizer\n * (i.e., used only during model training and not during inference).\n */\n group?: WeightGroup;\n\n /**\n * Information for dequantization of the weight.\n */\n quantization?: {\n scale?: number, // The scaling constant to multiply by.\n min?: number, // The (possibly nudged) minimum weight to add.\n dtype: 'uint16'|'uint8'|'float16' // The dtype of the quantized weights.\n };\n}\n\n/**\n * Options for saving a model.\n * @innamespace io\n */\nexport interface SaveConfig {\n /**\n * Whether to save only the trainable weights of the model, ignoring the\n * non-trainable ones.\n */\n trainableOnly?: boolean;\n\n /**\n * Whether the optimizer will be saved (if exists).\n *\n * Default: `false`.\n */\n includeOptimizer?: boolean;\n}\n\n/**\n * Result of a saving operation.\n */\nexport interface SaveResult {\n /**\n * Information about the model artifacts saved.\n */\n modelArtifactsInfo: ModelArtifactsInfo;\n\n /**\n * HTTP responses from the server that handled the model-saving request (if\n * any). This is applicable only to server-based saving routes.\n */\n responses?: Response[];\n\n /**\n * Error messages and related data (if any).\n */\n errors?: Array<{}|string>;\n}\n\nexport declare interface ModelArtifactsInfo {\n /**\n * Timestamp for when the model is saved.\n */\n dateSaved: Date;\n\n /**\n * TODO (cais,yassogba) consider removing GraphDef as GraphDefs now\n * come in a JSON format and none of our IOHandlers support a non json\n * format. We could conder replacing this with 'Binary' if we want to\n * allow future handlers to save to non json formats (though they will\n * probably want more information than 'Binary').\n * Type of the model topology\n *\n * Type of the model topology\n *\n * Possible values:\n * - JSON: JSON config (human-readable, e.g., Keras JSON).\n * - GraphDef: TensorFlow\n * [GraphDef](https://www.tensorflow.org/extend/tool_developers/#graphdef)\n * protocol buffer (binary).\n */\n modelTopologyType: 'JSON'|'GraphDef';\n\n /**\n * Size of model topology (Keras JSON or GraphDef), in bytes.\n */\n modelTopologyBytes?: number;\n\n /**\n * Size of weight specification or manifest, in bytes.\n */\n weightSpecsBytes?: number;\n\n /**\n * Size of weight value data, in bytes.\n */\n weightDataBytes?: number;\n}\n\n/** Model training configuration. */\nexport declare interface TrainingConfig {\n // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n // See\n // tslint:disable-next-line:max-line-length\n // https://github.com/tensorflow/tfjs-layers/blob/master/src/keras_format/training_config.ts\n /** Optimizer used for the model training. */\n optimizer_config: {};\n\n // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n /** Loss function(s) for the model's output(s). */\n loss: string|string[]|{[key: string]: string};\n\n // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n /** Metric function(s) for the model's output(s). */\n metrics?: string[]|{[key: string]: string};\n\n // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n weighted_metrics?: string[];\n\n // TODO(cais): Tighten the typing once keras spec is available to tfjs-core.\n sample_weight_mode?: string;\n\n loss_weights?: number[]|{[key: string]: number};\n}\n\n/**\n * The serialized artifacts of a model, including topology and weights.\n *\n * The `modelTopology`, `trainingConfig`, `weightSpecs` and `weightData` fields\n * of this interface are optional, in order to support topology- or weights-only\n * saving and loading.\n *\n * Note this interface is used internally in IOHandlers. For the file format\n * written to disk as `model.json`, see `ModelJSON`.\n */\nexport declare interface ModelArtifacts {\n /**\n * Model topology.\n *\n * For Keras-style `tf.Model`s, this is a JSON object.\n * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n * encoding of the `GraphDef` protocol buffer.\n */\n modelTopology?: {}|ArrayBuffer;\n\n /**\n * Serialized configuration for the model's training.\n */\n trainingConfig?: TrainingConfig;\n\n /**\n * Weight specifications.\n *\n * This corresponds to the weightsData below.\n */\n weightSpecs?: WeightsManifestEntry[];\n\n /**\n * Binary buffer for all weight values concatenated in the order specified\n * by `weightSpecs`.\n */\n weightData?: ArrayBuffer;\n\n /**\n * Hard-coded format name for models saved from TensorFlow.js or converted\n * by TensorFlow.js Converter.\n */\n format?: string;\n\n /**\n * What library is responsible for originally generating this artifact.\n *\n * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n */\n generatedBy?: string;\n\n /**\n * What library or tool is responsible for converting the original model\n * to this format, applicable only if the model is output by a converter.\n *\n * Used for debugging purposes. E.g., 'TensorFlow.js Converter v1.0.0'.\n *\n * A value of `null` means the model artifacts are generated without any\n * conversion process (e.g., saved directly from a TensorFlow.js\n * `tf.LayersModel` instance.)\n */\n convertedBy?: string|null;\n\n /**\n * User-defined metadata about the model.\n */\n userDefinedMetadata?: {};\n\n /**\n * Initializer for the model.\n */\n modelInitializer?: {};\n}\n\n/**\n * The on-disk format of the `model.json` file.\n *\n * TF.js 1.0 always populates the optional fields when writing model.json.\n * Prior versions did not provide those fields.\n */\nexport declare interface ModelJSON {\n /**\n * Model topology.\n *\n * For Keras-style `tf.Model`s, this is a JSON object.\n * For TensorFlow-style models (e.g., `SavedModel`), this is the JSON\n * encoding of the `GraphDef` protocol buffer.\n */\n modelTopology: {};\n\n /** Model training configuration. */\n trainingConfig?: TrainingConfig;\n\n /**\n * Weights manifest.\n *\n * The weights manifest consists of an ordered list of weight-manifest\n * groups. Each weight-manifest group consists of a number of weight values\n * stored in a number of paths. See the documentation of\n * `WeightsManifestConfig` for more details.\n */\n weightsManifest: WeightsManifestConfig;\n\n /**\n * Hard-coded format name for models saved from TensorFlow.js or converted\n * by TensorFlow.js Converter.\n */\n format?: string;\n\n /**\n * What library is responsible for originally generating this artifact.\n *\n * Used for debugging purposes. E.g., 'TensorFlow.js v1.0.0'.\n */\n generatedBy?: string;\n\n /**\n * What library or tool is responsible for converting the original model\n * to this format, applicable only if the model is output by a converter.\n *\n * Used for debugging purposes. E.g., 'TensorFlow.js Converter v1.0.0'.\n *\n * A value of `null` means the model artifacts are generated without any\n * conversion process (e.g., saved directly from a TensorFlow.js\n * `tf.LayersModel` instance.)\n */\n convertedBy?: string|null;\n\n /**\n * User-defined metadata about the model.\n */\n userDefinedMetadata?: {};\n\n /**\n * Initializer for the model.\n */\n modelInitializer?: {};\n}\n\n/**\n * Type definition for handlers of loading operations.\n */\nexport type LoadHandler = () => Promise;\n\n/**\n * Type definition for handlers of saving operations.\n */\nexport type SaveHandler = (modelArtifact: ModelArtifacts) =>\n Promise;\n\n/**\n * Interface for a model import/export handler.\n *\n * The `save` and `load` handlers are both optional, in order to allow handlers\n * that support only saving or loading.\n */\n// tslint:disable-next-line:interface-name\nexport interface IOHandler {\n save?: SaveHandler;\n load?: LoadHandler;\n}\n\n/**\n * An interface for the manager of a model store.\n *\n * A model store is defined as a storage medium on which multiple models can\n * be stored. Each stored model has a unique `path` as its identifier.\n * A `ModelStoreManager` for the store allows actions including\n *\n * - Listing the models stored in the store.\n * - Deleting a model from the store.\n */\nexport interface ModelStoreManager {\n /**\n * List all models in the model store.\n *\n * @returns A dictionary mapping paths of existing models to their\n * model artifacts info. Model artifacts info include type of the model's\n * topology, byte sizes of the topology, weights, etc.\n */\n listModels(): Promise<{[path: string]: ModelArtifactsInfo}>;\n\n /**\n * Remove a model specified by `path`.\n *\n * @param path\n * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n * is successful).\n * @throws Error if deletion fails, e.g., if no model exists at `path`.\n */\n removeModel(path: string): Promise;\n}\n\n/**\n * Callback for the progress of a long-running action such as an HTTP\n * request for a large binary object.\n *\n * `fraction` should be a number in the [0, 1] interval, indicating how\n * much of the action has completed.\n */\nexport type OnProgressCallback = (fraction: number) => void;\n\n/** @innamespace io */\nexport interface LoadOptions {\n /**\n * RequestInit (options) for HTTP requests.\n *\n * For detailed information on the supported fields, see\n * [https://developer.mozilla.org/en-US/docs/Web/API/Request/Request](\n * https://developer.mozilla.org/en-US/docs/Web/API/Request/Request)\n */\n requestInit?: RequestInit;\n\n /**\n * Progress callback.\n */\n onProgress?: OnProgressCallback;\n\n /**\n * A function used to override the `window.fetch` function.\n */\n fetchFunc?: Function;\n\n /**\n * Strict loading model: whether extraneous weights or missing\n * weights should trigger an `Error`.\n *\n * If `true`, require that the provided weights exactly match those\n * required by the layers. `false` means that both extra weights\n * and missing weights will be silently ignored.\n *\n * Default: `true`.\n */\n strict?: boolean;\n\n /**\n * Path prefix for weight files, by default this is calculated from the\n * path of the model JSON file.\n *\n * For instance, if the path to the model JSON file is\n * `http://localhost/foo/model.json`, then the default path prefix will be\n * `http://localhost/foo/`. If a weight file has the path value\n * `group1-shard1of2` in the weight manifest, then the weight file will be\n * loaded from `http://localhost/foo/group1-shard1of2` by default. However,\n * if you provide a `weightPathPrefix` value of\n * `http://localhost/foo/alt-weights`, then the weight file will be loaded\n * from the path `http://localhost/foo/alt-weights/group1-shard1of2` instead.\n */\n weightPathPrefix?: string;\n\n /**\n * Whether the module or model is to be loaded from TF Hub.\n *\n * Setting this to `true` allows passing a TF-Hub module URL, omitting the\n * standard model file name and the query parameters.\n *\n * Default: `false`.\n */\n fromTFHub?: boolean;\n\n /**\n * An async function to convert weight file name to URL. The weight file\n * names are stored in model.json's weightsManifest.paths field. By default we\n * consider weight files are colocated with the model.json file. For example:\n * model.json URL: https://www.google.com/models/1/model.json\n * group1-shard1of1.bin url:\n * https://www.google.com/models/1/group1-shard1of1.bin\n *\n * With this func you can convert the weight file name to any URL.\n */\n weightUrlConverter?: (weightFileName: string) => Promise;\n}\n\n/**\n * Additional options for Platform.fetch\n */\nexport interface RequestDetails {\n /**\n * Is this request for a binary file (as opposed to a json file)\n */\n isBinary?: boolean;\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {complex} from '../ops/complex';\n\nimport {tensor} from '../ops/tensor';\nimport {NamedTensor, NamedTensorMap} from '../tensor_types';\nimport {TypedArray} from '../types';\nimport {sizeFromShape} from '../util';\n\nimport {DTYPE_VALUE_SIZE_MAP, ModelArtifacts, ModelArtifactsInfo, WeightGroup, WeightsManifestEntry} from './types';\n\n/** Number of bytes reserved for the length of the string. (32bit integer). */\nconst NUM_BYTES_STRING_LENGTH = 4;\n\n/**\n * Encode a map from names to weight values as an ArrayBuffer, along with an\n * `Array` of `WeightsManifestEntry` as specification of the encoded weights.\n *\n * This function does not perform sharding.\n *\n * This function is the reverse of `decodeWeights`.\n *\n * @param tensors A map (\"dict\") from names to tensors.\n * @param group Group to which the weights belong (optional).\n * @returns A `Promise` of\n * - A flat `ArrayBuffer` with all the binary values of the `Tensor`s\n * concatenated.\n * - An `Array` of `WeightManifestEntry`s, carrying information including\n * tensor names, `dtype`s and shapes.\n * @throws Error: on unsupported tensor `dtype`.\n */\nexport async function encodeWeights(\n tensors: NamedTensorMap|NamedTensor[], group?: WeightGroup):\n Promise<{data: ArrayBuffer, specs: WeightsManifestEntry[]}> {\n // TODO(adarob, cais): Support quantization.\n const specs: WeightsManifestEntry[] = [];\n const dataPromises: Array> = [];\n\n const names: string[] = Array.isArray(tensors) ?\n tensors.map(tensor => tensor.name) :\n Object.keys(tensors);\n\n for (let i = 0; i < names.length; ++i) {\n const name = names[i];\n const t = Array.isArray(tensors) ? tensors[i].tensor : tensors[name];\n if (t.dtype !== 'float32' && t.dtype !== 'int32' && t.dtype !== 'bool' &&\n t.dtype !== 'string' && t.dtype !== 'complex64') {\n throw new Error(`Unsupported dtype in weight '${name}': ${t.dtype}`);\n }\n const spec: WeightsManifestEntry = {name, shape: t.shape, dtype: t.dtype};\n if (t.dtype === 'string') {\n const utf8bytes = new Promise(async resolve => {\n const vals = await t.bytes() as Uint8Array[];\n const totalNumBytes = vals.reduce((p, c) => p + c.length, 0) +\n NUM_BYTES_STRING_LENGTH * vals.length;\n const bytes = new Uint8Array(totalNumBytes);\n let offset = 0;\n for (let i = 0; i < vals.length; i++) {\n const val = vals[i];\n const bytesOfLength =\n new Uint8Array(new Uint32Array([val.length]).buffer);\n bytes.set(bytesOfLength, offset);\n offset += NUM_BYTES_STRING_LENGTH;\n bytes.set(val, offset);\n offset += val.length;\n }\n resolve(bytes);\n });\n dataPromises.push(utf8bytes);\n } else {\n dataPromises.push(t.data());\n }\n if (group != null) {\n spec.group = group;\n }\n specs.push(spec);\n }\n\n const tensorValues = await Promise.all(dataPromises);\n return {data: concatenateTypedArrays(tensorValues), specs};\n}\n\n/**\n * Decode flat ArrayBuffer as weights.\n *\n * This function does not handle sharding.\n *\n * This function is the reverse of `encodeWeights`.\n *\n * @param buffer A flat ArrayBuffer carrying the binary values of the tensors\n * concatenated in the order specified in `specs`.\n * @param specs Specifications of the names, dtypes and shapes of the tensors\n * whose value are encoded by `buffer`.\n * @return A map from tensor name to tensor value, with the names corresponding\n * to names in `specs`.\n * @throws Error, if any of the tensors has unsupported dtype.\n */\nexport function decodeWeights(\n buffer: ArrayBuffer, specs: WeightsManifestEntry[]): NamedTensorMap {\n // TODO(adarob, cais): Support quantization.\n const out: NamedTensorMap = {};\n let float16Decode: (buffer: Uint16Array) => Float32Array | undefined;\n let offset = 0;\n for (const spec of specs) {\n const name = spec.name;\n const dtype = spec.dtype;\n const shape = spec.shape;\n const size = sizeFromShape(shape);\n let values: TypedArray|string[]|Uint8Array[];\n\n if ('quantization' in spec) {\n const quantization = spec.quantization;\n if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n if (!('min' in quantization && 'scale' in quantization)) {\n throw new Error(\n `Weight ${spec.name} with quantization ${quantization.dtype} ` +\n `doesn't have corresponding metadata min and scale.`);\n }\n } else if (quantization.dtype === 'float16') {\n if (dtype !== 'float32') {\n throw new Error(\n `Weight ${spec.name} is quantized with ${quantization.dtype} ` +\n `which only supports weights of type float32 not ${dtype}.`);\n }\n } else {\n throw new Error(\n `Weight ${spec.name} has unknown ` +\n `quantization dtype ${quantization.dtype}. ` +\n `Supported quantization dtypes are: ` +\n `'uint8', 'uint16', and 'float16'.`);\n }\n const quantizationSizeFactor = DTYPE_VALUE_SIZE_MAP[quantization.dtype];\n const byteBuffer =\n buffer.slice(offset, offset + size * quantizationSizeFactor);\n const quantizedArray = (quantization.dtype === 'uint8') ?\n new Uint8Array(byteBuffer) :\n new Uint16Array(byteBuffer);\n if (dtype === 'float32') {\n if (quantization.dtype === 'uint8' || quantization.dtype === 'uint16') {\n values = new Float32Array(quantizedArray.length);\n for (let i = 0; i < quantizedArray.length; i++) {\n const v = quantizedArray[i];\n values[i] = v * quantization.scale + quantization.min;\n }\n } else if (quantization.dtype === 'float16') {\n if (float16Decode === undefined) {\n float16Decode = getFloat16Decoder();\n }\n values = float16Decode(quantizedArray as Uint16Array);\n } else {\n throw new Error(\n `Unsupported quantization type ${quantization.dtype} ` +\n `for weight type float32.`);\n }\n } else if (dtype === 'int32') {\n if (quantization.dtype !== 'uint8' && quantization.dtype !== 'uint16') {\n throw new Error(\n `Unsupported quantization type ${quantization.dtype} ` +\n `for weight type int32.`);\n }\n values = new Int32Array(quantizedArray.length);\n for (let i = 0; i < quantizedArray.length; i++) {\n const v = quantizedArray[i];\n values[i] = Math.round(v * quantization.scale + quantization.min);\n }\n } else {\n throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n }\n offset += size * quantizationSizeFactor;\n } else if (dtype === 'string') {\n const size = sizeFromShape(spec.shape);\n values = [];\n for (let i = 0; i < size; i++) {\n const byteLength = new Uint32Array(\n buffer.slice(offset, offset + NUM_BYTES_STRING_LENGTH))[0];\n offset += NUM_BYTES_STRING_LENGTH;\n const bytes = new Uint8Array(buffer.slice(offset, offset + byteLength));\n (values as Uint8Array[]).push(bytes);\n offset += byteLength;\n }\n } else {\n const dtypeFactor = DTYPE_VALUE_SIZE_MAP[dtype];\n const byteBuffer = buffer.slice(offset, offset + size * dtypeFactor);\n\n if (dtype === 'float32') {\n values = new Float32Array(byteBuffer);\n } else if (dtype === 'int32') {\n values = new Int32Array(byteBuffer);\n } else if (dtype === 'bool') {\n values = new Uint8Array(byteBuffer);\n } else if (dtype === 'complex64') {\n values = new Float32Array(byteBuffer);\n const real = new Float32Array(values.length / 2);\n const image = new Float32Array(values.length / 2);\n for (let i = 0; i < real.length; i++) {\n real[i] = values[i * 2];\n image[i] = values[i * 2 + 1];\n }\n const realTensor = tensor(real, shape, 'float32');\n const imageTensor = tensor(image, shape, 'float32');\n out[name] = complex(realTensor, imageTensor);\n } else {\n throw new Error(`Unsupported dtype in weight '${name}': ${dtype}`);\n }\n offset += size * dtypeFactor;\n }\n if (dtype !== 'complex64') {\n out[name] = tensor(values, shape, dtype);\n }\n }\n return out;\n}\n\n/**\n * Concatenate TypedArrays into an ArrayBuffer.\n */\nexport function concatenateTypedArrays(xs: TypedArray[]): ArrayBuffer {\n // TODO(adarob, cais): Support quantization.\n if (xs === null) {\n throw new Error(`Invalid input value: ${JSON.stringify(xs)}`);\n }\n\n let totalByteLength = 0;\n\n // `normalizedXs` is here for this reason: a `TypedArray`'s `buffer'\n // can have a different byte length from that of the `TypedArray` itself,\n // for example, when the `TypedArray` is created from an offset in an\n // `ArrayBuffer`. `normliazedXs` holds `TypedArray`s whose `buffer`s match\n // the `TypedArray` in byte length. If an element of `xs` does not show\n // this property, a new `TypedArray` that satisfy this property will be\n // constructed and pushed into `normalizedXs`.\n const normalizedXs: TypedArray[] = [];\n xs.forEach((x: TypedArray) => {\n totalByteLength += x.byteLength;\n // tslint:disable:no-any\n normalizedXs.push(\n x.byteLength === x.buffer.byteLength ? x :\n new (x.constructor as any)(x));\n if (!(x as any instanceof Float32Array || x as any instanceof Int32Array ||\n x as any instanceof Uint8Array)) {\n throw new Error(`Unsupported TypedArray subtype: ${x.constructor.name}`);\n }\n // tslint:enable:no-any\n });\n\n const y = new Uint8Array(totalByteLength);\n let offset = 0;\n normalizedXs.forEach((x: TypedArray) => {\n y.set(new Uint8Array(x.buffer), offset);\n offset += x.byteLength;\n });\n\n return y.buffer;\n}\n\n// Use Buffer on Node.js instead of Blob/atob/btoa\nconst useNodeBuffer = typeof Buffer !== 'undefined' &&\n (typeof Blob === 'undefined' || typeof atob === 'undefined' ||\n typeof btoa === 'undefined');\n\n/**\n * Calculate the byte length of a JavaScript string.\n *\n * Note that a JavaScript string can contain wide characters, therefore the\n * length of the string is not necessarily equal to the byte length.\n *\n * @param str Input string.\n * @returns Byte length.\n */\nexport function stringByteLength(str: string): number {\n if (useNodeBuffer) {\n return Buffer.byteLength(str);\n }\n return new Blob([str]).size;\n}\n\n/**\n * Encode an ArrayBuffer as a base64 encoded string.\n *\n * @param buffer `ArrayBuffer` to be converted.\n * @returns A string that base64-encodes `buffer`.\n */\nexport function arrayBufferToBase64String(buffer: ArrayBuffer): string {\n if (useNodeBuffer) {\n return Buffer.from(buffer).toString('base64');\n }\n const buf = new Uint8Array(buffer);\n let s = '';\n for (let i = 0, l = buf.length; i < l; i++) {\n s += String.fromCharCode(buf[i]);\n }\n return btoa(s);\n}\n\n/**\n * Decode a base64 string as an ArrayBuffer.\n *\n * @param str Base64 string.\n * @returns Decoded `ArrayBuffer`.\n */\nexport function base64StringToArrayBuffer(str: string): ArrayBuffer {\n if (useNodeBuffer) {\n const buf = Buffer.from(str, 'base64');\n return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength);\n }\n const s = atob(str);\n const buffer = new Uint8Array(s.length);\n for (let i = 0; i < s.length; ++i) {\n buffer.set([s.charCodeAt(i)], i);\n }\n return buffer.buffer;\n}\n\n/**\n * Concatenate a number of ArrayBuffers into one.\n *\n * @param buffers A number of array buffers to concatenate.\n * @returns Result of concatenating `buffers` in order.\n */\nexport function concatenateArrayBuffers(buffers: ArrayBuffer[]): ArrayBuffer {\n if (buffers.length === 1) {\n return buffers[0];\n }\n\n let totalByteLength = 0;\n buffers.forEach((buffer: ArrayBuffer) => {\n totalByteLength += buffer.byteLength;\n });\n\n const temp = new Uint8Array(totalByteLength);\n let offset = 0;\n buffers.forEach((buffer: ArrayBuffer) => {\n temp.set(new Uint8Array(buffer), offset);\n offset += buffer.byteLength;\n });\n return temp.buffer;\n}\n\n/**\n * Get the basename of a path.\n *\n * Behaves in a way analogous to Linux's basename command.\n *\n * @param path\n */\nexport function basename(path: string): string {\n const SEPARATOR = '/';\n path = path.trim();\n while (path.endsWith(SEPARATOR)) {\n path = path.slice(0, path.length - 1);\n }\n const items = path.split(SEPARATOR);\n return items[items.length - 1];\n}\n\n/**\n * Populate ModelArtifactsInfo fields for a model with JSON topology.\n * @param modelArtifacts\n * @returns A ModelArtifactsInfo object.\n */\nexport function getModelArtifactsInfoForJSON(modelArtifacts: ModelArtifacts):\n ModelArtifactsInfo {\n if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n throw new Error('Expected JSON model topology, received ArrayBuffer.');\n }\n\n return {\n dateSaved: new Date(),\n modelTopologyType: 'JSON',\n modelTopologyBytes: modelArtifacts.modelTopology == null ?\n 0 :\n stringByteLength(JSON.stringify(modelArtifacts.modelTopology)),\n weightSpecsBytes: modelArtifacts.weightSpecs == null ?\n 0 :\n stringByteLength(JSON.stringify(modelArtifacts.weightSpecs)),\n weightDataBytes: modelArtifacts.weightData == null ?\n 0 :\n modelArtifacts.weightData.byteLength,\n };\n}\n\n/**\n * Computes mantisa table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 2048 mantissa lookup values.\n */\nfunction computeFloat16MantisaTable(): Uint32Array {\n const convertMantissa = (i: number): number => {\n let m = i << 13;\n let e = 0;\n\n while ((m & 0x00800000) === 0) {\n e -= 0x00800000;\n m <<= 1;\n }\n m &= ~0x00800000;\n e += 0x38800000;\n\n return m | e;\n };\n\n const mantisaTable = new Uint32Array(2048);\n\n mantisaTable[0] = 0;\n for (let i = 1; i < 1024; i++) {\n mantisaTable[i] = convertMantissa(i);\n }\n for (let i = 1024; i < 2048; i++) {\n mantisaTable[i] = 0x38000000 + ((i - 1024) << 13);\n }\n\n return mantisaTable;\n}\n\n/**\n * Computes exponent table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 64 exponent lookup values.\n */\nfunction computeFloat16ExponentTable(): Uint32Array {\n const exponentTable = new Uint32Array(64);\n\n exponentTable[0] = 0;\n exponentTable[31] = 0x47800000;\n exponentTable[32] = 0x80000000;\n exponentTable[63] = 0xc7800000;\n for (let i = 1; i < 31; i++) {\n exponentTable[i] = i << 23;\n }\n for (let i = 33; i < 63; i++) {\n exponentTable[i] = 0x80000000 + ((i - 32) << 23);\n }\n\n return exponentTable;\n}\n\n/**\n * Computes offset table for casting Float16 to Float32\n * See http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n *\n * @returns Uint32Array, 6d offset values.\n */\nfunction computeFloat16OffsetTable(): Uint32Array {\n const offsetTable = new Uint32Array(64);\n\n for (let i = 0; i < 64; i++) {\n offsetTable[i] = 1024;\n }\n offsetTable[0] = offsetTable[32] = 0;\n\n return offsetTable;\n}\n\n/**\n * Retrieve a Float16 decoder which will decode a ByteArray of Float16 values\n * to a Float32Array.\n *\n * @returns Function (buffer: Uint16Array) => Float32Array which decodes\n * the Uint16Array of Float16 bytes to a Float32Array.\n */\nexport function getFloat16Decoder(): (buffer: Uint16Array) => Float32Array {\n // Algorithm is based off of\n // http://www.fox-toolkit.org/ftp/fasthalffloatconversion.pdf\n\n // Cache lookup tables\n const mantisaTable = computeFloat16MantisaTable();\n const exponentTable = computeFloat16ExponentTable();\n const offsetTable = computeFloat16OffsetTable();\n\n return (quantizedArray: Uint16Array) => {\n const buffer = new ArrayBuffer(4 * quantizedArray.length);\n const bufferUint32View = new Uint32Array(buffer);\n for (let index = 0; index < quantizedArray.length; index++) {\n const float16Bits = quantizedArray[index];\n const float32Bits =\n mantisaTable[offsetTable[float16Bits >> 10] + (float16Bits & 0x3ff)] +\n exponentTable[float16Bits >> 10];\n bufferUint32View[index] = float32Bits;\n }\n return new Float32Array(buffer);\n };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {IOHandler, LoadOptions} from './types';\n\nexport type IORouter = (url: string|string[], loadOptions?: LoadOptions) =>\n IOHandler;\n\nexport class IORouterRegistry {\n // Singleton instance.\n private static instance: IORouterRegistry;\n\n private saveRouters: IORouter[];\n private loadRouters: IORouter[];\n\n private constructor() {\n this.saveRouters = [];\n this.loadRouters = [];\n }\n\n private static getInstance(): IORouterRegistry {\n if (IORouterRegistry.instance == null) {\n IORouterRegistry.instance = new IORouterRegistry();\n }\n return IORouterRegistry.instance;\n }\n\n /**\n * Register a save-handler router.\n *\n * @param saveRouter A function that maps a URL-like string onto an instance\n * of `IOHandler` with the `save` method defined or `null`.\n */\n static registerSaveRouter(saveRouter: IORouter) {\n IORouterRegistry.getInstance().saveRouters.push(saveRouter);\n }\n\n /**\n * Register a load-handler router.\n *\n * @param loadRouter A function that maps a URL-like string onto an instance\n * of `IOHandler` with the `load` method defined or `null`.\n */\n static registerLoadRouter(loadRouter: IORouter) {\n IORouterRegistry.getInstance().loadRouters.push(loadRouter);\n }\n\n /**\n * Look up IOHandler for saving, given a URL-like string.\n *\n * @param url\n * @returns If only one match is found, an instance of IOHandler with the\n * `save` method defined. If no match is found, `null`.\n * @throws Error, if more than one match is found.\n */\n static getSaveHandlers(url: string|string[]): IOHandler[] {\n return IORouterRegistry.getHandlers(url, 'save');\n }\n\n /**\n * Look up IOHandler for loading, given a URL-like string.\n *\n * @param url\n * @param loadOptions Optional, custom load options.\n * @returns All valid handlers for `url`, given the currently registered\n * handler routers.\n */\n static getLoadHandlers(url: string|string[], loadOptions?: LoadOptions):\n IOHandler[] {\n return IORouterRegistry.getHandlers(url, 'load', loadOptions);\n }\n\n private static getHandlers(\n url: string|string[], handlerType: 'save'|'load',\n loadOptions?: LoadOptions): IOHandler[] {\n const validHandlers: IOHandler[] = [];\n const routers = handlerType === 'load' ?\n IORouterRegistry.getInstance().loadRouters :\n IORouterRegistry.getInstance().saveRouters;\n routers.forEach(router => {\n const handler = router(url, loadOptions);\n if (handler !== null) {\n validHandlers.push(handler);\n }\n });\n return validHandlers;\n }\n}\n\nexport const registerSaveRouter = (loudRouter: IORouter) =>\n IORouterRegistry.registerSaveRouter(loudRouter);\nexport const registerLoadRouter = (loudRouter: IORouter) =>\n IORouterRegistry.registerLoadRouter(loudRouter);\nexport const getSaveHandlers = (url: string|string[]) =>\n IORouterRegistry.getSaveHandlers(url);\nexport const getLoadHandlers =\n (url: string|string[], loadOptions?: LoadOptions) =>\n IORouterRegistry.getLoadHandlers(url, loadOptions);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\n\nimport {getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst DATABASE_NAME = 'tensorflowjs';\nconst DATABASE_VERSION = 1;\n\n// Model data and ModelArtifactsInfo (metadata) are stored in two separate\n// stores for efficient access of the list of stored models and their metadata.\n// 1. The object store for model data: topology, weights and weight manifests.\nconst MODEL_STORE_NAME = 'models_store';\n// 2. The object store for ModelArtifactsInfo, including meta-information such\n// as the type of topology (JSON vs binary), byte size of the topology, byte\n// size of the weights, etc.\nconst INFO_STORE_NAME = 'model_info_store';\n\n/**\n * Delete the entire database for tensorflow.js, including the models store.\n */\nexport async function deleteDatabase(): Promise {\n const idbFactory = getIndexedDBFactory();\n\n return new Promise((resolve, reject) => {\n const deleteRequest = idbFactory.deleteDatabase(DATABASE_NAME);\n deleteRequest.onsuccess = () => resolve();\n deleteRequest.onerror = error => reject(error);\n });\n}\n\nfunction getIndexedDBFactory(): IDBFactory {\n if (!env().getBool('IS_BROWSER')) {\n // TODO(cais): Add more info about what IOHandler subtypes are available.\n // Maybe point to a doc page on the web and/or automatically determine\n // the available IOHandlers and print them in the error message.\n throw new Error(\n 'Failed to obtain IndexedDB factory because the current environment' +\n 'is not a web browser.');\n }\n // tslint:disable-next-line:no-any\n const theWindow: any = typeof window === 'undefined' ? self : window;\n const factory = theWindow.indexedDB || theWindow.mozIndexedDB ||\n theWindow.webkitIndexedDB || theWindow.msIndexedDB ||\n theWindow.shimIndexedDB;\n if (factory == null) {\n throw new Error(\n 'The current browser does not appear to support IndexedDB.');\n }\n return factory;\n}\n\nfunction setUpDatabase(openRequest: IDBRequest) {\n const db = openRequest.result as IDBDatabase;\n db.createObjectStore(MODEL_STORE_NAME, {keyPath: 'modelPath'});\n db.createObjectStore(INFO_STORE_NAME, {keyPath: 'modelPath'});\n}\n\n/**\n * IOHandler subclass: Browser IndexedDB.\n *\n * See the doc string of `browserIndexedDB` for more details.\n */\nexport class BrowserIndexedDB implements IOHandler {\n protected readonly indexedDB: IDBFactory;\n protected readonly modelPath: string;\n\n static readonly URL_SCHEME = 'indexeddb://';\n\n constructor(modelPath: string) {\n this.indexedDB = getIndexedDBFactory();\n\n if (modelPath == null || !modelPath) {\n throw new Error(\n 'For IndexedDB, modelPath must not be null, undefined or empty.');\n }\n this.modelPath = modelPath;\n }\n\n async save(modelArtifacts: ModelArtifacts): Promise {\n // TODO(cais): Support saving GraphDef models.\n if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n throw new Error(\n 'BrowserLocalStorage.save() does not support saving model topology ' +\n 'in binary formats yet.');\n }\n\n return this.databaseAction(this.modelPath, modelArtifacts) as\n Promise;\n }\n\n async load(): Promise {\n return this.databaseAction(this.modelPath) as Promise;\n }\n\n /**\n * Perform database action to put model artifacts into or read model artifacts\n * from IndexedDB object store.\n *\n * Whether the action is put or get depends on whether `modelArtifacts` is\n * specified. If it is specified, the action will be put; otherwise the action\n * will be get.\n *\n * @param modelPath A unique string path for the model.\n * @param modelArtifacts If specified, it will be the model artifacts to be\n * stored in IndexedDB.\n * @returns A `Promise` of `SaveResult`, if the action is put, or a `Promise`\n * of `ModelArtifacts`, if the action is get.\n */\n private databaseAction(modelPath: string, modelArtifacts?: ModelArtifacts):\n Promise {\n return new Promise((resolve, reject) => {\n const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n openRequest.onsuccess = () => {\n const db = openRequest.result;\n\n if (modelArtifacts == null) {\n // Read model out from object store.\n const modelTx = db.transaction(MODEL_STORE_NAME, 'readonly');\n const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n const getRequest = modelStore.get(this.modelPath);\n getRequest.onsuccess = () => {\n if (getRequest.result == null) {\n db.close();\n return reject(new Error(\n `Cannot find model with path '${this.modelPath}' ` +\n `in IndexedDB.`));\n } else {\n resolve(getRequest.result.modelArtifacts);\n }\n };\n getRequest.onerror = error => {\n db.close();\n return reject(getRequest.error);\n };\n modelTx.oncomplete = () => db.close();\n } else {\n // Put model into object store.\n const modelArtifactsInfo: ModelArtifactsInfo =\n getModelArtifactsInfoForJSON(modelArtifacts);\n // First, put ModelArtifactsInfo into info store.\n const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n let infoStore = infoTx.objectStore(INFO_STORE_NAME);\n const putInfoRequest =\n infoStore.put({modelPath: this.modelPath, modelArtifactsInfo});\n let modelTx: IDBTransaction;\n putInfoRequest.onsuccess = () => {\n // Second, put model data into model store.\n modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n const putModelRequest = modelStore.put({\n modelPath: this.modelPath,\n modelArtifacts,\n modelArtifactsInfo\n });\n putModelRequest.onsuccess = () => resolve({modelArtifactsInfo});\n putModelRequest.onerror = error => {\n // If the put-model request fails, roll back the info entry as\n // well.\n infoStore = infoTx.objectStore(INFO_STORE_NAME);\n const deleteInfoRequest = infoStore.delete(this.modelPath);\n deleteInfoRequest.onsuccess = () => {\n db.close();\n return reject(putModelRequest.error);\n };\n deleteInfoRequest.onerror = error => {\n db.close();\n return reject(putModelRequest.error);\n };\n };\n };\n putInfoRequest.onerror = error => {\n db.close();\n return reject(putInfoRequest.error);\n };\n infoTx.oncomplete = () => {\n if (modelTx == null) {\n db.close();\n } else {\n modelTx.oncomplete = () => db.close();\n }\n };\n }\n };\n openRequest.onerror = error => reject(openRequest.error);\n });\n }\n}\n\nexport const indexedDBRouter: IORouter = (url: string|string[]) => {\n if (!env().getBool('IS_BROWSER')) {\n return null;\n } else {\n if (!Array.isArray(url) && url.startsWith(BrowserIndexedDB.URL_SCHEME)) {\n return browserIndexedDB(url.slice(BrowserIndexedDB.URL_SCHEME.length));\n } else {\n return null;\n }\n }\n};\nIORouterRegistry.registerSaveRouter(indexedDBRouter);\nIORouterRegistry.registerLoadRouter(indexedDBRouter);\n\n/**\n * Creates a browser IndexedDB IOHandler for saving and loading models.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n * tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save('indexeddb://MyModel'));\n * console.log(saveResult);\n * ```\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n * non-empty string.\n * @returns An instance of `BrowserIndexedDB` (sublcass of `IOHandler`),\n * which can be used with, e.g., `tf.Model.save`.\n */\nexport function browserIndexedDB(modelPath: string): IOHandler {\n return new BrowserIndexedDB(modelPath);\n}\n\nfunction maybeStripScheme(key: string) {\n return key.startsWith(BrowserIndexedDB.URL_SCHEME) ?\n key.slice(BrowserIndexedDB.URL_SCHEME.length) :\n key;\n}\n\nexport class BrowserIndexedDBManager implements ModelStoreManager {\n private indexedDB: IDBFactory;\n\n constructor() {\n this.indexedDB = getIndexedDBFactory();\n }\n\n async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n return new Promise<{[path: string]: ModelArtifactsInfo}>(\n (resolve, reject) => {\n const openRequest =\n this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n openRequest.onsuccess = () => {\n const db = openRequest.result;\n const tx = db.transaction(INFO_STORE_NAME, 'readonly');\n const store = tx.objectStore(INFO_STORE_NAME);\n // tslint:disable:max-line-length\n // Need to cast `store` as `any` here because TypeScript's DOM\n // library does not have the `getAll()` method even though the\n // method is supported in the latest version of most mainstream\n // browsers:\n // https://developer.mozilla.org/en-US/docs/Web/API/IDBObjectStore/getAll\n // tslint:enable:max-line-length\n // tslint:disable-next-line:no-any\n const getAllInfoRequest = (store as any).getAll() as IDBRequest;\n getAllInfoRequest.onsuccess = () => {\n const out: {[path: string]: ModelArtifactsInfo} = {};\n for (const item of getAllInfoRequest.result) {\n out[item.modelPath] = item.modelArtifactsInfo;\n }\n resolve(out);\n };\n getAllInfoRequest.onerror = error => {\n db.close();\n return reject(getAllInfoRequest.error);\n };\n tx.oncomplete = () => db.close();\n };\n openRequest.onerror = error => reject(openRequest.error);\n });\n }\n\n async removeModel(path: string): Promise {\n path = maybeStripScheme(path);\n return new Promise((resolve, reject) => {\n const openRequest = this.indexedDB.open(DATABASE_NAME, DATABASE_VERSION);\n openRequest.onupgradeneeded = () => setUpDatabase(openRequest);\n\n openRequest.onsuccess = () => {\n const db = openRequest.result;\n const infoTx = db.transaction(INFO_STORE_NAME, 'readwrite');\n const infoStore = infoTx.objectStore(INFO_STORE_NAME);\n\n const getInfoRequest = infoStore.get(path);\n let modelTx: IDBTransaction;\n getInfoRequest.onsuccess = () => {\n if (getInfoRequest.result == null) {\n db.close();\n return reject(new Error(\n `Cannot find model with path '${path}' ` +\n `in IndexedDB.`));\n } else {\n // First, delete the entry in the info store.\n const deleteInfoRequest = infoStore.delete(path);\n const deleteModelData = () => {\n // Second, delete the entry in the model store.\n modelTx = db.transaction(MODEL_STORE_NAME, 'readwrite');\n const modelStore = modelTx.objectStore(MODEL_STORE_NAME);\n const deleteModelRequest = modelStore.delete(path);\n deleteModelRequest.onsuccess = () =>\n resolve(getInfoRequest.result.modelArtifactsInfo);\n deleteModelRequest.onerror = error =>\n reject(getInfoRequest.error);\n };\n // Proceed with deleting model data regardless of whether deletion\n // of info data succeeds or not.\n deleteInfoRequest.onsuccess = deleteModelData;\n deleteInfoRequest.onerror = error => {\n deleteModelData();\n db.close();\n return reject(getInfoRequest.error);\n };\n }\n };\n getInfoRequest.onerror = error => {\n db.close();\n return reject(getInfoRequest.error);\n };\n\n infoTx.oncomplete = () => {\n if (modelTx == null) {\n db.close();\n } else {\n modelTx.oncomplete = () => db.close();\n }\n };\n };\n openRequest.onerror = error => reject(openRequest.error);\n });\n }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {arrayBufferToBase64String, base64StringToArrayBuffer, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelArtifactsInfo, ModelStoreManager, SaveResult} from './types';\n\nconst PATH_SEPARATOR = '/';\nconst PATH_PREFIX = 'tensorflowjs_models';\nconst INFO_SUFFIX = 'info';\nconst MODEL_TOPOLOGY_SUFFIX = 'model_topology';\nconst WEIGHT_SPECS_SUFFIX = 'weight_specs';\nconst WEIGHT_DATA_SUFFIX = 'weight_data';\nconst MODEL_METADATA_SUFFIX = 'model_metadata';\n\n/**\n * Purge all tensorflow.js-saved model artifacts from local storage.\n *\n * @returns Paths of the models purged.\n */\nexport function purgeLocalStorageArtifacts(): string[] {\n if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n typeof window.localStorage === 'undefined') {\n throw new Error(\n 'purgeLocalStorageModels() cannot proceed because local storage is ' +\n 'unavailable in the current environment.');\n }\n const LS = window.localStorage;\n const purgedModelPaths: string[] = [];\n for (let i = 0; i < LS.length; ++i) {\n const key = LS.key(i);\n const prefix = PATH_PREFIX + PATH_SEPARATOR;\n if (key.startsWith(prefix) && key.length > prefix.length) {\n LS.removeItem(key);\n const modelName = getModelPathFromKey(key);\n if (purgedModelPaths.indexOf(modelName) === -1) {\n purgedModelPaths.push(modelName);\n }\n }\n }\n return purgedModelPaths;\n}\n\nfunction getModelKeys(path: string): {\n info: string,\n topology: string,\n weightSpecs: string,\n weightData: string,\n modelMetadata: string\n} {\n return {\n info: [PATH_PREFIX, path, INFO_SUFFIX].join(PATH_SEPARATOR),\n topology: [PATH_PREFIX, path, MODEL_TOPOLOGY_SUFFIX].join(PATH_SEPARATOR),\n weightSpecs: [PATH_PREFIX, path, WEIGHT_SPECS_SUFFIX].join(PATH_SEPARATOR),\n weightData: [PATH_PREFIX, path, WEIGHT_DATA_SUFFIX].join(PATH_SEPARATOR),\n modelMetadata:\n [PATH_PREFIX, path, MODEL_METADATA_SUFFIX].join(PATH_SEPARATOR)\n };\n}\n\n/**\n * Get model path from a local-storage key.\n *\n * E.g., 'tensorflowjs_models/my/model/1/info' --> 'my/model/1'\n *\n * @param key\n */\nfunction getModelPathFromKey(key: string) {\n const items = key.split(PATH_SEPARATOR);\n if (items.length < 3) {\n throw new Error(`Invalid key format: ${key}`);\n }\n return items.slice(1, items.length - 1).join(PATH_SEPARATOR);\n}\n\nfunction maybeStripScheme(key: string) {\n return key.startsWith(BrowserLocalStorage.URL_SCHEME) ?\n key.slice(BrowserLocalStorage.URL_SCHEME.length) :\n key;\n}\n\ndeclare type LocalStorageKeys = {\n info: string,\n topology: string,\n weightSpecs: string,\n weightData: string,\n modelMetadata: string\n};\n\n/**\n * IOHandler subclass: Browser Local Storage.\n *\n * See the doc string to `browserLocalStorage` for more details.\n */\nexport class BrowserLocalStorage implements IOHandler {\n protected readonly LS: Storage;\n protected readonly modelPath: string;\n protected readonly keys: LocalStorageKeys;\n\n static readonly URL_SCHEME = 'localstorage://';\n\n constructor(modelPath: string) {\n if (!env().getBool('IS_BROWSER') || typeof window === 'undefined' ||\n typeof window.localStorage === 'undefined') {\n // TODO(cais): Add more info about what IOHandler subtypes are\n // available.\n // Maybe point to a doc page on the web and/or automatically determine\n // the available IOHandlers and print them in the error message.\n throw new Error(\n 'The current environment does not support local storage.');\n }\n this.LS = window.localStorage;\n\n if (modelPath == null || !modelPath) {\n throw new Error(\n 'For local storage, modelPath must not be null, undefined or empty.');\n }\n this.modelPath = modelPath;\n this.keys = getModelKeys(this.modelPath);\n }\n\n /**\n * Save model artifacts to browser local storage.\n *\n * See the documentation to `browserLocalStorage` for details on the saved\n * artifacts.\n *\n * @param modelArtifacts The model artifacts to be stored.\n * @returns An instance of SaveResult.\n */\n async save(modelArtifacts: ModelArtifacts): Promise {\n if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n throw new Error(\n 'BrowserLocalStorage.save() does not support saving model topology ' +\n 'in binary formats yet.');\n } else {\n const topology = JSON.stringify(modelArtifacts.modelTopology);\n const weightSpecs = JSON.stringify(modelArtifacts.weightSpecs);\n\n const modelArtifactsInfo: ModelArtifactsInfo =\n getModelArtifactsInfoForJSON(modelArtifacts);\n\n try {\n this.LS.setItem(this.keys.info, JSON.stringify(modelArtifactsInfo));\n this.LS.setItem(this.keys.topology, topology);\n this.LS.setItem(this.keys.weightSpecs, weightSpecs);\n this.LS.setItem(\n this.keys.weightData,\n arrayBufferToBase64String(modelArtifacts.weightData));\n this.LS.setItem(this.keys.modelMetadata, JSON.stringify({\n format: modelArtifacts.format,\n generatedBy: modelArtifacts.generatedBy,\n convertedBy: modelArtifacts.convertedBy,\n userDefinedMetadata: modelArtifacts.userDefinedMetadata\n }));\n\n return {modelArtifactsInfo};\n } catch (err) {\n // If saving failed, clean up all items saved so far.\n this.LS.removeItem(this.keys.info);\n this.LS.removeItem(this.keys.topology);\n this.LS.removeItem(this.keys.weightSpecs);\n this.LS.removeItem(this.keys.weightData);\n this.LS.removeItem(this.keys.modelMetadata);\n\n throw new Error(\n `Failed to save model '${this.modelPath}' to local storage: ` +\n `size quota being exceeded is a possible cause of this failure: ` +\n `modelTopologyBytes=${modelArtifactsInfo.modelTopologyBytes}, ` +\n `weightSpecsBytes=${modelArtifactsInfo.weightSpecsBytes}, ` +\n `weightDataBytes=${modelArtifactsInfo.weightDataBytes}.`);\n }\n }\n }\n\n /**\n * Load a model from local storage.\n *\n * See the documentation to `browserLocalStorage` for details on the saved\n * artifacts.\n *\n * @returns The loaded model (if loading succeeds).\n */\n async load(): Promise {\n const info =\n JSON.parse(this.LS.getItem(this.keys.info)) as ModelArtifactsInfo;\n if (info == null) {\n throw new Error(\n `In local storage, there is no model with name '${this.modelPath}'`);\n }\n\n if (info.modelTopologyType !== 'JSON') {\n throw new Error(\n 'BrowserLocalStorage does not support loading non-JSON model ' +\n 'topology yet.');\n }\n\n const out: ModelArtifacts = {};\n\n // Load topology.\n const topology = JSON.parse(this.LS.getItem(this.keys.topology));\n if (topology == null) {\n throw new Error(\n `In local storage, the topology of model '${this.modelPath}' ` +\n `is missing.`);\n }\n out.modelTopology = topology;\n\n // Load weight specs.\n const weightSpecs = JSON.parse(this.LS.getItem(this.keys.weightSpecs));\n if (weightSpecs == null) {\n throw new Error(\n `In local storage, the weight specs of model '${this.modelPath}' ` +\n `are missing.`);\n }\n out.weightSpecs = weightSpecs;\n\n // Load meta-data fields.\n const metadataString = this.LS.getItem(this.keys.modelMetadata);\n if (metadataString != null) {\n const metadata = JSON.parse(metadataString) as ModelArtifacts;\n out.format = metadata['format'];\n out.generatedBy = metadata['generatedBy'];\n out.convertedBy = metadata['convertedBy'];\n out.userDefinedMetadata = metadata['userDefinedMetadata'];\n }\n\n // Load weight data.\n const weightDataBase64 = this.LS.getItem(this.keys.weightData);\n if (weightDataBase64 == null) {\n throw new Error(\n `In local storage, the binary weight values of model ` +\n `'${this.modelPath}' are missing.`);\n }\n out.weightData = base64StringToArrayBuffer(weightDataBase64);\n\n return out;\n }\n}\n\nexport const localStorageRouter: IORouter = (url: string|string[]) => {\n if (!env().getBool('IS_BROWSER')) {\n return null;\n } else {\n if (!Array.isArray(url) && url.startsWith(BrowserLocalStorage.URL_SCHEME)) {\n return browserLocalStorage(\n url.slice(BrowserLocalStorage.URL_SCHEME.length));\n } else {\n return null;\n }\n }\n};\nIORouterRegistry.registerSaveRouter(localStorageRouter);\nIORouterRegistry.registerLoadRouter(localStorageRouter);\n\n/**\n * Factory function for local storage IOHandler.\n *\n * This `IOHandler` supports both `save` and `load`.\n *\n * For each model's saved artifacts, four items are saved to local storage.\n * - `${PATH_SEPARATOR}/${modelPath}/info`: Contains meta-info about the\n * model, such as date saved, type of the topology, size in bytes, etc.\n * - `${PATH_SEPARATOR}/${modelPath}/topology`: Model topology. For Keras-\n * style models, this is a stringized JSON.\n * - `${PATH_SEPARATOR}/${modelPath}/weight_specs`: Weight specs of the\n * model, can be used to decode the saved binary weight values (see\n * item below).\n * - `${PATH_SEPARATOR}/${modelPath}/weight_data`: Concatenated binary\n * weight values, stored as a base64-encoded string.\n *\n * Saving may throw an `Error` if the total size of the artifacts exceed the\n * browser-specific quota.\n *\n * @param modelPath A unique identifier for the model to be saved. Must be a\n * non-empty string.\n * @returns An instance of `IOHandler`, which can be used with, e.g.,\n * `tf.Model.save`.\n */\nexport function browserLocalStorage(modelPath: string): IOHandler {\n return new BrowserLocalStorage(modelPath);\n}\n\nexport class BrowserLocalStorageManager implements ModelStoreManager {\n private readonly LS: Storage;\n\n constructor() {\n assert(\n env().getBool('IS_BROWSER'),\n () => 'Current environment is not a web browser');\n assert(\n typeof window === 'undefined' ||\n typeof window.localStorage !== 'undefined',\n () => 'Current browser does not appear to support localStorage');\n this.LS = window.localStorage;\n }\n\n async listModels(): Promise<{[path: string]: ModelArtifactsInfo}> {\n const out: {[path: string]: ModelArtifactsInfo} = {};\n const prefix = PATH_PREFIX + PATH_SEPARATOR;\n const suffix = PATH_SEPARATOR + INFO_SUFFIX;\n for (let i = 0; i < this.LS.length; ++i) {\n const key = this.LS.key(i);\n if (key.startsWith(prefix) && key.endsWith(suffix)) {\n const modelPath = getModelPathFromKey(key);\n out[modelPath] = JSON.parse(this.LS.getItem(key)) as ModelArtifactsInfo;\n }\n }\n return out;\n }\n\n async removeModel(path: string): Promise {\n path = maybeStripScheme(path);\n const keys = getModelKeys(path);\n if (this.LS.getItem(keys.info) == null) {\n throw new Error(`Cannot find model at path '${path}'`);\n }\n const info = JSON.parse(this.LS.getItem(keys.info)) as ModelArtifactsInfo;\n\n this.LS.removeItem(keys.info);\n this.LS.removeItem(keys.topology);\n this.LS.removeItem(keys.weightSpecs);\n this.LS.removeItem(keys.weightData);\n return info;\n }\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Classes and functions for model management across multiple storage mediums.\n *\n * Supported client actions:\n * - Listing models on all registered storage mediums.\n * - Remove model by URL from any registered storage mediums, by using URL\n * string.\n * - Moving or copying model from one path to another in the same medium or from\n * one medium to another, by using URL strings.\n */\n\nimport {assert} from '../util';\n\nimport {IORouterRegistry} from './router_registry';\nimport {ModelArtifactsInfo, ModelStoreManager} from './types';\n\nconst URL_SCHEME_SUFFIX = '://';\n\nexport class ModelStoreManagerRegistry {\n // Singleton instance.\n private static instance: ModelStoreManagerRegistry;\n\n private managers: {[scheme: string]: ModelStoreManager};\n\n private constructor() {\n this.managers = {};\n }\n\n private static getInstance(): ModelStoreManagerRegistry {\n if (ModelStoreManagerRegistry.instance == null) {\n ModelStoreManagerRegistry.instance = new ModelStoreManagerRegistry();\n }\n return ModelStoreManagerRegistry.instance;\n }\n\n /**\n * Register a save-handler router.\n *\n * @param saveRouter A function that maps a URL-like string onto an instance\n * of `IOHandler` with the `save` method defined or `null`.\n */\n static registerManager(scheme: string, manager: ModelStoreManager) {\n assert(scheme != null, () => 'scheme must not be undefined or null.');\n if (scheme.endsWith(URL_SCHEME_SUFFIX)) {\n scheme = scheme.slice(0, scheme.indexOf(URL_SCHEME_SUFFIX));\n }\n assert(scheme.length > 0, () => 'scheme must not be an empty string.');\n const registry = ModelStoreManagerRegistry.getInstance();\n assert(\n registry.managers[scheme] == null,\n () => `A model store manager is already registered for scheme '${\n scheme}'.`);\n registry.managers[scheme] = manager;\n }\n\n static getManager(scheme: string): ModelStoreManager {\n const manager = this.getInstance().managers[scheme];\n if (manager == null) {\n throw new Error(`Cannot find model manager for scheme '${scheme}'`);\n }\n return manager;\n }\n\n static getSchemes(): string[] {\n return Object.keys(this.getInstance().managers);\n }\n}\n\n/**\n * Helper method for parsing a URL string into a scheme and a path.\n *\n * @param url E.g., 'localstorage://my-model'\n * @returns A dictionary with two fields: scheme and path.\n * Scheme: e.g., 'localstorage' in the example above.\n * Path: e.g., 'my-model' in the example above.\n */\nfunction parseURL(url: string): {scheme: string, path: string} {\n if (url.indexOf(URL_SCHEME_SUFFIX) === -1) {\n throw new Error(\n `The url string provided does not contain a scheme. ` +\n `Supported schemes are: ` +\n `${ModelStoreManagerRegistry.getSchemes().join(',')}`);\n }\n return {\n scheme: url.split(URL_SCHEME_SUFFIX)[0],\n path: url.split(URL_SCHEME_SUFFIX)[1],\n };\n}\n\nasync function cloneModelInternal(\n sourceURL: string, destURL: string,\n deleteSource = false): Promise {\n assert(\n sourceURL !== destURL,\n () => `Old path and new path are the same: '${sourceURL}'`);\n\n const loadHandlers = IORouterRegistry.getLoadHandlers(sourceURL);\n assert(\n loadHandlers.length > 0,\n () => `Copying failed because no load handler is found for source URL ${\n sourceURL}.`);\n assert(\n loadHandlers.length < 2,\n () => `Copying failed because more than one (${loadHandlers.length}) ` +\n `load handlers for source URL ${sourceURL}.`);\n const loadHandler = loadHandlers[0];\n\n const saveHandlers = IORouterRegistry.getSaveHandlers(destURL);\n assert(\n saveHandlers.length > 0,\n () => `Copying failed because no save handler is found for destination ` +\n `URL ${destURL}.`);\n assert(\n saveHandlers.length < 2,\n () => `Copying failed because more than one (${loadHandlers.length}) ` +\n `save handlers for destination URL ${destURL}.`);\n const saveHandler = saveHandlers[0];\n\n const sourceScheme = parseURL(sourceURL).scheme;\n const sourcePath = parseURL(sourceURL).path;\n const sameMedium = sourceScheme === parseURL(sourceURL).scheme;\n\n const modelArtifacts = await loadHandler.load();\n\n // If moving within the same storage medium, remove the old model as soon as\n // the loading is done. Without doing this, it is possible that the combined\n // size of the two models will cause the cloning to fail.\n if (deleteSource && sameMedium) {\n await ModelStoreManagerRegistry.getManager(sourceScheme)\n .removeModel(sourcePath);\n }\n\n const saveResult = await saveHandler.save(modelArtifacts);\n\n // If moving between mediums, the deletion is done after the save succeeds.\n // This guards against the case in which saving to the destination medium\n // fails.\n if (deleteSource && !sameMedium) {\n await ModelStoreManagerRegistry.getManager(sourceScheme)\n .removeModel(sourcePath);\n }\n\n return saveResult.modelArtifactsInfo;\n}\n\n/**\n * List all models stored in registered storage mediums.\n *\n * For a web browser environment, the registered mediums are Local Storage and\n * IndexedDB.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n * {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @returns A `Promise` of a dictionary mapping URLs of existing models to\n * their model artifacts info. URLs include medium-specific schemes, e.g.,\n * 'indexeddb://my/model/1'. Model artifacts info include type of the\n * model's topology, byte sizes of the topology, weights, etc.\n *\n * @doc {\n * heading: 'Models',\n * subheading: 'Management',\n * namespace: 'io',\n * ignoreCI: true\n * }\n */\nasync function listModels(): Promise<{[url: string]: ModelArtifactsInfo}> {\n const schemes = ModelStoreManagerRegistry.getSchemes();\n const out: {[url: string]: ModelArtifactsInfo} = {};\n for (const scheme of schemes) {\n const schemeOut =\n await ModelStoreManagerRegistry.getManager(scheme).listModels();\n for (const path in schemeOut) {\n const url = scheme + URL_SCHEME_SUFFIX + path;\n out[url] = schemeOut[path];\n }\n }\n return out;\n}\n\n/**\n * Remove a model specified by URL from a reigstered storage medium.\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n * {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Delete the model.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n * ```\n *\n * @param url A URL to a stored model, with a scheme prefix, e.g.,\n * 'localstorage://my-model-1', 'indexeddb://my/model/2'.\n * @returns ModelArtifactsInfo of the deleted model (if and only if deletion\n * is successful).\n * @throws Error if deletion fails, e.g., if no model exists at `path`.\n *\n * @doc {\n * heading: 'Models',\n * subheading: 'Management',\n * namespace: 'io',\n * ignoreCI: true\n * }\n */\nasync function removeModel(url: string): Promise {\n const schemeAndPath = parseURL(url);\n const manager = ModelStoreManagerRegistry.getManager(schemeAndPath.scheme);\n return manager.removeModel(schemeAndPath.path);\n}\n\n/**\n * Copy a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Copying within a storage medium, e.g.,\n * `tf.io.copyModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Copying between two storage mediums, e.g.,\n * `tf.io.copyModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n * {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Copy the model, from Local Storage to IndexedDB.\n * await tf.io.copyModel(\n * 'localstorage://demo/management/model1',\n * 'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove both models.\n * await tf.io.removeModel('localstorage://demo/management/model1');\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of copying.\n * @param destURL Destination URL of copying.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n * is successful).\n * @throws Error if copying fails, e.g., if no model exists at `sourceURL`, or\n * if `oldPath` and `newPath` are identical.\n *\n * @doc {\n * heading: 'Models',\n * subheading: 'Management',\n * namespace: 'io',\n * ignoreCI: true\n * }\n */\nasync function copyModel(\n sourceURL: string, destURL: string): Promise {\n const deleteSource = false;\n return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\n/**\n * Move a model from one URL to another.\n *\n * This function supports:\n *\n * 1. Moving within a storage medium, e.g.,\n * `tf.io.moveModel('localstorage://model-1', 'localstorage://model-2')`\n * 2. Moving between two storage mediums, e.g.,\n * `tf.io.moveModel('localstorage://model-1', 'indexeddb://model-1')`\n *\n * ```js\n * // First create and save a model.\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n * {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * await model.save('localstorage://demo/management/model1');\n *\n * // Then list existing models.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Move the model, from Local Storage to IndexedDB.\n * await tf.io.moveModel(\n * 'localstorage://demo/management/model1',\n * 'indexeddb://demo/management/model1');\n *\n * // List models again.\n * console.log(JSON.stringify(await tf.io.listModels()));\n *\n * // Remove the moved model.\n * await tf.io.removeModel('indexeddb://demo/management/model1');\n * ```\n *\n * @param sourceURL Source URL of moving.\n * @param destURL Destination URL of moving.\n * @returns ModelArtifactsInfo of the copied model (if and only if copying\n * is successful).\n * @throws Error if moving fails, e.g., if no model exists at `sourceURL`, or\n * if `oldPath` and `newPath` are identical.\n *\n * @doc {\n * heading: 'Models',\n * subheading: 'Management',\n * namespace: 'io',\n * ignoreCI: true\n * }\n */\nasync function moveModel(\n sourceURL: string, destURL: string): Promise {\n const deleteSource = true;\n return cloneModelInternal(sourceURL, destURL, deleteSource);\n}\n\nexport {moveModel, copyModel, removeModel, listModels};\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport '../flags';\n\nimport {env} from '../environment';\nimport {BrowserIndexedDB, BrowserIndexedDBManager} from '../io/indexed_db';\nimport {BrowserLocalStorage, BrowserLocalStorageManager} from '../io/local_storage';\nimport {ModelStoreManagerRegistry} from '../io/model_management';\n\nimport {Platform} from './platform';\n\nexport class PlatformBrowser implements Platform {\n // According to the spec, the built-in encoder can do only UTF-8 encoding.\n // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n private textEncoder: TextEncoder;\n\n fetch(path: string, init?: RequestInit): Promise {\n return fetch(path, init);\n }\n\n now(): number {\n return performance.now();\n }\n\n encode(text: string, encoding: string): Uint8Array {\n if (encoding !== 'utf-8' && encoding !== 'utf8') {\n throw new Error(\n `Browser's encoder only supports utf-8, but got ${encoding}`);\n }\n if (this.textEncoder == null) {\n this.textEncoder = new TextEncoder();\n }\n return this.textEncoder.encode(text);\n }\n decode(bytes: Uint8Array, encoding: string): string {\n return new TextDecoder(encoding).decode(bytes);\n }\n}\n\nif (env().get('IS_BROWSER')) {\n env().setPlatform('browser', new PlatformBrowser());\n\n // Register LocalStorage IOHandler\n try {\n ModelStoreManagerRegistry.registerManager(\n BrowserLocalStorage.URL_SCHEME, new BrowserLocalStorageManager());\n } catch (err) {\n }\n\n // Register IndexedDB IOHandler\n try {\n ModelStoreManagerRegistry.registerManager(\n BrowserIndexedDB.URL_SCHEME, new BrowserIndexedDBManager());\n } catch (err) {\n }\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {env} from '../environment';\n\nimport {Platform} from './platform';\n\n// We are wrapping this within an object so it can be stubbed by Jasmine.\nexport const getNodeFetch = {\n // tslint:disable-next-line:no-require-imports\n importFetch: () => require('node-fetch')\n};\n\ntype FetchFn = (url: string, init?: RequestInit) => Promise;\nlet systemFetch: FetchFn;\n// These getters and setters are for testing so we don't export a mutable\n// variable.\nexport function resetSystemFetch() {\n systemFetch = null;\n}\nexport function setSystemFetch(fetchFn: FetchFn) {\n systemFetch = fetchFn;\n}\nexport function getSystemFetch(): FetchFn {\n return systemFetch;\n}\n\nexport class PlatformNode implements Platform {\n private textEncoder: TextEncoder;\n // tslint:disable-next-line:no-any\n util: any;\n\n constructor() {\n // tslint:disable-next-line:no-require-imports\n this.util = require('util');\n // According to the spec, the built-in encoder can do only UTF-8 encoding.\n // https://developer.mozilla.org/en-US/docs/Web/API/TextEncoder/TextEncoder\n this.textEncoder = new this.util.TextEncoder();\n }\n\n fetch(path: string, requestInits?: RequestInit): Promise {\n if (env().global.fetch != null) {\n return env().global.fetch(path, requestInits);\n }\n\n if (systemFetch == null) {\n systemFetch = getNodeFetch.importFetch();\n }\n return systemFetch(path, requestInits);\n }\n\n now(): number {\n const time = process.hrtime();\n return time[0] * 1000 + time[1] / 1000000;\n }\n\n encode(text: string, encoding: string): Uint8Array {\n if (encoding !== 'utf-8' && encoding !== 'utf8') {\n throw new Error(\n `Node built-in encoder only supports utf-8, but got ${encoding}`);\n }\n return this.textEncoder.encode(text);\n }\n decode(bytes: Uint8Array, encoding: string): string {\n if (bytes.length === 0) {\n return '';\n }\n return new this.util.TextDecoder(encoding).decode(bytes);\n }\n}\n\nif (env().get('IS_NODE')) {\n env().setPlatform('node', new PlatformNode());\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {TensorBuffer} from '../tensor';\nimport {DataType, DataTypeMap, Rank, ShapeMap} from '../types';\nimport * as util from '../util';\n\n/**\n * Creates an empty `tf.TensorBuffer` with the specified `shape` and `dtype`.\n *\n * The values are stored in CPU as `TypedArray`. Fill the buffer using\n * `buffer.set()`, or by modifying directly `buffer.values`.\n *\n * When done, call `buffer.toTensor()` to get an immutable `tf.Tensor` with\n * those values.\n *\n * ```js\n * // Create a buffer and set values at particular indices.\n * const buffer = tf.buffer([2, 2]);\n * buffer.set(3, 0, 0);\n * buffer.set(5, 1, 0);\n *\n * // Convert the buffer back to a tensor.\n * buffer.toTensor().print();\n * ```\n *\n * @param shape An array of integers defining the output tensor shape.\n * @param dtype The dtype of the buffer. Defaults to 'float32'.\n * @param values The values of the buffer as `TypedArray`. Defaults to\n * zeros.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function buffer(\n shape: ShapeMap[R], dtype: D = 'float32' as D,\n values?: DataTypeMap[D]): TensorBuffer {\n dtype = dtype || 'float32' as D;\n util.assertNonNegativeIntegerDimensions(shape);\n return new TensorBuffer(shape, dtype, values);\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE} from '../engine';\nimport {Cast, CastAttrs, CastInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {DataType, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Casts a `tf.Tensor` to a new dtype.\n *\n * ```js\n * const x = tf.tensor1d([1.5, 2.5, 3]);\n * tf.cast(x, 'int32').print();\n * ```\n * @param x The input tensor to be casted.\n * @param dtype The dtype to cast the input tensor to.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction cast_(x: T|TensorLike, dtype: DataType): T {\n const $x = convertToTensor(x, 'x', 'cast');\n\n // Sanity checks.\n if (!util.isValidDtype(dtype)) {\n throw new Error(`Failed to cast to unknown dtype ${dtype}`);\n }\n if (dtype === 'string' && $x.dtype !== 'string' ||\n dtype !== 'string' && $x.dtype === 'string') {\n throw new Error('Only strings can be casted to strings');\n }\n\n const inputs: CastInputs = {x: $x};\n const attrs: CastAttrs = {dtype};\n\n return ENGINE.runKernelFunc(\n backend => backend.cast($x, dtype), inputs as {} as NamedTensorMap,\n null /* grad */, Cast, attrs as {} as NamedAttrMap);\n}\n\nexport const cast = op({cast_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Identity, IdentityInputs} from '../kernel_names';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\n\n/**\n * Creates a new tensor with the same values and shape as the specified\n * tensor.\n *\n * ```js\n * const x = tf.tensor([1, 2]);\n *\n * x.clone().print();\n * ```\n *\n * @param x The tensor to clone.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction clone_(x: T|TensorLike): T {\n const $x = convertToTensor(x, 'x', 'clone', null);\n const forward = () =>\n ENGINE.makeTensorFromDataId($x.dataId, $x.shape, $x.dtype) as T;\n\n const inputs: IdentityInputs = {x: $x};\n\n // Note this op is called tf.identity in python. Hence the kernel name used\n // here.\n return ENGINE.runKernelFunc(\n forward, inputs as {} as NamedTensorMap, null /* grad */, Identity);\n}\n\nexport const clone = op({clone_});\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor} from '../tensor';\n\n/**\n * Prints information about the `tf.Tensor` including its data.\n *\n * ```js\n * const verbose = true;\n * tf.tensor2d([1, 2, 3, 4], [2, 2]).print(verbose);\n * ```\n * @param x The tensor to be printed.\n * @param verbose Whether to print verbose information about the ` Tensor`,\n * including dtype and size.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function print(x: T, verbose = false): void {\n console.log(x.toString(verbose));\n}\n", "/**\n * @license\n * Copyright 2020 Google Inc. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Required side effectful code for tfjs-core\n\n// Set up Engine and ENV\nimport {getOrMakeEngine} from './engine';\ngetOrMakeEngine();\n\n// Register backend-agnostic flags.\nimport './flags';\n// Register platforms\nimport './platforms/platform_browser';\nimport './platforms/platform_node';\n\n// Set up OpHandler\nimport {buffer} from './ops/buffer';\nimport {cast} from './ops/cast';\nimport {clone} from './ops/clone';\nimport {print} from './ops/print';\nimport {OpHandler, setOpHandler} from './tensor';\nconst opHandler: OpHandler = {\n buffer,\n cast,\n clone,\n print\n};\nsetOpHandler(opHandler);\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers related to files, such as browser-triggered file downloads,\n * user-selected files in browser.\n */\n\nimport '../flags';\nimport {env} from '../environment';\n\nimport {basename, concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, ModelArtifacts, ModelJSON, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\nconst DEFAULT_FILE_NAME_PREFIX = 'model';\nconst DEFAULT_JSON_EXTENSION_NAME = '.json';\nconst DEFAULT_WEIGHT_DATA_EXTENSION_NAME = '.weights.bin';\n\nfunction defer(f: () => T): Promise {\n return new Promise(resolve => setTimeout(resolve)).then(f);\n}\n\nexport class BrowserDownloads implements IOHandler {\n private readonly modelTopologyFileName: string;\n private readonly weightDataFileName: string;\n private readonly jsonAnchor: HTMLAnchorElement;\n private readonly weightDataAnchor: HTMLAnchorElement;\n\n static readonly URL_SCHEME = 'downloads://';\n\n constructor(fileNamePrefix?: string) {\n if (!env().getBool('IS_BROWSER')) {\n // TODO(cais): Provide info on what IOHandlers are available under the\n // current environment.\n throw new Error(\n 'browserDownloads() cannot proceed because the current environment ' +\n 'is not a browser.');\n }\n\n if (fileNamePrefix.startsWith(BrowserDownloads.URL_SCHEME)) {\n fileNamePrefix = fileNamePrefix.slice(BrowserDownloads.URL_SCHEME.length);\n }\n if (fileNamePrefix == null || fileNamePrefix.length === 0) {\n fileNamePrefix = DEFAULT_FILE_NAME_PREFIX;\n }\n\n this.modelTopologyFileName = fileNamePrefix + DEFAULT_JSON_EXTENSION_NAME;\n this.weightDataFileName =\n fileNamePrefix + DEFAULT_WEIGHT_DATA_EXTENSION_NAME;\n }\n\n async save(modelArtifacts: ModelArtifacts): Promise {\n if (typeof (document) === 'undefined') {\n throw new Error(\n 'Browser downloads are not supported in ' +\n 'this environment since `document` is not present');\n }\n const weightsURL = window.URL.createObjectURL(new Blob(\n [modelArtifacts.weightData], {type: 'application/octet-stream'}));\n\n if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n throw new Error(\n 'BrowserDownloads.save() does not support saving model topology ' +\n 'in binary formats yet.');\n } else {\n const weightsManifest: WeightsManifestConfig = [{\n paths: ['./' + this.weightDataFileName],\n weights: modelArtifacts.weightSpecs\n }];\n const modelTopologyAndWeightManifest: ModelJSON = {\n modelTopology: modelArtifacts.modelTopology,\n format: modelArtifacts.format,\n generatedBy: modelArtifacts.generatedBy,\n convertedBy: modelArtifacts.convertedBy,\n weightsManifest\n };\n const modelTopologyAndWeightManifestURL =\n window.URL.createObjectURL(new Blob(\n [JSON.stringify(modelTopologyAndWeightManifest)],\n {type: 'application/json'}));\n\n // If anchor elements are not provided, create them without attaching them\n // to parents, so that the downloaded file names can be controlled.\n const jsonAnchor = this.jsonAnchor == null ? document.createElement('a') :\n this.jsonAnchor;\n jsonAnchor.download = this.modelTopologyFileName;\n jsonAnchor.href = modelTopologyAndWeightManifestURL;\n // Trigger downloads by evoking a click event on the download anchors.\n // When multiple downloads are started synchronously, Firefox will only\n // save the last one.\n await defer(() => jsonAnchor.dispatchEvent(new MouseEvent('click')));\n\n if (modelArtifacts.weightData != null) {\n const weightDataAnchor = this.weightDataAnchor == null ?\n document.createElement('a') :\n this.weightDataAnchor;\n weightDataAnchor.download = this.weightDataFileName;\n weightDataAnchor.href = weightsURL;\n await defer(\n () => weightDataAnchor.dispatchEvent(new MouseEvent('click')));\n }\n\n return {modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts)};\n }\n }\n}\n\nclass BrowserFiles implements IOHandler {\n private readonly files: File[];\n\n constructor(files: File[]) {\n if (files == null || files.length < 1) {\n throw new Error(\n `When calling browserFiles, at least 1 file is required, ` +\n `but received ${files}`);\n }\n this.files = files;\n }\n\n async load(): Promise {\n const jsonFile = this.files[0];\n const weightFiles = this.files.slice(1);\n\n return new Promise((resolve, reject) => {\n const jsonReader = new FileReader();\n jsonReader.onload = (event: Event) => {\n // tslint:disable-next-line:no-any\n const modelJSON = JSON.parse((event.target as any).result) as ModelJSON;\n const modelTopology = modelJSON.modelTopology;\n if (modelTopology == null) {\n reject(new Error(\n `modelTopology field is missing from file ${jsonFile.name}`));\n return;\n }\n\n if (weightFiles.length === 0) {\n resolve({modelTopology});\n }\n\n const weightsManifest = modelJSON.weightsManifest;\n if (weightsManifest == null) {\n reject(new Error(\n `weightManifest field is missing from file ${jsonFile.name}`));\n return;\n }\n\n let pathToFile: {[path: string]: File};\n try {\n pathToFile =\n this.checkManifestAndWeightFiles(weightsManifest, weightFiles);\n } catch (err) {\n reject(err);\n return;\n }\n\n const weightSpecs: WeightsManifestEntry[] = [];\n const paths: string[] = [];\n const perFileBuffers: ArrayBuffer[] = [];\n weightsManifest.forEach(weightsGroup => {\n weightsGroup.paths.forEach(path => {\n paths.push(path);\n perFileBuffers.push(null);\n });\n weightSpecs.push(...weightsGroup.weights);\n });\n\n weightsManifest.forEach(weightsGroup => {\n weightsGroup.paths.forEach(path => {\n const weightFileReader = new FileReader();\n weightFileReader.onload = (event: Event) => {\n // tslint:disable-next-line:no-any\n const weightData = (event.target as any).result as ArrayBuffer;\n const index = paths.indexOf(path);\n perFileBuffers[index] = weightData;\n if (perFileBuffers.indexOf(null) === -1) {\n resolve({\n modelTopology,\n weightSpecs,\n weightData: concatenateArrayBuffers(perFileBuffers),\n format: modelJSON.format,\n generatedBy: modelJSON.generatedBy,\n convertedBy: modelJSON.convertedBy,\n userDefinedMetadata: modelJSON.userDefinedMetadata\n });\n }\n };\n weightFileReader.onerror = error =>\n reject(`Failed to weights data from file of path '${path}'.`);\n weightFileReader.readAsArrayBuffer(pathToFile[path]);\n });\n });\n };\n jsonReader.onerror = error => reject(\n `Failed to read model topology and weights manifest JSON ` +\n `from file '${jsonFile.name}'. BrowserFiles supports loading ` +\n `Keras-style tf.Model artifacts only.`);\n jsonReader.readAsText(jsonFile);\n });\n }\n\n /**\n * Check the compatibility between weights manifest and weight files.\n */\n private checkManifestAndWeightFiles(\n manifest: WeightsManifestConfig, files: File[]): {[path: string]: File} {\n const basenames: string[] = [];\n const fileNames = files.map(file => basename(file.name));\n const pathToFile: {[path: string]: File} = {};\n for (const group of manifest) {\n group.paths.forEach(path => {\n const pathBasename = basename(path);\n if (basenames.indexOf(pathBasename) !== -1) {\n throw new Error(\n `Duplicate file basename found in weights manifest: ` +\n `'${pathBasename}'`);\n }\n basenames.push(pathBasename);\n if (fileNames.indexOf(pathBasename) === -1) {\n throw new Error(\n `Weight file with basename '${pathBasename}' is not provided.`);\n } else {\n pathToFile[path] = files[fileNames.indexOf(pathBasename)];\n }\n });\n }\n\n if (basenames.length !== files.length) {\n throw new Error(\n `Mismatch in the number of files in weights manifest ` +\n `(${basenames.length}) and the number of weight files provided ` +\n `(${files.length}).`);\n }\n return pathToFile;\n }\n}\n\nexport const browserDownloadsRouter: IORouter = (url: string|string[]) => {\n if (!env().getBool('IS_BROWSER')) {\n return null;\n } else {\n if (!Array.isArray(url) && url.startsWith(BrowserDownloads.URL_SCHEME)) {\n return browserDownloads(url.slice(BrowserDownloads.URL_SCHEME.length));\n } else {\n return null;\n }\n }\n};\nIORouterRegistry.registerSaveRouter(browserDownloadsRouter);\n\n/**\n * Creates an IOHandler that triggers file downloads from the browser.\n *\n * The returned `IOHandler` instance can be used as model exporting methods such\n * as `tf.Model.save` and supports only saving.\n *\n * ```js\n * const model = tf.sequential();\n * model.add(tf.layers.dense(\n * {units: 1, inputShape: [10], activation: 'sigmoid'}));\n * const saveResult = await model.save('downloads://mymodel');\n * // This will trigger downloading of two files:\n * // 'mymodel.json' and 'mymodel.weights.bin'.\n * console.log(saveResult);\n * ```\n *\n * @param fileNamePrefix Prefix name of the files to be downloaded. For use with\n * `tf.Model`, `fileNamePrefix` should follow either of the following two\n * formats:\n * 1. `null` or `undefined`, in which case the default file\n * names will be used:\n * - 'model.json' for the JSON file containing the model topology and\n * weights manifest.\n * - 'model.weights.bin' for the binary file containing the binary weight\n * values.\n * 2. A single string or an Array of a single string, as the file name prefix.\n * For example, if `'foo'` is provided, the downloaded JSON\n * file and binary weights file will be named 'foo.json' and\n * 'foo.weights.bin', respectively.\n * @param config Additional configuration for triggering downloads.\n * @returns An instance of `BrowserDownloads` `IOHandler`.\n *\n * @doc {\n * heading: 'Models',\n * subheading: 'Loading',\n * namespace: 'io',\n * ignoreCI: true\n * }\n */\nexport function browserDownloads(fileNamePrefix = 'model'): IOHandler {\n return new BrowserDownloads(fileNamePrefix);\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from user-selected files.\n *\n * This method can be used for loading from files such as user-selected files\n * in the browser.\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * // Note: This code snippet won't run properly without the actual file input\n * // elements in the HTML DOM.\n *\n * // Suppose there are two HTML file input (``)\n * // elements.\n * const uploadJSONInput = document.getElementById('upload-json');\n * const uploadWeightsInput = document.getElementById('upload-weights');\n * const model = await tf.loadLayersModel(tf.io.browserFiles(\n * [uploadJSONInput.files[0], uploadWeightsInput.files[0]]));\n * ```\n *\n * @param files `File`s to load from. Currently, this function supports only\n * loading from files that contain Keras-style models (i.e., `tf.Model`s), for\n * which an `Array` of `File`s is expected (in that order):\n * - A JSON file containing the model topology and weight manifest.\n * - Optionally, One or more binary files containing the binary weights.\n * These files must have names that match the paths in the `weightsManifest`\n * contained by the aforementioned JSON file, or errors will be thrown\n * during loading. These weights files have the same format as the ones\n * generated by `tensorflowjs_converter` that comes with the `tensorflowjs`\n * Python PIP package. If no weights files are provided, only the model\n * topology will be loaded from the JSON file above.\n * @returns An instance of `Files` `IOHandler`.\n *\n * @doc {\n * heading: 'Models',\n * subheading: 'Loading',\n * namespace: 'io',\n * ignoreCI: true\n * }\n */\nexport function browserFiles(files: File[]): IOHandler {\n return new BrowserFiles(files);\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {assert} from '../util';\n\nimport {OnProgressCallback} from './types';\n\n/**\n * Monitor Promise.all progress, fire onProgress callback function.\n *\n * @param promises Promise list going to be monitored\n * @param onProgress Callback function. Fired when a promise resolved.\n * @param startFraction Optional fraction start. Default to 0.\n * @param endFraction Optional fraction end. Default to 1.\n */\nexport function monitorPromisesProgress(\n promises: Array>, onProgress: OnProgressCallback,\n startFraction?: number, endFraction?: number) {\n checkPromises(promises);\n startFraction = startFraction == null ? 0 : startFraction;\n endFraction = endFraction == null ? 1 : endFraction;\n checkFraction(startFraction, endFraction);\n let resolvedPromise = 0;\n\n const registerMonitor = (promise: Promise<{}>) => {\n promise.then(value => {\n const fraction = startFraction +\n ++resolvedPromise / promises.length * (endFraction - startFraction);\n // pass fraction as parameter to callback function.\n onProgress(fraction);\n return value;\n });\n return promise;\n };\n\n function checkPromises(promises: Array>): void {\n assert(\n promises != null && Array.isArray(promises) && promises.length > 0,\n () => 'promises must be a none empty array');\n }\n\n function checkFraction(startFraction: number, endFraction: number): void {\n assert(\n startFraction >= 0 && startFraction <= 1,\n () => `Progress fraction must be in range [0, 1], but ` +\n `got startFraction ${startFraction}`);\n assert(\n endFraction >= 0 && endFraction <= 1,\n () => `Progress fraction must be in range [0, 1], but ` +\n `got endFraction ${endFraction}`);\n assert(\n endFraction >= startFraction,\n () => `startFraction must be no more than endFraction, but ` +\n `got startFraction ${startFraction} and endFraction ` +\n `${endFraction}`);\n }\n\n return Promise.all(promises.map(registerMonitor));\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {env} from '../environment';\n\nimport {NamedTensorMap} from '../tensor_types';\nimport * as util from '../util';\nimport {decodeWeights} from './io_utils';\nimport {monitorPromisesProgress} from './progress';\nimport {DTYPE_VALUE_SIZE_MAP, LoadOptions, WeightsManifestConfig, WeightsManifestEntry} from './types';\n\n/**\n * Reads binary weights data from a number of URLs.\n *\n * @param fetchURLs URLs to send the HTTP requests at, using `fetch` calls.\n * @param requestOptions RequestInit (options) for the HTTP requests.\n * @param fetchFunc Optional overriding value for the `window.fetch` function.\n * @param onProgress Optional, progress callback function, fired periodically\n * before the load is completed.\n * @returns A `Promise` of an Array of `ArrayBuffer`. The Array has the same\n * length as `fetchURLs`.\n */\nexport async function loadWeightsAsArrayBuffer(\n fetchURLs: string[], loadOptions?: LoadOptions): Promise {\n if (loadOptions == null) {\n loadOptions = {};\n }\n\n const fetchFunc = loadOptions.fetchFunc == null ? env().platform.fetch :\n loadOptions.fetchFunc;\n\n // Create the requests for all of the weights in parallel.\n const requests = fetchURLs.map(\n fetchURL =>\n fetchFunc(fetchURL, loadOptions.requestInit, {isBinary: true}));\n\n const fetchStartFraction = 0;\n const fetchEndFraction = 0.5;\n\n const responses = loadOptions.onProgress == null ?\n await Promise.all(requests) :\n await monitorPromisesProgress(\n requests, loadOptions.onProgress, fetchStartFraction,\n fetchEndFraction);\n\n const bufferPromises = responses.map(response => response.arrayBuffer());\n\n const bufferStartFraction = 0.5;\n const bufferEndFraction = 1;\n\n const buffers = loadOptions.onProgress == null ?\n await Promise.all(bufferPromises) :\n await monitorPromisesProgress(\n bufferPromises, loadOptions.onProgress, bufferStartFraction,\n bufferEndFraction);\n return buffers;\n}\n\n/**\n * Reads a weights manifest JSON configuration, fetches the weights and\n * returns them as `Tensor`s.\n *\n * @param manifest The weights manifest JSON.\n * @param filePathPrefix The path prefix for filenames given in the manifest.\n * Defaults to the empty string.\n * @param weightNames The names of the weights to be fetched.\n */\nexport async function loadWeights(\n manifest: WeightsManifestConfig, filePathPrefix = '',\n weightNames?: string[],\n requestInit?: RequestInit): Promise {\n // TODO(nsthorat): Groups are currently fetched atomically. If you need a\n // single weight from a group, the whole group will be fetched. At a future\n // date, we should support fetching only the individual shards within a\n // group that are needed to reconstruct the requested weight.\n // TODO(cais): Use `decodeWeights` for implementation.\n\n const fetchWeights = (fetchUrls: string[]) =>\n loadWeightsAsArrayBuffer(fetchUrls, {requestInit});\n const loadWeights = weightsLoaderFactory(fetchWeights);\n\n return loadWeights(manifest, filePathPrefix, weightNames);\n}\n\n/**\n * Creates a function, which reads a weights manifest JSON configuration,\n * fetches the weight files using the specified function and returns them as\n * `Tensor`s.\n *\n * ```js\n * // example for creating a nodejs weight loader, which reads the weight files\n * // from disk using fs.readFileSync\n *\n * import * as fs from 'fs'\n *\n * const fetchWeightsFromDisk = (filePaths: string[]) =>\n * filePaths.map(filePath => fs.readFileSync(filePath).buffer)\n *\n * const loadWeights = tf.io.weightsLoaderFactory(fetchWeightsFromDisk)\n *\n * const manifest = JSON.parse(\n * fs.readFileSync('./my_model-weights_manifest').toString()\n * )\n * const weightMap = await loadWeights(manifest, './')\n * ```\n * @param fetchWeightsFunction The function used for fetching the weight files.\n * @returns Weight loading function.\n */\nexport function weightsLoaderFactory(\n fetchWeightsFunction: (fetchUrls: string[]) => Promise):\n (manifest: WeightsManifestConfig, filePathPrefix?: string,\n weightNames?: string[]) => Promise {\n return async(\n manifest: WeightsManifestConfig, filePathPrefix = '',\n weightNames?: string[]): Promise => {\n // Collect all the groups, weights, and their relative offsets to be\n // fetched.\n const groupIndicesToFetchMap = manifest.map(() => false);\n const groupWeightsToFetch: {\n [group: number]: Array<{\n manifestEntry: WeightsManifestEntry; groupOffset: number;\n sizeBytes: number;\n }>\n } = {};\n const weightsFound =\n weightNames != null ? weightNames.map(() => false) : [];\n const allManifestWeightNames: string[] = [];\n manifest.forEach((manifestGroupConfig, groupIndex) => {\n let groupOffset = 0;\n manifestGroupConfig.weights.forEach(weightsEntry => {\n const rawDtype = ('quantization' in weightsEntry) ?\n weightsEntry.quantization.dtype :\n weightsEntry.dtype;\n\n const weightsBytes = DTYPE_VALUE_SIZE_MAP[rawDtype] *\n util.sizeFromShape(weightsEntry.shape);\n\n const enqueueWeightsForFetchingFn = () => {\n groupIndicesToFetchMap[groupIndex] = true;\n if (groupWeightsToFetch[groupIndex] == null) {\n groupWeightsToFetch[groupIndex] = [];\n }\n\n groupWeightsToFetch[groupIndex].push({\n manifestEntry: weightsEntry,\n groupOffset,\n sizeBytes: weightsBytes\n });\n };\n\n if (weightNames != null) {\n weightNames.forEach((weightName, weightIndex) => {\n if (weightName === weightsEntry.name) {\n enqueueWeightsForFetchingFn();\n weightsFound[weightIndex] = true;\n }\n });\n } else {\n enqueueWeightsForFetchingFn();\n }\n\n allManifestWeightNames.push(weightsEntry.name);\n groupOffset += weightsBytes;\n });\n });\n\n if (!weightsFound.every(found => found)) {\n const weightsNotFound = weightNames.filter((_, i) => !weightsFound[i]);\n throw new Error(\n `Could not find weights in manifest with names: ` +\n `${weightsNotFound.join(', ')}. \\n` +\n `Manifest JSON has weights with names: ` +\n `${allManifestWeightNames.join(', ')}.`);\n }\n\n // Convert the one-hot boolean groupId => shouldFetch map to a list of group\n // IDs.\n const groupIndicesToFetch =\n groupIndicesToFetchMap.reduce((accumulator, shouldFetch, i) => {\n if (shouldFetch) {\n accumulator.push(i);\n }\n return accumulator;\n }, []);\n\n const fetchUrls: string[] = [];\n groupIndicesToFetch.forEach(i => {\n manifest[i].paths.forEach(filepath => {\n const fetchUrl = filePathPrefix +\n (!filePathPrefix.endsWith('/') ? '/' : '') + filepath;\n fetchUrls.push(fetchUrl);\n });\n });\n const buffers = await fetchWeightsFunction(fetchUrls);\n\n const weightsTensorMap: NamedTensorMap = {};\n let bufferIndexOffset = 0;\n groupIndicesToFetch.forEach(i => {\n const numBuffers = manifest[i].paths.length;\n\n let groupBytes = 0;\n for (let i = 0; i < numBuffers; i++) {\n groupBytes += buffers[bufferIndexOffset + i].byteLength;\n }\n\n // Create a buffer for the whole group.\n const groupBuffer = new ArrayBuffer(groupBytes);\n const groupByteBuffer = new Uint8Array(groupBuffer);\n let groupBufferOffset = 0;\n for (let i = 0; i < numBuffers; i++) {\n const buffer = new Uint8Array(buffers[bufferIndexOffset + i]);\n groupByteBuffer.set(buffer, groupBufferOffset);\n groupBufferOffset += buffer.byteLength;\n }\n\n const weightsEntries = groupWeightsToFetch[i];\n weightsEntries.forEach(weightsEntry => {\n const byteBuffer = groupBuffer.slice(\n weightsEntry.groupOffset,\n weightsEntry.groupOffset + weightsEntry.sizeBytes);\n const nameToTensorMap =\n decodeWeights(byteBuffer, [weightsEntry.manifestEntry]);\n for (const name in nameToTensorMap) {\n weightsTensorMap[name] = nameToTensorMap[name];\n }\n });\n\n bufferIndexOffset += numBuffers;\n });\n\n return weightsTensorMap;\n };\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandler implementations based on HTTP requests in the web browser.\n *\n * Uses [`fetch`](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API).\n */\n\nimport {env} from '../environment';\n\nimport {assert} from '../util';\nimport {concatenateArrayBuffers, getModelArtifactsInfoForJSON} from './io_utils';\nimport {IORouter, IORouterRegistry} from './router_registry';\nimport {IOHandler, LoadOptions, ModelArtifacts, ModelJSON, OnProgressCallback, SaveResult, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeightsAsArrayBuffer} from './weights_loader';\n\nconst OCTET_STREAM_MIME_TYPE = 'application/octet-stream';\nconst JSON_TYPE = 'application/json';\nexport class HTTPRequest implements IOHandler {\n protected readonly path: string;\n protected readonly requestInit: RequestInit;\n\n private readonly fetch: Function;\n private readonly weightUrlConverter: (weightName: string) => Promise;\n\n readonly DEFAULT_METHOD = 'POST';\n\n static readonly URL_SCHEME_REGEX = /^https?:\\/\\//;\n\n private readonly weightPathPrefix: string;\n private readonly onProgress: OnProgressCallback;\n\n constructor(path: string, loadOptions?: LoadOptions) {\n if (loadOptions == null) {\n loadOptions = {};\n }\n this.weightPathPrefix = loadOptions.weightPathPrefix;\n this.onProgress = loadOptions.onProgress;\n this.weightUrlConverter = loadOptions.weightUrlConverter;\n\n if (loadOptions.fetchFunc != null) {\n assert(\n typeof loadOptions.fetchFunc === 'function',\n () => 'Must pass a function that matches the signature of ' +\n '`fetch` (see ' +\n 'https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)');\n this.fetch = loadOptions.fetchFunc;\n } else {\n this.fetch = env().platform.fetch;\n }\n\n assert(\n path != null && path.length > 0,\n () => 'URL path for http must not be null, undefined or ' +\n 'empty.');\n\n if (Array.isArray(path)) {\n assert(\n path.length === 2,\n () => 'URL paths for http must have a length of 2, ' +\n `(actual length is ${path.length}).`);\n }\n this.path = path;\n\n if (loadOptions.requestInit != null &&\n loadOptions.requestInit.body != null) {\n throw new Error(\n 'requestInit is expected to have no pre-existing body, but has one.');\n }\n this.requestInit = loadOptions.requestInit || {};\n }\n\n async save(modelArtifacts: ModelArtifacts): Promise {\n if (modelArtifacts.modelTopology instanceof ArrayBuffer) {\n throw new Error(\n 'BrowserHTTPRequest.save() does not support saving model topology ' +\n 'in binary formats yet.');\n }\n\n const init = Object.assign({method: this.DEFAULT_METHOD}, this.requestInit);\n init.body = new FormData();\n\n const weightsManifest: WeightsManifestConfig = [{\n paths: ['./model.weights.bin'],\n weights: modelArtifacts.weightSpecs,\n }];\n const modelTopologyAndWeightManifest: ModelJSON = {\n modelTopology: modelArtifacts.modelTopology,\n format: modelArtifacts.format,\n generatedBy: modelArtifacts.generatedBy,\n convertedBy: modelArtifacts.convertedBy,\n userDefinedMetadata: modelArtifacts.userDefinedMetadata,\n weightsManifest\n };\n\n init.body.append(\n 'model.json',\n new Blob(\n [JSON.stringify(modelTopologyAndWeightManifest)],\n {type: JSON_TYPE}),\n 'model.json');\n\n if (modelArtifacts.weightData != null) {\n init.body.append(\n 'model.weights.bin',\n new Blob([modelArtifacts.weightData], {type: OCTET_STREAM_MIME_TYPE}),\n 'model.weights.bin');\n }\n\n const response = await this.fetch(this.path, init);\n\n if (response.ok) {\n return {\n modelArtifactsInfo: getModelArtifactsInfoForJSON(modelArtifacts),\n responses: [response],\n };\n } else {\n throw new Error(\n `BrowserHTTPRequest.save() failed due to HTTP response status ` +\n `${response.status}.`);\n }\n }\n\n /**\n * Load model artifacts via HTTP request(s).\n *\n * See the documentation to `tf.io.http` for details on the saved\n * artifacts.\n *\n * @returns The loaded model artifacts (if loading succeeds).\n */\n async load(): Promise {\n const modelConfigRequest = await this.fetch(this.path, this.requestInit);\n\n if (!modelConfigRequest.ok) {\n throw new Error(\n `Request to ${this.path} failed with status code ` +\n `${modelConfigRequest.status}. Please verify this URL points to ` +\n `the model JSON of the model to load.`);\n }\n let modelConfig: ModelJSON;\n try {\n modelConfig = await modelConfigRequest.json();\n } catch (e) {\n let message = `Failed to parse model JSON of response from ${this.path}.`;\n // TODO(nsthorat): Remove this after some time when we're comfortable that\n // .pb files are mostly gone.\n if (this.path.endsWith('.pb')) {\n message += ' Your path contains a .pb file extension. ' +\n 'Support for .pb models have been removed in TensorFlow.js 1.0 ' +\n 'in favor of .json models. You can re-convert your Python ' +\n 'TensorFlow model using the TensorFlow.js 1.0 conversion scripts ' +\n 'or you can convert your.pb models with the \\'pb2json\\'' +\n 'NPM script in the tensorflow/tfjs-converter repository.';\n } else {\n message += ' Please make sure the server is serving valid ' +\n 'JSON for this request.';\n }\n throw new Error(message);\n }\n const modelTopology = modelConfig.modelTopology;\n const weightsManifest = modelConfig.weightsManifest;\n const generatedBy = modelConfig.generatedBy;\n const convertedBy = modelConfig.convertedBy;\n const format = modelConfig.format;\n const userDefinedMetadata = modelConfig.userDefinedMetadata;\n\n // We do not allow both modelTopology and weightsManifest to be missing.\n if (modelTopology == null && weightsManifest == null) {\n throw new Error(\n `The JSON from HTTP path ${this.path} contains neither model ` +\n `topology or manifest for weights.`);\n }\n\n let weightSpecs: WeightsManifestEntry[];\n let weightData: ArrayBuffer;\n if (weightsManifest != null) {\n const results = await this.loadWeights(weightsManifest);\n [weightSpecs, weightData] = results;\n }\n\n const artifacts: ModelArtifacts = {\n modelTopology,\n weightSpecs,\n weightData,\n userDefinedMetadata,\n generatedBy,\n convertedBy,\n format\n };\n\n const initializer = modelConfig.modelInitializer;\n if (initializer) {\n artifacts.modelInitializer = initializer;\n }\n\n return artifacts;\n }\n\n private async loadWeights(weightsManifest: WeightsManifestConfig):\n Promise<[WeightsManifestEntry[], ArrayBuffer]> {\n const weightPath = Array.isArray(this.path) ? this.path[1] : this.path;\n const [prefix, suffix] = parseUrl(weightPath);\n const pathPrefix = this.weightPathPrefix || prefix;\n\n const weightSpecs = [];\n for (const entry of weightsManifest) {\n weightSpecs.push(...entry.weights);\n }\n\n const fetchURLs: string[] = [];\n const urlPromises: Array> = [];\n for (const weightsGroup of weightsManifest) {\n for (const path of weightsGroup.paths) {\n if (this.weightUrlConverter != null) {\n urlPromises.push(this.weightUrlConverter(path));\n } else {\n fetchURLs.push(pathPrefix + path + suffix);\n }\n }\n }\n\n if (this.weightUrlConverter) {\n fetchURLs.push(...await Promise.all(urlPromises));\n }\n\n const buffers = await loadWeightsAsArrayBuffer(fetchURLs, {\n requestInit: this.requestInit,\n fetchFunc: this.fetch,\n onProgress: this.onProgress\n });\n return [weightSpecs, concatenateArrayBuffers(buffers)];\n }\n}\n\n/**\n * Extract the prefix and suffix of the url, where the prefix is the path before\n * the last file, and suffix is the search params after the last file.\n * ```\n * const url = 'http://tfhub.dev/model/1/tensorflowjs_model.pb?tfjs-format=file'\n * [prefix, suffix] = parseUrl(url)\n * // prefix = 'http://tfhub.dev/model/1/'\n * // suffix = '?tfjs-format=file'\n * ```\n * @param url the model url to be parsed.\n */\nexport function parseUrl(url: string): [string, string] {\n const lastSlash = url.lastIndexOf('/');\n const lastSearchParam = url.lastIndexOf('?');\n const prefix = url.substring(0, lastSlash);\n const suffix =\n lastSearchParam > lastSlash ? url.substring(lastSearchParam) : '';\n return [prefix + '/', suffix];\n}\n\nexport function isHTTPScheme(url: string): boolean {\n return url.match(HTTPRequest.URL_SCHEME_REGEX) != null;\n}\n\nexport const httpRouter: IORouter =\n (url: string, loadOptions?: LoadOptions) => {\n if (typeof fetch === 'undefined' &&\n (loadOptions == null || loadOptions.fetchFunc == null)) {\n // `http` uses `fetch` or `node-fetch`, if one wants to use it in\n // an environment that is not the browser or node they have to setup a\n // global fetch polyfill.\n return null;\n } else {\n let isHTTP = true;\n if (Array.isArray(url)) {\n isHTTP = url.every(urlItem => isHTTPScheme(urlItem));\n } else {\n isHTTP = isHTTPScheme(url);\n }\n if (isHTTP) {\n return http(url, loadOptions);\n }\n }\n return null;\n };\nIORouterRegistry.registerSaveRouter(httpRouter);\nIORouterRegistry.registerLoadRouter(httpRouter);\n\n/**\n * Creates an IOHandler subtype that sends model artifacts to HTTP server.\n *\n * An HTTP request of the `multipart/form-data` mime type will be sent to the\n * `path` URL. The form data includes artifacts that represent the topology\n * and/or weights of the model. In the case of Keras-style `tf.Model`, two\n * blobs (files) exist in form-data:\n * - A JSON file consisting of `modelTopology` and `weightsManifest`.\n * - A binary weights file consisting of the concatenated weight values.\n * These files are in the same format as the one generated by\n * [tfjs_converter](https://js.tensorflow.org/tutorials/import-keras.html).\n *\n * The following code snippet exemplifies the client-side code that uses this\n * function:\n *\n * ```js\n * const model = tf.sequential();\n * model.add(\n * tf.layers.dense({units: 1, inputShape: [100], activation: 'sigmoid'}));\n *\n * const saveResult = await model.save(tf.io.http(\n * 'http://model-server:5000/upload', {requestInit: {method: 'PUT'}}));\n * console.log(saveResult);\n * ```\n *\n * If the default `POST` method is to be used, without any custom parameters\n * such as headers, you can simply pass an HTTP or HTTPS URL to `model.save`:\n *\n * ```js\n * const saveResult = await model.save('http://model-server:5000/upload');\n * ```\n *\n * The following GitHub Gist\n * https://gist.github.com/dsmilkov/1b6046fd6132d7408d5257b0976f7864\n * implements a server based on [flask](https://github.com/pallets/flask) that\n * can receive the request. Upon receiving the model artifacts via the requst,\n * this particular server reconsistutes instances of [Keras\n * Models](https://keras.io/models/model/) in memory.\n *\n *\n * @param path A URL path to the model.\n * Can be an absolute HTTP path (e.g.,\n * 'http://localhost:8000/model-upload)') or a relative path (e.g.,\n * './model-upload').\n * @param requestInit Request configurations to be used when sending\n * HTTP request to server using `fetch`. It can contain fields such as\n * `method`, `credentials`, `headers`, `mode`, etc. See\n * https://developer.mozilla.org/en-US/docs/Web/API/Request/Request\n * for more information. `requestInit` must not have a body, because the\n * body will be set by TensorFlow.js. File blobs representing the model\n * topology (filename: 'model.json') and the weights of the model (filename:\n * 'model.weights.bin') will be appended to the body. If `requestInit` has a\n * `body`, an Error will be thrown.\n * @param loadOptions Optional configuration for the loading. It includes the\n * following fields:\n * - weightPathPrefix Optional, this specifies the path prefix for weight\n * files, by default this is calculated from the path param.\n * - fetchFunc Optional, custom `fetch` function. E.g., in Node.js,\n * the `fetch` from node-fetch can be used here.\n * - onProgress Optional, progress callback function, fired periodically\n * before the load is completed.\n * @returns An instance of `IOHandler`.\n *\n * @doc {\n * heading: 'Models',\n * subheading: 'Loading',\n * namespace: 'io',\n * ignoreCI: true\n * }\n */\nexport function http(path: string, loadOptions?: LoadOptions): IOHandler {\n return new HTTPRequest(path, loadOptions);\n}\n\n/**\n * Deprecated. Use `tf.io.http`.\n * @param path\n * @param loadOptions\n */\nexport function browserHTTPRequest(\n path: string, loadOptions?: LoadOptions): IOHandler {\n return http(path, loadOptions);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * IOHandlers that pass through the in-memory ModelArtifacts format.\n */\n\nimport {IOHandler, ModelArtifacts, SaveResult, TrainingConfig, WeightsManifestEntry} from './types';\n\nclass PassthroughLoader implements IOHandler {\n constructor(private readonly modelArtifacts?: ModelArtifacts) {}\n\n async load(): Promise {\n return this.modelArtifacts;\n }\n}\n\nclass PassthroughSaver implements IOHandler {\n constructor(\n private readonly saveHandler:\n (artifacts: ModelArtifacts) => Promise) {}\n\n async save(modelArtifacts: ModelArtifacts) {\n return this.saveHandler(modelArtifacts);\n }\n}\n\n/**\n * Creates an IOHandler that loads model artifacts from memory.\n *\n * When used in conjunction with `tf.loadLayersModel`, an instance of\n * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.\n *\n * ```js\n * const model = await tf.loadLayersModel(tf.io.fromMemory(\n * modelTopology, weightSpecs, weightData));\n * ```\n *\n * @param modelArtifacts a object containing model topology (i.e., parsed from\n * the JSON format).\n * @param weightSpecs An array of `WeightsManifestEntry` objects describing the\n * names, shapes, types, and quantization of the weight data.\n * @param weightData A single `ArrayBuffer` containing the weight data,\n * concatenated in the order described by the weightSpecs.\n * @param trainingConfig Model training configuration. Optional.\n *\n * @returns A passthrough `IOHandler` that simply loads the provided data.\n */\nexport function fromMemory(\n modelArtifacts: {}|ModelArtifacts, weightSpecs?: WeightsManifestEntry[],\n weightData?: ArrayBuffer, trainingConfig?: TrainingConfig): IOHandler {\n if (arguments.length === 1) {\n const isModelArtifacts =\n (modelArtifacts as ModelArtifacts).modelTopology != null ||\n (modelArtifacts as ModelArtifacts).weightSpecs != null;\n if (isModelArtifacts) {\n return new PassthroughLoader(modelArtifacts as ModelArtifacts);\n } else {\n // Legacy support: with only modelTopology.\n // TODO(cais): Remove this deprecated API.\n console.warn(\n 'Please call tf.io.fromMemory() with only one argument. ' +\n 'The argument should be of type ModelArtifacts. ' +\n 'The multi-argument signature of tf.io.fromMemory() has been ' +\n 'deprecated and will be removed in a future release.');\n return new PassthroughLoader({modelTopology: modelArtifacts as {}});\n }\n } else {\n // Legacy support.\n // TODO(cais): Remove this deprecated API.\n console.warn(\n 'Please call tf.io.fromMemory() with only one argument. ' +\n 'The argument should be of type ModelArtifacts. ' +\n 'The multi-argument signature of tf.io.fromMemory() has been ' +\n 'deprecated and will be removed in a future release.');\n return new PassthroughLoader({\n modelTopology: modelArtifacts as {},\n weightSpecs,\n weightData,\n trainingConfig\n });\n }\n}\n\n/**\n * Creates an IOHandler that passes saved model artifacts to a callback.\n *\n * ```js\n * function handleSave(artifacts) {\n * // ... do something with the artifacts ...\n * return {modelArtifactsInfo: {...}, ...};\n * }\n *\n * const saveResult = model.save(tf.io.withSaveHandler(handleSave));\n * ```\n *\n * @param saveHandler A function that accepts a `ModelArtifacts` and returns a\n * `SaveResult`.\n */\nexport function withSaveHandler(\n saveHandler: (artifacts: ModelArtifacts) =>\n Promise): IOHandler {\n return new PassthroughSaver(saveHandler);\n}\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n// Importing local_storage and indexed_db is necessary for the routers to be\n// registered.\nimport './indexed_db';\nimport './local_storage';\n\nimport {browserFiles} from './browser_files';\nimport {browserHTTPRequest, http, isHTTPScheme} from './http';\nimport {concatenateArrayBuffers, decodeWeights, encodeWeights, getModelArtifactsInfoForJSON} from './io_utils';\nimport {fromMemory, withSaveHandler} from './passthrough';\nimport {getLoadHandlers, getSaveHandlers, registerLoadRouter, registerSaveRouter} from './router_registry';\nimport {IOHandler, LoadHandler, LoadOptions, ModelArtifacts, ModelArtifactsInfo, ModelJSON, ModelStoreManager, OnProgressCallback, RequestDetails, SaveConfig, SaveHandler, SaveResult, WeightGroup, WeightsManifestConfig, WeightsManifestEntry} from './types';\nimport {loadWeights, weightsLoaderFactory} from './weights_loader';\n\nexport {copyModel, listModels, moveModel, removeModel} from './model_management';\nexport {\n browserFiles,\n browserHTTPRequest,\n concatenateArrayBuffers,\n decodeWeights,\n encodeWeights,\n fromMemory,\n getLoadHandlers,\n getModelArtifactsInfoForJSON,\n getSaveHandlers,\n http,\n IOHandler,\n isHTTPScheme,\n LoadHandler,\n LoadOptions,\n loadWeights,\n ModelArtifacts,\n ModelArtifactsInfo,\n ModelJSON,\n ModelStoreManager,\n OnProgressCallback,\n registerLoadRouter,\n registerSaveRouter,\n RequestDetails,\n SaveConfig,\n SaveHandler,\n SaveResult,\n WeightGroup,\n weightsLoaderFactory,\n WeightsManifestConfig,\n WeightsManifestEntry,\n withSaveHandler\n};\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {KernelBackend} from '../backends/backend';\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {Reshape, ReshapeAttrs, ReshapeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {GradSaveFunc, NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {Rank, ShapeMap, TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Reshapes a `tf.Tensor` to a given shape.\n *\n * Given an input tensor, returns a new tensor with the same values as the\n * input tensor with shape `shape`.\n *\n * If one component of shape is the special value -1, the size of that\n * dimension is computed so that the total size remains constant. In\n * particular, a shape of [-1] flattens into 1-D. At most one component of\n * shape can be -1.\n *\n * If shape is 1-D or higher, then the operation returns a tensor with shape\n * shape filled with the values of tensor. In this case, the number of\n * elements implied by shape must be the same as the number of elements in\n * tensor.\n *\n * ```js\n * const x = tf.tensor1d([1, 2, 3, 4]);\n * x.reshape([2, 2]).print();\n * ```\n *\n * @param x The input tensor to be reshaped.\n * @param shape An array of integers defining the output tensor shape.\n *\n * @doc {heading: 'Tensors', subheading: 'Transformations'}\n */\nfunction reshape_(\n x: Tensor|TensorLike, shape: ShapeMap[R]): Tensor {\n const $x = convertToTensor(x, 'x', 'reshape', null);\n\n const inputs: ReshapeInputs = {x: $x};\n const attrs: ReshapeAttrs = {shape};\n const forward: ForwardFunc<\n Tensor> = (backend: KernelBackend, save: GradSaveFunc) => {\n shape = util.inferFromImplicitShape(shape, $x.size) as ShapeMap[R];\n util.assert(\n $x.size === util.sizeFromShape(shape),\n () => 'new shape and old shape must have the same number of elements.');\n save([$x]);\n return backend.reshape($x, shape);\n };\n return ENGINE.runKernelFunc(\n forward, inputs as {} as NamedTensorMap, null /* grad */, Reshape,\n attrs as {} as NamedAttrMap);\n}\nexport const reshape = op({reshape_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {BatchMatMul, BatchMatMulAttrs, BatchMatMulInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {makeTypesMatch} from '../tensor_util';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Computes the dot product of two matrices, A * B. These must be matrices.\n *\n * ```js\n * const a = tf.tensor2d([1, 2], [1, 2]);\n * const b = tf.tensor2d([1, 2, 3, 4], [2, 2]);\n *\n * a.matMul(b).print(); // or tf.matMul(a, b)\n * ```\n * @param a First matrix in dot product operation.\n * @param b Second matrix in dot product operation.\n * @param transposeA If true, `a` is transposed before multiplication.\n * @param transposeB If true, `b` is transposed before multiplication.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction matMul_(\n a: T|TensorLike, b: T|TensorLike, transposeA = false,\n transposeB = false): T {\n let $a = convertToTensor(a, 'a', 'matMul');\n let $b = convertToTensor(b, 'b', 'matMul');\n [$a, $b] = makeTypesMatch($a, $b);\n\n util.assert(\n $a.rank >= 2 && $b.rank >= 2 && $a.rank === $b.rank,\n () => `Error in matMul: inputs must have the same rank of at least 2, ` +\n `got ranks ${$a.rank} and ${$b.rank}.`);\n\n const innerShapeA =\n transposeA ? $a.shape[$a.rank - 2] : $a.shape[$a.rank - 1];\n const innerShapeB =\n transposeB ? $b.shape[$b.rank - 1] : $b.shape[$b.rank - 2];\n\n const outerShapeA =\n transposeA ? $a.shape[$a.rank - 1] : $a.shape[$a.rank - 2];\n const outerShapeB =\n transposeB ? $b.shape[$b.rank - 2] : $b.shape[$b.rank - 1];\n\n const outerDimsA = $a.shape.slice(0, -2);\n const outerDimsB = $b.shape.slice(0, -2);\n const batchDimA = util.sizeFromShape(outerDimsA);\n const batchDimB = util.sizeFromShape(outerDimsB);\n\n util.assert(\n util.arraysEqual(outerDimsA, outerDimsB),\n () => `Error in matMul: outer dimensions (${outerDimsA}) and (` +\n `${outerDimsB}) of Tensors with shapes ${$a.shape} and ` +\n `${$b.shape} must match.`);\n\n util.assert(\n innerShapeA === innerShapeB,\n () => `Error in matMul: inner shapes (${innerShapeA}) and (` +\n `${innerShapeB}) of Tensors with shapes ${$a.shape} and ` +\n `${$b.shape} and transposeA=${transposeA}` +\n ` and transposeB=${transposeB} must match.`);\n\n const outShape = $a.shape.slice(0, -2).concat([outerShapeA, outerShapeB]);\n\n const a3D = transposeA ? reshape($a, [batchDimA, innerShapeA, outerShapeA]) :\n reshape($a, [batchDimA, outerShapeA, innerShapeA]);\n const b3D = transposeB ? reshape($b, [batchDimB, outerShapeB, innerShapeB]) :\n reshape($b, [batchDimB, innerShapeB, outerShapeB]);\n\n const forward: ForwardFunc = (backend, save) => {\n save([a3D, b3D]);\n\n return backend.batchMatMul(\n a3D as Tensor3D, b3D as Tensor3D, transposeA, transposeB);\n };\n\n const inputs: BatchMatMulInputs = {a: a3D, b: b3D};\n\n const attrs: BatchMatMulAttrs = {transposeA, transposeB};\n\n const res = ENGINE.runKernelFunc(\n forward, inputs as {} as NamedTensorMap, null /* grad */, BatchMatMul,\n attrs as {} as NamedAttrMap);\n\n return reshape(res, outShape) as T;\n}\n\nexport const matMul = op({matMul_});\n", "/**\n * @license\n * Copyright 2020 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE, ForwardFunc} from '../engine';\nimport {OneHot, OneHotAttrs, OneHotInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\n\nimport {op} from './operation';\nimport {reshape} from './reshape';\n\n/**\n * Creates a one-hot `tf.Tensor`. The locations represented by `indices` take\n * value `onValue` (defaults to 1), while all other locations take value\n * `offValue` (defaults to 0). If `indices` is rank `R`, the output has rank\n * `R+1` with the last axis of size `depth`.\n *\n * ```js\n * tf.oneHot(tf.tensor1d([0, 1], 'int32'), 3).print();\n * ```\n *\n * @param indices `tf.Tensor` of indices with dtype `int32`.\n * @param depth The depth of the one hot dimension.\n * @param onValue A number used to fill in the output when the index matches\n * the location.\n * @param offValue A number used to fill in the output when the index does\n * not match the location.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nfunction oneHot_(\n indices: Tensor|TensorLike, depth: number, onValue = 1,\n offValue = 0): Tensor {\n if (depth < 2) {\n throw new Error(`Error in oneHot: depth must be >=2, but it is ${depth}`);\n }\n const $indices = convertToTensor(indices, 'indices', 'oneHot', 'int32');\n const outShape = [...$indices.shape, depth];\n\n const forward: ForwardFunc = (backend, save) => {\n save([$indices]);\n return reshape(\n backend.oneHot(\n reshape($indices, [$indices.size]), depth, onValue, offValue),\n outShape);\n };\n\n const inputs: OneHotInputs = {indices: $indices};\n const attrs: OneHotAttrs = {depth, onValue, offValue};\n\n return ENGINE.runKernelFunc(\n forward, inputs as unknown as NamedTensorMap, null /* grad */, OneHot,\n attrs as unknown as NamedAttrMap);\n}\n\nexport const oneHot = op({oneHot_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {Transpose, TransposeAttrs, TransposeInputs} from '../kernel_names';\nimport {NamedAttrMap} from '../kernel_registry';\nimport {Tensor} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {op} from './operation';\n\n/**\n * Transposes the `tf.Tensor`. Permutes the dimensions according to `perm`.\n *\n * The returned `tf.Tensor`'s dimension `i` will correspond to the input\n * dimension `perm[i]`. If `perm` is not given, it is set to `[n-1...0]`,\n * where `n` is the rank of the input `tf.Tensor`. Hence by default, this\n * operation performs a regular matrix transpose on 2-D input `tf.Tensor`s.\n *\n * ```js\n * const a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);\n *\n * a.transpose().print(); // or tf.transpose(a)\n * ```\n *\n * @param x The tensor to transpose.\n * @param perm The permutation of the dimensions of a.\n *\n * @doc {heading: 'Operations', subheading: 'Matrices'}\n */\nfunction transpose_(x: T|TensorLike, perm?: number[]): T {\n const $x = convertToTensor(x, 'x', 'transpose');\n\n if (perm == null) {\n perm = $x.shape.map((s, i) => i).reverse();\n }\n util.assert(\n $x.rank === perm.length,\n () => `Error in transpose: rank of input ${$x.rank} ` +\n `must match length of perm ${perm}.`);\n perm.forEach(axis => {\n util.assert(\n axis >= 0 && axis < $x.rank,\n () => `All entries in 'perm' must be between 0 and ${$x.rank - 1}` +\n ` but got ${perm}`);\n });\n\n if ($x.rank <= 1) {\n return $x.clone();\n }\n\n const inputs: TransposeInputs = {x: $x};\n const attrs: TransposeAttrs = {perm};\n\n return ENGINE.runKernelFunc(\n backend => backend.transpose($x, perm), inputs as {} as NamedTensorMap,\n null /* gradient */, Transpose, attrs as {} as NamedAttrMap);\n}\n\nexport const transpose = op({transpose_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor1D, Tensor2D} from '../tensor';\nimport {convertToTensor} from '../tensor_util_env';\nimport {TensorLike} from '../types';\nimport * as util from '../util';\n\nimport {cast} from './cast';\nimport {matMul} from './mat_mul';\nimport {oneHot} from './one_hot';\nimport {op} from './operation';\nimport {transpose} from './transpose';\n\n/**\n * Computes the confusion matrix from true labels and predicted labels.\n *\n * ```js\n * const labels = tf.tensor1d([0, 1, 2, 1, 0], 'int32');\n * const predictions = tf.tensor1d([0, 2, 2, 1, 0], 'int32');\n * const numClasses = 3;\n * const out = tf.math.confusionMatrix(labels, predictions, numClasses);\n * out.print();\n * // Expected output matrix:\n * // [[2, 0, 0],\n * // [0, 1, 1],\n * // [0, 0, 1]]\n * ```\n *\n * @param labels The target labels, assumed to be 0-based integers\n * for the classes. The shape is `[numExamples]`, where\n * `numExamples` is the number of examples included.\n * @param predictions The predicted classes, assumed to be\n * 0-based integers for the classes. Must have the same shape as `labels`.\n * @param numClasses Number of all classes, as an integer.\n * Its value must be larger than the largest element in `labels` and\n * `predictions`.\n * @returns The confusion matrix as a int32-type 2D tensor. The value at\n * row `r` and column `c` is the number of times examples of actual class\n * `r` were predicted as class `c`.\n *\n * @doc {heading: 'Operations', subheading: 'Evaluation'}\n */\nexport function confusionMatrix_(\n labels: Tensor1D|TensorLike, predictions: Tensor1D|TensorLike,\n numClasses: number): Tensor2D {\n const $labels = convertToTensor(labels, 'labels', 'confusionMatrix');\n const $predictions =\n convertToTensor(predictions, 'predictions', 'confusionMatrix');\n\n util.assert(\n numClasses == null || numClasses > 0 && Number.isInteger(numClasses),\n () => `If provided, numClasses must be a positive integer, ` +\n `but got ${numClasses}`);\n util.assert(\n $labels.rank === 1,\n () => `Expected the rank of labels to be 1, but got ${$labels.rank}`);\n util.assert(\n $predictions.rank === 1,\n () => `Expected the rank of predictions to be 1, ` +\n `but got ${$predictions.rank}`);\n util.assert(\n $labels.shape[0] === $predictions.shape[0],\n () => `Mismatch in the number of examples: ` +\n `${$labels.shape[0]} vs. ${$predictions.shape[0]}. ` +\n `Labels and predictions should have the same number of elements.`);\n util.assert(\n numClasses > 0 && Number.isInteger(numClasses),\n () => `numClasses is required to be a positive integer, but got ` +\n `${numClasses}`);\n // TODO(cais): In the future, if oneHot supports tensors inputs for\n // `numClasses`, `confusionMatrix` can make `numClasses` optional.\n\n const oneHotLabels = oneHot(cast($labels, 'int32'), numClasses) as Tensor2D;\n const oneHotPredictions =\n oneHot(cast($predictions, 'int32'), numClasses) as Tensor2D;\n const oneHotLabelsT: Tensor2D = transpose(oneHotLabels);\n return cast(matMul(oneHotLabelsT, oneHotPredictions), 'int32');\n}\n\nexport const confusionMatrix = op({confusionMatrix_});\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\n/**\n * Exports under the tf.math.* namespace.\n */\n\nimport {confusionMatrix} from './ops/confusion_matrix';\n\nexport {confusionMatrix};\n", "/**\n * @license\n * Copyright 2018 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {Tensor3D} from '../tensor';\nimport {inferShape} from '../tensor_util_env';\nimport {TensorLike3D} from '../types';\nimport {DataType} from '../types';\nimport {assertNonNull} from '../util';\nimport {makeTensor} from './tensor_ops_util';\n\n/**\n * Creates rank-3 `tf.Tensor` with the provided values, shape and dtype.\n *\n * The same functionality can be achieved with `tf.tensor`, but in general\n * we recommend using `tf.tensor3d` as it makes the code more readable.\n *\n * ```js\n * // Pass a nested array.\n * tf.tensor3d([[[1], [2]], [[3], [4]]]).print();\n * ```\n * ```js\n * // Pass a flat array and specify a shape.\n * tf.tensor3d([1, 2, 3, 4], [2, 2, 1]).print();\n * ```\n *\n * @param values The values of the tensor. Can be nested array of numbers,\n * or a flat array, or a `TypedArray`.\n * @param shape The shape of the tensor. If not provided, it is inferred from\n * `values`.\n * @param dtype The data type.\n *\n * @doc {heading: 'Tensors', subheading: 'Creation'}\n */\nexport function tensor3d(\n values: TensorLike3D, shape?: [number, number, number],\n dtype?: DataType): Tensor3D {\n assertNonNull(values);\n if (shape != null && shape.length !== 3) {\n throw new Error('tensor3d() requires shape to have three numbers');\n }\n const inferredShape = inferShape(values, dtype);\n if (inferredShape.length !== 3 && inferredShape.length !== 1) {\n throw new Error(\n 'tensor3d() requires values to be number[][][] or flat/TypedArray');\n }\n if (inferredShape.length === 1 && shape == null) {\n throw new Error(\n 'tensor3d() requires shape to be provided when `values` ' +\n 'are a flat array');\n }\n return makeTensor(values, shape, inferredShape, dtype) as Tensor3D;\n}\n", "/**\n * @license\n * Copyright 2019 Google LLC. All Rights Reserved.\n * Licensed under the Apache License, Version 2.0 (the \"License\");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \"AS IS\" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n * =============================================================================\n */\n\nimport {ENGINE} from '../engine';\nimport {FromPixels, FromPixelsAttrs, FromPixelsInputs} from '../kernel_names';\nimport {getKernel, NamedAttrMap} from '../kernel_registry';\nimport {Tensor, Tensor2D, Tensor3D} from '../tensor';\nimport {NamedTensorMap} from '../tensor_types';\nimport {convertToTensor} from '../tensor_util_env';\nimport {PixelData, TensorLike} from '../types';\n\nimport {cast} from './cast';\nimport {op} from './operation';\nimport {tensor3d} from './tensor3d';\n\nlet fromPixels2DContext: CanvasRenderingContext2D;\n\n/**\n * Creates a `tf.Tensor` from an image.\n *\n * ```js\n * const image = new ImageData(1, 1);\n * image.data[0] = 100;\n * image.data[1] = 150;\n * image.data[2] = 200;\n * image.data[3] = 255;\n *\n * tf.browser.fromPixels(image).print();\n * ```\n *\n * @param pixels The input image to construct the tensor from. The\n * supported image types are all 4-channel. You can also pass in an image\n * object with following attributes:\n * `{data: Uint8Array; width: number; height: number}`\n * @param numChannels The number of channels of the output tensor. A\n * numChannels value less than 4 allows you to ignore channels. Defaults to\n * 3 (ignores alpha channel of input image).\n *\n * @doc {heading: 'Browser', namespace: 'browser', ignoreCI: true}\n */\nfunction fromPixels_(\n pixels: PixelData|ImageData|HTMLImageElement|HTMLCanvasElement|\n HTMLVideoElement,\n numChannels = 3): Tensor3D {\n // Sanity checks.\n if (numChannels > 4) {\n throw new Error(\n 'Cannot construct Tensor with more than 4 channels from pixels.');\n }\n if (pixels == null) {\n throw new Error('pixels passed to tf.browser.fromPixels() can not be null');\n }\n let isPixelData = false;\n let isImageData = false;\n let isVideo = false;\n let isImage = false;\n let isCanvasLike = false;\n if ((pixels as PixelData).data instanceof Uint8Array) {\n isPixelData = true;\n } else if (\n typeof (ImageData) !== 'undefined' && pixels instanceof ImageData) {\n isImageData = true;\n } else if (\n typeof (HTMLVideoElement) !== 'undefined' &&\n pixels instanceof HTMLVideoElement) {\n isVideo = true;\n } else if (\n typeof (HTMLImageElement) !== 'undefined' &&\n pixels instanceof HTMLImageElement) {\n isImage = true;\n // tslint:disable-next-line: no-any\n } else if ((pixels as any).getContext != null) {\n isCanvasLike = true;\n } else {\n throw new Error(\n 'pixels passed to tf.browser.fromPixels() must be either an ' +\n `HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData ` +\n `in browser, or OffscreenCanvas, ImageData in webworker` +\n ` or {data: Uint32Array, width: number, height: number}, ` +\n `but was ${(pixels as {}).constructor.name}`);\n }\n if (isVideo) {\n const HAVE_CURRENT_DATA_READY_STATE = 2;\n if (isVideo &&\n (pixels as HTMLVideoElement).readyState <\n HAVE_CURRENT_DATA_READY_STATE) {\n throw new Error(\n 'The video element has not loaded data yet. Please wait for ' +\n '`loadeddata` event on the