/**
 * @license
 * Copyright 2023 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.
 * =============================================================================
 */
!function(t,e){"object"==typeof exports&&"undefined"!=typeof module?e(exports,require("@tensorflow/tfjs-core")):"function"==typeof define&&define.amd?define(["exports","@tensorflow/tfjs-core"],e):e((t="undefined"!=typeof globalThis?globalThis:t||self).tf=t.tf||{},t.tf)}(this,(function(t,e){"use strict";function n(t){var e=Object.create(null);return t&&Object.keys(t).forEach((function(n){if("default"!==n){var s=Object.getOwnPropertyDescriptor(t,n);Object.defineProperty(e,n,s.get?s:{enumerable:!0,get:function(){return t[n]}})}})),e.default=t,e}function s(t,e){return e.forEach((function(e){e&&"string"!=typeof e&&!Array.isArray(e)&&Object.keys(e).forEach((function(n){if("default"!==n&&!(n in t)){var s=Object.getOwnPropertyDescriptor(e,n);Object.defineProperty(t,n,s.get?s:{enumerable:!0,get:function(){return e[n]}})}}))})),t}var i=n(e);class r extends Error{constructor(t){super(t),Object.setPrototypeOf(this,r.prototype)}}class a extends Error{constructor(t){super(t),Object.setPrototypeOf(this,a.prototype)}}class o extends Error{constructor(t){super(t),Object.setPrototypeOf(this,o.prototype)}}class l extends Error{constructor(t){super(t),Object.setPrototypeOf(this,l.prototype)}}class u extends Error{constructor(t){super(t),Object.setPrototypeOf(this,u.prototype)}}class h{constructor(t){this.maxEntries=t||100,this.cache=new Map}get(t){let e;return this.cache.has(t)&&(e=this.cache.get(t),this.cache.delete(t),this.cache.set(t,e)),e}put(t,e){if(this.cache.has(t))this.cache.delete(t);else if(this.cache.size>=this.maxEntries){const t=this.cache.keys().next().value;this.cache.delete(t)}this.cache.set(t,e)}getMaxEntries(){return this.maxEntries}setMaxEntries(t){if(t<0)throw new Error(`The maxEntries of LRU caches must be at least 0, but got ${t}.`);if(this.maxEntries>t)for(let e=0;e<this.maxEntries-t;e++){const t=this.cache.keys().next().value;this.cache.delete(t)}this.maxEntries=t}}function c(t,e){if(Array.isArray(t)){let n=[];for(let s=0;s<e;s++)n=n.concat(t);return n}{const n=new Array(e);return n.fill(t),n}}function p(t,e){if(!t)throw new u(e)}function d(t,e){let n=0;for(const s of t)s===e&&n++;return n}function f(t){return 1===t.length?t[0]:t}function g(t){return Array.isArray(t)?t:[t]}function m(t){const e=t.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2").replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return"_"!==e[0]?e:"private"+e}function y(t){return t.length<=1||-1===t.indexOf("_")?t:t.replace(/[_]+(\w|$)/g,((t,e)=>e.toUpperCase()))}let b={};function w(t){if(null==t)return null;const e={};return e.className=t.getClassName(),e.config=t.getConfig(),e}function k(t){if(null!=t&&"object"==typeof t)if(Array.isArray(t))t.forEach((t=>k(t)));else{const e=Object.keys(t);for(const n of e){const e=t[n];null!=e&&"object"==typeof e&&(Array.isArray(e)||"ndarray"!==e.type||"number"!=typeof e.value?k(e):t[n]=e.value)}}}function v(t,e={},n={},s="object",i=!1){if("string"==typeof t){const i=t;let r;if(i in n)r=n[i];else if(i in b)r=b[i];else if(r=e[i],null==r)throw new o(`Unknown ${s}: ${t}. This may be due to one of the following reasons:\n1. The ${s} is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.\n2. The custom ${s} is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().`);return r}{const r=t;if(null==r.className||null==r.config)throw new o(`${s}: Improper config format: ${JSON.stringify(r)}.\n'className' and 'config' must set.`);const a=r.className;let l,u;if(a in n?[l,u]=n[a]:a in b?[l,u]=b.className:a in e&&([l,u]=e[a]),null==l)throw new o(`Unknown ${s}: ${a}. This may be due to one of the following reasons:\n1. The ${s} is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.\n2. The custom ${s} is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().`);if(null!=u){const t={};for(const e of Object.keys(b))t[e]=b[e];for(const e of Object.keys(n))t[e]=n[e];r.config.customObjects=t;const e=Object.assign({},b);for(const t of Object.keys(n))b[t]=n[t];k(r.config);const s=u(l,r.config,n,i);return b=Object.assign({},e),s}{const t=Object.assign({},b);for(const t of Object.keys(n))b[t]=n[t];const e=new l(r.config);return b=Object.assign({},t),e}}}function S(t,e){return-1*function(t,e){return t<e?-1:t>e?1:0}(t,e)}function x(t){if(null==t)return t;const e=[];for(const n of t)-1===e.indexOf(n)&&e.push(n);return e}function N(t){if(null==t)throw new o(`Invalid value in obj: ${JSON.stringify(t)}`);for(const e in t)if(t.hasOwnProperty(e))return!1;return!0}function I(t,e,n){if(null!=n&&t.indexOf(n)<0)throw new o(`${n} is not a valid ${e}.  Valid values are ${t} or null/undefined.`)}function A(t,e,n=0,s=1/0){return p(n>=0),p(s>=n),Array.isArray(t)&&t.length>=n&&t.length<=s&&t.every((t=>typeof t===e))}function z(t,n){Array.isArray(t)?(e.util.assert(t.length>0,(()=>`${n} is unexpectedly an empty array.`)),t.forEach(((t,e)=>z(t,`element ${e+1} of ${n}`)))):e.util.assert(Number.isInteger(t)&&t>0,(()=>`Expected ${n} to be a positive integer, but got ${E(t)}.`))}function E(t){return null===t?"null":Array.isArray(t)?"["+t.map((t=>E(t))).join(",")+"]":"string"==typeof t?`"${t}"`:`${t}`}function T(t){return"relu"===t?"relu":"linear"===t?"linear":"elu"===t?"elu":null}let C=0;function $(){return C++}const F={};function D(t=""){return t in F||(F[t]=0),F[t]+=1,t+F[t].toString()}const L=["channelsFirst","channelsLast"],_=["nearest","bilinear"],R=["valid","same","causal"],O=["max","avg"],M=["sum","mul","concat","ave"],B=new Map;function P(t){I(L,"DataFormat",t)}function U(t){I(R,"PaddingMode",t)}function W(t){I(O,"PoolMode",t)}const j=[];function q(t,e){j.push(t);try{const t=e();return j.pop(),t}catch(t){throw j.pop(),t}}function V(t){if(!H(t))throw new Error("Not a valid tensor name: '"+t+"'");return(0===j.length?"":j.join("/")+"/")+t}function K(t){if(!H(t))throw new Error("Not a valid tensor name: '"+t+"'");B.has(t)||B.set(t,0);const e=B.get(t);if(B.set(t,B.get(t)+1),e>0){const n=`${t}_${e}`;return B.set(n,1),n}return t}const G=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function H(t){return!!t.match(G)}function J(t,e,n){null==e&&(e=0),null==n&&(n=t.length);let s=1;for(let i=e;i<n;++i)s*=t[i];return s}function Z(t){if(0===t.length)return Number.NaN;let e=Number.POSITIVE_INFINITY;for(let n=0;n<t.length;n++){const s=t[n];s<e&&(e=s)}return e}function Y(t){if(0===t.length)return Number.NaN;let e=Number.NEGATIVE_INFINITY;for(let n=0;n<t.length;n++){const s=t[n];s>e&&(e=s)}return e}function X(t,e){if(e<t)throw new o(`end (${e}) < begin (${t}) is forbidden.`);const n=[];for(let s=t;s<e;++s)n.push(s);return n}let Q;function tt(){return null==Q&&(Q=e.backend().epsilon()),Q}function et(t,e){return i.cast(t,e)}function nt(t,e=-1){const n=t.shape.slice();return e<0&&(e=n.length+e+1),n.splice(e,0,1),i.reshape(t,n)}function st(t,n,s){return e.tidy((()=>{switch(t.rank){case 1:return i.slice1d(t,n,s);case 2:return i.slice2d(t,[n,0],[s,t.shape[1]]);case 3:return i.slice3d(t,[n,0,0],[s,t.shape[1],t.shape[2]]);case 4:return i.slice4d(t,[n,0,0,0],[s,t.shape[1],t.shape[2],t.shape[3]]);case 5:return i.slice(t,[n,0,0,0,0],[s,t.shape[1],t.shape[2],t.shape[3],t.shape[4]]);case 6:return i.slice(t,[n,0,0,0,0,0],[s,t.shape[1],t.shape[2],t.shape[3],t.shape[4],t.shape[5]]);default:throw new o(`sliceAlongFirstAxis() received an unsupported tensor rank: ${t.rank}`)}}))}function it(t,n,s){return e.tidy((()=>{switch(t.rank){case 1:return i.slice1d(t,n,s);case 2:return i.slice2d(t,[0,n],[t.shape[0],s]);case 3:return i.slice3d(t,[0,0,n],[t.shape[0],t.shape[1],s]);case 4:return i.slice4d(t,[0,0,0,n],[t.shape[0],t.shape[1],t.shape[2],s]);default:throw new o(`sliceAlongLastAxis() received an unsupported tensor rank: ${t.rank}`)}}))}function rt(t,n,s,r){return e.tidy((()=>{switch(t.rank){case 1:return i.slice1d(t,n,s);case 2:switch(r){case 1:return st(t,n,s);case 2:return it(t,n,s);default:throw new o(`The axis is not within the rank of the tensor ${r}`)}case 3:switch(r){case 1:return st(t,n,s);case 2:return i.slice3d(t,[0,n,0],[t.shape[0],s,t.shape[2]]);case 3:return it(t,n,s);default:throw new o(`The axis is not within the rank of the tensor ${r}`)}case 4:switch(r){case 1:return st(t,n,s);case 2:return i.slice4d(t,[0,n,0,0],[t.shape[0],s,t.shape[2],t.shape[3]]);case 3:return i.slice4d(t,[0,0,n,0],[t.shape[0],t.shape[1],s,t.shape[3]]);case 4:return it(t,n,s);default:throw new o(`The axis is not within the rank of the tensor ${r}`)}default:throw new o(`sliceAlongLastAxis() received an unsupported tensor rank: ${t.rank}`)}}))}function at(t,e=-1){let n;return e<0&&(n=t[0].rank,e=0!==n?n:0),e===t[0].rank&&(e=-1),i.concat(t,e)}function ot(t,e){switch(t.rank){case 1:return i.concat1d([t,e]);case 2:return i.concat2d([t,e],0);case 3:return i.concat3d([t,e],0);case 4:return i.concat4d([t,e],0);default:throw new o(`concatAlongFirstAxis() received an unsupported tensor rank: ${t.rank}`)}}function lt(t,e){if(Array.isArray(e)||(e=[e]),t.rank!==e.length)throw new o(`The length of input n (${e.length}) does not match the number of dimensions in input x (${t.rank})`);return i.tile(t,e)}function ut(t,e=0,n=1,s,r){return i.randomNormal(t,e,n,s,r)}function ht(t,e,n,s){if(t.rank<2||e.rank<2)throw new l(`dot requires both inputs to be rank >= 2 but got x shape = ${t.shape} and y shape = ${e.shape}`);if(e.rank>=3){if(t.shape.slice(-1)[0]!==e.shape.slice(-2)[0])throw new l(`If rank y >= 3, then the second last dim of y must equal the last dim of x but got x shape = ${t.shape} and  y shape = ${e.shape}`)}if(2===t.rank&&2===e.rank){const r=!1,a=!1;return i.fused.matMul({a:t,b:e,transposeA:r,transposeB:a,bias:s?dt(t.rank,s,"channelsLast"):null,activation:n})}{const r=t.shape.slice(),a=r.pop();t=i.reshape(t,[-1,a]);const o=e.shape.slice(),l=o.pop(),u=o.pop(),h=[...o,l],c=Array.from({length:e.rank},((t,n)=>0===n?e.rank-2:n<=e.rank-2?n-1:n));e=i.reshape(i.transpose(e,c),[u,-1]);const p=[...r,...h],d=!1,f=!1;return i.reshape(i.fused.matMul({a:t,b:e,transposeA:d,transposeB:f,bias:s?dt(t.rank,s,"channelsLast"):null,activation:n}),p)}}function ct(t,n,s){return e.tidy((()=>(n=Array.isArray(n)?e.tensor1d(n,"int32"):i.cast(n,"int32"),i.gather(t,n,s))))}function pt(t){return i.mul(t,t)}function dt(t,e,n){const s=e.shape;if(1!==e.rank&&e.rank!==t)throw new o(`Unexpected bias dimensions: ${e.rank}; expected it to be 1 or ${t}`);if(5===t){if("channelsFirst"===n)return 1===s.length?i.reshape(e,[1,s[0],1,1,1]):i.reshape(e,[1,s[3],s[0],s[1],s[2]]);if("channelsLast"===n)return 1===s.length?i.reshape(e,[1,1,1,1,s[0]]):i.reshape(e,[1].concat(s))}else if(4===t){if("channelsFirst"===n)return 1===s.length?i.reshape(e,[1,s[0],1,1]):i.reshape(e,[1,s[2],s[0],s[1]]);if("channelsLast"===n)return 1===s.length?i.reshape(e,[1,1,1,s[0]]):i.reshape(e,[1].concat(s))}else if(3===t){if("channelsFirst"===n)return 1===s.length?i.reshape(e,[1,s[0],1]):i.reshape(e,[1,s[1],s[0]]);if("channelsLast"===n)return 1===s.length?i.reshape(e,[1,1,s[0]]):i.reshape(e,[1].concat(s))}else if(t<3)return e;throw new o(`Unsupported input rank by biasAdd: ${e.rank}`)}function ft(t,n,s){return e.tidy((()=>(null==s&&(s="channelsLast"),P(s),i.add(t,dt(t.rank,n,s)))))}function gt(t,n,s,r){return e.tidy((()=>i.dropout(t,n,s,r)))}function mt(t,e,n=!1){return n?t():e()}const yt=["fanIn","fanOut","fanAvg"],bt=["normal","uniform","truncatedNormal"];class wt extends e.serialization.Serializable{fromConfigUsesCustomObjects(){return!1}getConfig(){return{}}}class kt extends wt{apply(t,n){return e.zeros(t,n)}}kt.className="Zeros",e.serialization.registerClass(kt);class vt extends wt{apply(t,n){return e.ones(t,n)}}vt.className="Ones",e.serialization.registerClass(vt);class St extends wt{constructor(t){if(super(),"object"!=typeof t)throw new o(`Expected argument of type ConstantConfig but got ${t}`);if(void 0===t.value)throw new o(`config must have value set but got ${t}`);this.value=t.value}apply(t,n){return e.tidy((()=>e.mul(e.scalar(this.value),e.ones(t,n))))}getConfig(){return{value:this.value}}}St.className="Constant",e.serialization.registerClass(St);class xt extends wt{constructor(t){super(),this.DEFAULT_MINVAL=-.05,this.DEFAULT_MAXVAL=.05,this.minval=t.minval||this.DEFAULT_MINVAL,this.maxval=t.maxval||this.DEFAULT_MAXVAL,this.seed=t.seed}apply(t,n){return e.randomUniform(t,this.minval,this.maxval,n,this.seed)}getConfig(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}}}xt.className="RandomUniform",e.serialization.registerClass(xt);class Nt extends wt{constructor(t){super(),this.DEFAULT_MEAN=0,this.DEFAULT_STDDEV=.05,this.mean=t.mean||this.DEFAULT_MEAN,this.stddev=t.stddev||this.DEFAULT_STDDEV,this.seed=t.seed}apply(t,e){if("float32"!==(e=e||"float32")&&"int32"!==e)throw new l(`randomNormal does not support dType ${e}.`);return ut(t,this.mean,this.stddev,e,this.seed)}getConfig(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}}}Nt.className="RandomNormal",e.serialization.registerClass(Nt);class It extends wt{constructor(t){super(),this.DEFAULT_MEAN=0,this.DEFAULT_STDDEV=.05,this.mean=t.mean||this.DEFAULT_MEAN,this.stddev=t.stddev||this.DEFAULT_STDDEV,this.seed=t.seed}apply(t,n){if("float32"!==(n=n||"float32")&&"int32"!==n)throw new l(`truncatedNormal does not support dType ${n}.`);return e.truncatedNormal(t,this.mean,this.stddev,n,this.seed)}getConfig(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}}}It.className="TruncatedNormal",e.serialization.registerClass(It);let At=class extends wt{constructor(t){super(),this.gain=null!=t.gain?t.gain:1}apply(t,n){return e.tidy((()=>{if(2!==t.length||t[0]!==t[1])throw new o("Identity matrix initializer can only be used for 2D square matrices.");return e.mul(this.gain,e.eye(t[0]))}))}getConfig(){return{gain:this.gain}}};At.className="Identity",e.serialization.registerClass(At);class zt extends wt{constructor(t){if(super(),t.scale<0)throw new o(`scale must be a positive float. Got: ${t.scale}`);var e;this.scale=null==t.scale?1:t.scale,this.mode=null==t.mode?"fanIn":t.mode,e=this.mode,I(yt,"FanMode",e),this.distribution=null==t.distribution?"normal":t.distribution,function(t){I(bt,"Distribution",t)}(this.distribution),this.seed=t.seed}apply(t,n){const s=function(t,e="channelsLast"){let n,s;if(P(e),2===t.length)n=t[0],s=t[1];else if(-1!==[3,4,5].indexOf(t.length)){if("channelsFirst"===e){const e=J(t,2);n=t[1]*e,s=t[0]*e}else if("channelsLast"===e){const e=J(t,0,t.length-2);n=t[t.length-2]*e,s=t[t.length-1]*e}}else{const e=J(t);n=Math.sqrt(e),s=Math.sqrt(e)}return[n,s]}(t),i=s[0],r=s[1];let a=this.scale;if("fanIn"===this.mode?a/=Math.max(1,i):"fanOut"===this.mode?a/=Math.max(1,r):a/=Math.max(1,(i+r)/2),"normal"===this.distribution){const s=Math.sqrt(a);if("float32"!==(n=n||"float32")&&"int32"!==n)throw new l(`${this.getClassName()} does not support dType ${n}.`);return e.truncatedNormal(t,0,s,n,this.seed)}{const s=Math.sqrt(3*a);return e.randomUniform(t,-s,s,n,this.seed)}}getConfig(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}}}zt.className="VarianceScaling",e.serialization.registerClass(zt);class Et extends zt{constructor(t){super({scale:1,mode:"fanAvg",distribution:"uniform",seed:null==t?null:t.seed})}getClassName(){return zt.className}}Et.className="GlorotUniform",e.serialization.registerClass(Et);class Tt extends zt{constructor(t){super({scale:1,mode:"fanAvg",distribution:"normal",seed:null==t?null:t.seed})}getClassName(){return zt.className}}Tt.className="GlorotNormal",e.serialization.registerClass(Tt);class Ct extends zt{constructor(t){super({scale:2,mode:"fanIn",distribution:"normal",seed:null==t?null:t.seed})}getClassName(){return zt.className}}Ct.className="HeNormal",e.serialization.registerClass(Ct);class $t extends zt{constructor(t){super({scale:2,mode:"fanIn",distribution:"uniform",seed:null==t?null:t.seed})}getClassName(){return zt.className}}$t.className="HeUniform",e.serialization.registerClass($t);class Ft extends zt{constructor(t){super({scale:1,mode:"fanIn",distribution:"normal",seed:null==t?null:t.seed})}getClassName(){return zt.className}}Ft.className="LeCunNormal",e.serialization.registerClass(Ft);class Dt extends zt{constructor(t){super({scale:1,mode:"fanIn",distribution:"uniform",seed:null==t?null:t.seed})}getClassName(){return zt.className}}Dt.className="LeCunUniform",e.serialization.registerClass(Dt);class Lt extends wt{constructor(t){super(),this.DEFAULT_GAIN=1,this.ELEMENTS_WARN_SLOW=2e3,this.gain=null==t.gain?this.DEFAULT_GAIN:t.gain,this.seed=t.seed}apply(t,n){return e.tidy((()=>{if(t.length<2)throw new l("Shape must be at least 2D.");if("int32"!==n&&"float32"!==n&&void 0!==n)throw new TypeError(`Unsupported data type ${n}.`);const s=e.util.sizeFromShape(t.slice(0,-1)),i=t[t.length-1],r=s*i;r>this.ELEMENTS_WARN_SLOW&&console.warn(`Orthogonal initializer is being called on a matrix with more than ${this.ELEMENTS_WARN_SLOW} (${r}) elements: Slowness may result.`);const a=ut([Math.max(i,s),Math.min(i,s)],0,1,n,this.seed),o=e.linalg.qr(a,!1);let u=o[0];const h=o[1].flatten().stridedSlice([0],[Math.min(i,s)*Math.min(i,s)],[Math.min(i,s)+1]);return u=e.mul(u,h.sign()),s<i&&(u=u.transpose()),e.mul(e.scalar(this.gain),u.reshape(t))}))}getConfig(){return{gain:this.gain,seed:this.seed}}}Lt.className="Orthogonal",e.serialization.registerClass(Lt);const _t={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 Rt(t,n={}){return v(t,e.serialization.SerializationMap.getMap().classNameMap,n,"initializer")}function Ot(t){return w(t)}function Mt(t){if("string"==typeof t){const e=t in _t?_t[t]:t;if("GlorotNormal"===e)return new Tt;if("GlorotUniform"===e)return new Et;if("HeNormal"===e)return new Ct;if("HeUniform"===e)return new $t;if("LeCunNormal"===e)return new Ft;if("LeCunUniform"===e)return new Dt;{const t={};return t.className=e,t.config={},Rt(t)}}return t instanceof wt?t:Rt(t)}function Bt(t){return Array.isArray(t)&&Array.isArray(t[0])}function Pt(t){return 0===t.length?[]:Array.isArray(t[0])?t:[t]}function Ut(t){let e;if(Array.isArray(t)){if(1!==t.length)throw new o(`Expected Tensor length to be 1; got ${t.length}`);e=t[0]}else e=t;return e}function Wt(t){if(Array.isArray(t)&&Array.isArray(t[0])){if(1===t.length)return t[0];throw new o(`Expected exactly 1 Shape; got ${t.length}`)}return t}function jt(t){let e=0;for(const n of t)0===n.shape.length?e+=1:e+=n.shape.reduce(((t,e)=>t*e));return e}const qt="Variable";class Vt{constructor(t,e="float32",n="Variable",s=!0,r=null){this.dtype=null==e?"float32":e,this.shape=t.shape,this.id=$(),n=null==n?qt:n,this.originalName=V(n),this.name=K(this.originalName),this.trainable_=s,this.constraint=r,this.val=i.variable(t,this.trainable_,this.name,this.dtype)}read(){return this.assertNotDisposed(),this.val}write(t){return this.assertNotDisposed(),function(t,e){if(t.shape.toString()!==e.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(t.shape)+" vs. "+JSON.stringify(e.shape))}(this.val,t),this.val.id!==t.id&&(this.val.assign(t),null!=this.constraint&&this.val.assign(this.constraint.apply(this.val))),this}dispose(){this.assertNotDisposed(),this.val.dispose()}assertNotDisposed(){if(this.val.isDisposed)throw new Error(`LayersVariable ${this.name} is already disposed.`)}get trainable(){return this.trainable_}set trainable(t){this.trainable_=t,this.val.trainable=t}}function Kt(t){return t.map((t=>t.read()))}function Gt(t){t.forEach((t=>{t[0].write(t[1])}))}class Ht{constructor(t){this.dtype=t.dtype,this.shape=t.shape,null!=t.shape?this.ndim=t.shape.length:this.ndim=t.ndim,this.maxNDim=t.maxNDim,this.minNDim=t.minNDim,this.axes=t.axes||{}}}class Jt{constructor(t,e,n,s,i,r,a){this.dtype=t,this.shape=e,this.sourceLayer=n,this.inputs=s,this.callArgs=i,this.outputTensorIndex=a,this.id=$(),null!=r&&(this.originalName=V(r),this.name=K(this.originalName)),this.rank=e.length}}let Zt=0;class Yt{constructor(t,e){this.callArgs=e,this.id=Zt++,this.outboundLayer=t.outboundLayer,this.inboundLayers=t.inboundLayers,this.nodeIndices=t.nodeIndices,this.tensorIndices=t.tensorIndices,this.inputTensors=t.inputTensors,this.outputTensors=t.outputTensors,this.inputMasks=t.inputMasks,this.outputMasks=t.outputMasks,this.inputShapes=t.inputShapes,this.outputShapes=t.outputShapes;for(const e of t.inboundLayers)null!=e&&e.outboundNodes.push(this);t.outboundLayer.inboundNodes.push(this)}getConfig(){const t=[];for(const e of this.inboundLayers)null!=e?t.push(e.name):t.push(null);return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:t,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}}}let Xt=0;class Qt extends e.serialization.Serializable{constructor(t={}){super(),this._callHook=null,this._addedWeightNames=[],this._stateful=!1,this.id=Xt++,this.activityRegularizer=null,this.inputSpec=null,this.supportsMasking=!1,this._trainableWeights=[],this._nonTrainableWeights=[],this._losses=[],this._updates=[],this._built=!1,this.inboundNodes=[],this.outboundNodes=[];let e=t.name;if(!e){const t=this.getClassName();e=m(t)+"_"+D(t)}if(this.name=e,this.trainable_=null==t.trainable||t.trainable,null!=t.inputShape||null!=t.batchInputShape){let e;if(null!=t.batchInputShape)e=t.batchInputShape;else if(null!=t.inputShape){let n=null;null!=t.batchSize&&(n=t.batchSize),e=[n].concat(t.inputShape)}this.batchInputShape=e;let n=t.dtype;null==n&&(n=t.inputDType),null==n&&(n="float32"),this.dtype=n}null!=t.weights?this.initialWeights=t.weights:this.initialWeights=null,this._refCount=null,this.fastWeightInitDuringBuild=!1}static nodeKey(t,e){return t.name+"_ib-"+e.toString()}getNodeAtIndex(t,e){if(0===this.inboundNodes.length)throw new a(`The layer has never been called and thus has no defined ${e}.`);if(this.inboundNodes.length<=t)throw new o(`Asked to get ${e} at node ${t}, but the layer has only ${this.inboundNodes.length} inbound nodes.`);return this.inboundNodes[t]}getInputAt(t){return f(this.getNodeAtIndex(t,"input").inputTensors)}getOutputAt(t){return f(this.getNodeAtIndex(t,"output").outputTensors)}get input(){if(this.inboundNodes.length>1)throw new r(`Layer ${this.name} has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use \`getInputAt(nodeIndex)\` instead.`);if(0===this.inboundNodes.length)throw new r(`Layer ${this.name} is not connected, no input to return.`);return f(this.getNodeAtIndex(0,"input").inputTensors)}get output(){if(0===this.inboundNodes.length)throw new r(`Layer ${this.name} has no inbound nodes.`);if(this.inboundNodes.length>1)throw new r(`Layer ${this.name} has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use \`getOutputAt(nodeIndex)\` instead.`);return f(this.getNodeAtIndex(0,"output").outputTensors)}get losses(){return this._losses}calculateLosses(){return this.losses.map((t=>t()))}get updates(){return this._updates}get built(){return this._built}set built(t){this._built=t}get trainable(){return this.trainable_}set trainable(t){this._trainableWeights.forEach((e=>e.trainable=t)),this.trainable_=t}get trainableWeights(){return this.trainable_?this._trainableWeights.filter((t=>t.trainable)):[]}set trainableWeights(t){this._trainableWeights=t}get nonTrainableWeights(){return this.trainable?this._trainableWeights.filter((t=>!t.trainable)).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)}set nonTrainableWeights(t){this._nonTrainableWeights=t}get weights(){return this.trainableWeights.concat(this.nonTrainableWeights)}get stateful(){return this._stateful}resetStates(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")}assertInputCompatibility(t){const e=g(t);if(null==this.inputSpec||0===this.inputSpec.length)return;const n=g(this.inputSpec);if(e.length!==n.length)throw new o(`Layer ${this.name} expects ${n.length} inputs, but it received ${e.length} input tensors. Input received: ${t}`);for(let t=0;t<e.length;t++){const s=e[t],i=n[t];if(null==i)continue;const r=s.rank;if(null!=i.ndim&&r!==i.ndim)throw new o(`Input ${t} is incompatible with layer ${this.name}: expected ndim=${i.ndim}, found ndim=${r}`);if(null!=i.maxNDim&&r>i.maxNDim)throw new o(`Input ${t} is incompatible with layer ${this.name}: expected max_ndim=${i.maxNDim}, found ndim=${r}`);if(null!=i.minNDim&&r<i.minNDim)throw new o(`Input ${t} is incompatible with layer ${this.name}: expected min_ndim=${i.minNDim}, found ndim=${r}.`);if(null!=i.dtype&&s.dtype!==i.dtype)throw new o(`Input ${t} is incompatible with layer ${this.name} : expected dtype=${i.dtype}, found dtype=${s.dtype}.`);if(i.axes){const e=s.shape;for(const n in i.axes){const s=Number(n),r=i.axes[n],a=s>=0?e[s]:e[e.length+s];if(null!=r&&-1===[r,null].indexOf(a))throw new o(`Input ${t} is incompatible with layer ${this.name}: expected axis ${s} of input shape to have value ${r} but got shape ${e}.`)}}if(null!=i.shape)for(let e=0;e<i.shape.length;++e){const n=i.shape[e],r=s.shape[e];if(null!=n&&null!=r&&n!==r)throw new o(`Input ${t} is incompatible with layer ${this.name}: expected shape=${i.shape}, found shape=${s.shape}.`)}}}call(t,e){return t}invokeCallHook(t,e){null!=this._callHook&&this._callHook(t,e)}setCallHook(t){this._callHook=t}clearCallHook(){this._callHook=null}apply(t,e){e=e||{},this.assertNotDisposed();const n=g(t),s=function(t){let e=!0;for(const n of g(t))if(!(n instanceof Jt)){e=!1;break}return e}(t),i=function(t){let e=!0;for(const n of g(t))if(n instanceof Jt){e=!1;break}return e}(t);if(s===i)throw new o("Arguments to apply() must be all SymbolicTensors or all Tensors");return q(this.name,(()=>{if(!this.built){this.assertInputCompatibility(t);const e=[];for(const n of g(t))e.push(n.shape);this.build(f(e)),this.built=!0,this.initialWeights&&this.setWeights(this.initialWeights),null===this._refCount&&i&&(this._refCount=1)}if(this.assertInputCompatibility(t),i){let s=this.call(t,e);this.supportsMasking&&this.setMaskMetadata(t,s);const i=g(s),r=[];for(let t of i)-1!==n.indexOf(t)&&(t=t.clone()),r.push(t);if(s=f(r),null!=this.activityRegularizer)throw new l("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return s}{const n=function(t){t=g(t);const e=[];for(const n of t)e.push(n.shape);return f(e)}(t),s=this.computeOutputShape(n);let i;const r="float32";if(this.warnOnIncompatibleInputShape(Array.isArray(t)?n[0]:n),i=null!=s&&s.length>0&&Array.isArray(s[0])?s.map(((n,s)=>new Jt(r,n,this,g(t),e,this.name,s))):new Jt(r,s,this,g(t),e,this.name),this.addInboundNode(t,i,null,null,n,s,e),this._refCount++,null!=this.activityRegularizer)throw new l("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return i}}))}warnOnIncompatibleInputShape(t){if(null!=this.batchInputShape)if(t.length!==this.batchInputShape.length)console.warn(`The rank of the input tensor provided (shape: ${JSON.stringify(t)}) does not match that of the batchInputShape (${JSON.stringify(this.batchInputShape)}) of the layer ${this.name}`);else{let e=!1;this.batchInputShape.forEach(((n,s)=>{null!=n&&null!=t[s]&&t[s]!==n&&(e=!0)})),e&&console.warn(`The shape of the input tensor (${JSON.stringify(t)}) does not match the expectation of layer ${this.name}: ${JSON.stringify(this.batchInputShape)}`)}}get outputShape(){if(null==this.inboundNodes||0===this.inboundNodes.length)throw new r(`The layer ${this.name} has never been called and thus has no defined output shape.`);const t=[];for(const e of this.inboundNodes){const n=JSON.stringify(e.outputShapes);-1===t.indexOf(n)&&t.push(n)}if(1===t.length){const t=this.inboundNodes[0].outputShapes;return Array.isArray(t)&&Array.isArray(t[0])&&1===t.length?t[0]:t}throw new r(`The layer ${this.name} has multiple inbound nodes with different output shapes. Hence the notion of "output shape" is ill-defined for the layer.`)}countParams(){if(!this.built)throw new a(`You tried to call countParams() on ${this.name}, but the layer is not built yet. Build it first by calling build(batchInputShape).`);return jt(this.weights)}build(t){this.built=!0}getWeights(t=!1){return Kt(t?this.trainableWeights:this.weights)}setWeights(t){e.tidy((()=>{const n=this.weights;if(n.length!==t.length)throw new o(`You called setWeights(weights) on layer "${this.name}" with a weight list of length ${t.length}, but the layer was expecting ${n.length} weights. Provided weights: ${t}...`);if(0===n.length)return;const s=[],i=Kt(n);for(let r=0;r<i.length;++r){const a=i[r],l=n[r],u=t[r];if(!e.util.arraysEqual(a.shape,u.shape))throw new o(`Layer weight shape ${a.shape} not compatible with provided weight shape ${u.shape}`);s.push([l,u])}Gt(s)}))}addWeight(t,e,n,s,i,r,a,l){if(-1!==this._addedWeightNames.indexOf(t))throw new o(`Duplicate weight name ${t} for layer ${this.name}`);this._addedWeightNames.push(t),null==n&&(n="float32"),this.fastWeightInitDuringBuild&&(s=null!=l?l():Mt("zeros"));const u=s.apply(e,n),h=new Vt(u,n,t,r,a);return u.dispose(),null!=i&&this.addLoss((()=>i.apply(h.read()))),null==r&&(r=!0),r?this._trainableWeights.push(h):this._nonTrainableWeights.push(h),h}setFastWeightInitDuringBuild(t){this.fastWeightInitDuringBuild=t}addLoss(t){null==t||Array.isArray(t)&&0===t.length||(t=g(t),void 0!==this._losses&&null!==this._losses&&this.losses.push(...t))}computeOutputShape(t){return t}computeMask(t,e){if(!this.supportsMasking){if(null!=e){if(!Array.isArray(e))throw new TypeError(`Layer ${this.name} does not support masking, but was passed an inputMask.`);e.forEach((t=>{if(null!=t)throw new TypeError(`Layer ${this.name} does not support masking, but was passed an inputMask.`)}))}return null}return e}setMaskMetadata(t,e,n){if(!this.supportsMasking)return;const s=this.computeMask(t,n),i=g(e),r=g(s);if(i.length!==r.length)throw new Error(`${this.name} outputs ${i.length} tensors but ${i.length} masks for those tensors`);for(let t=0;t<i.length;t++)i[t].kerasMask=r[t]}addInboundNode(t,e,n,s,i,r,a=null){const o=g(t);e=g(e),n=g(n),s=g(s),i=Pt(i),r=Pt(r);const l=[],u=[],h=[];for(const t of o)l.push(t.sourceLayer),u.push(t.nodeIndex),h.push(t.tensorIndex);new Yt({outboundLayer:this,inboundLayers:l,nodeIndices:u,tensorIndices:h,inputTensors:o,outputTensors:e,inputMasks:n,outputMasks:s,inputShapes:i,outputShapes:r},a);for(let t=0;t<e.length;t++)e[t].sourceLayer=this,e[t].nodeIndex=this.inboundNodes.length-1,e[t].tensorIndex=t}getConfig(){const t={name:this.name,trainable:this.trainable};return null!=this.batchInputShape&&(t.batchInputShape=this.batchInputShape),null!=this.dtype&&(t.dtype=this.dtype),t}disposeWeights(){return this.weights.forEach((t=>t.dispose())),this.weights.length}assertNotDisposed(){if(0===this._refCount)throw new Error(`Layer '${this.name}' is already disposed.`)}dispose(){if(!this.built)throw new Error(`Cannot dispose Layer ${this.name} because it has not been built yet.`);if(null===this._refCount)throw new Error(`Cannot dispose Layer ${this.name} because it has not been used yet.`);this.assertNotDisposed();let t=0;return 0==--this._refCount&&(t=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:t}}}function te(t,e,n){if((null==e||null!=n&&n>0)&&(e=t.sourceLayer,n=t.nodeIndex),0===e.inboundNodes.length)return[t];{const t=e.inboundNodes[n];if(0===t.inboundLayers.length)return t.inputTensors;{const e=[];for(let n=0;n<t.inboundLayers.length;n++){const s=te(t.inputTensors[n],t.inboundLayers[n],t.nodeIndices[n]);for(const t of s)-1===e.indexOf(t)&&e.push(t)}return e}}}class ee extends Qt{constructor(t){if(super({dtype:t.dtype,name:null!=t.name?t.name:D("input").toString()}),null==t.batchSize&&(t.batchSize=null),null==t.sparse&&(t.sparse=!1),this.trainable=!1,this.built=!0,this.sparse=t.sparse,null!=t.inputShape&&null!=t.batchInputShape)throw new o("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");let e=t.batchInputShape;if(null==e){if(null==t.inputShape)throw new o("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");e=[t.batchSize].concat(t.inputShape)}else if(null!=t.batchSize)throw new o("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");const n=t.dtype||"float32";this.batchInputShape=e,this.dtype=n,this.inputSpec=[{shape:e}];const s=new Jt(this.dtype,this.batchInputShape,this,[],{},this.name);s.nodeIndex=0,s.tensorIndex=0,new Yt({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[s],outputTensors:[s],inputMasks:[null],outputMasks:[null],inputShapes:[e],outputShapes:[e]})}apply(t,e){throw new o(`Cannot pass any input to an InputLayer's apply() method. InputLayer name: ${this.name}`)}dispose(){return{refCountAfterDispose:this._refCount,numDisposedVariables:0}}getConfig(){return{batchInputShape:this.batchInputShape,dtype:this.dtype,sparse:this.sparse,name:this.name}}}function ne(t){if(null==t.batchShape&&null==t.shape)throw new Error("Please provide to Input either a `shape` or a `batchShape` argument. Note that `shape` does not include the batch dimension.");if(null!=t.batchShape&&null!=t.shape)throw new o("Please provide either a `shape` or `batchShape` argument to Input, but not both.");let e=t.batchShape;null!=t.shape&&null==e&&(e=[null].concat(t.shape));let n=t.dtype;null==n&&(n="float32");return new ee({batchInputShape:e,name:t.name,dtype:n,sparse:t.sparse}).inboundNodes[0].outputTensors[0]}ee.className="InputLayer",e.serialization.registerClass(ee);class se{constructor(t){if(this.id2Value={},this.id2Mask={},this.name2Id={},t instanceof se)for(const e in t.id2Value)this.id2Value[e]=t.id2Value[e],e in t.id2Mask&&(this.id2Mask[e]=t.id2Mask[e]);else{if(null==t)return;for(const e of t)this.add(e.key,e.value)}}add(t,n,s){if(null!=this.id2Value[t.id])throw new o(`Duplicate key: name=${t.name}, id=${t.id}`);return this.id2Value[t.id]=function(t,n){if(null==t.dtype||t.dtype===n.dtype)return n;try{return e.cast(n,t.dtype)}catch(e){throw new o(`The dtype of the feed (${n.dtype}) can not be cast to the dtype of the key '${t.name}' (${t.dtype}).`)}}(t,n),this.name2Id[t.name]=t.id,null!=s&&(this.id2Mask[t.id]=s),this}addFeed(t){this.add(t.key,t.value)}hasKey(t){return null!=this.id2Value[t.id]}names(){return Object.keys(this.name2Id)}getValue(t){if(t instanceof Jt){if(null==this.id2Value[t.id])throw new o(`Nonexistent key: ${t.name}`);return this.id2Value[t.id]}{const e=this.name2Id[t];if(null==e)throw new o(`Feed dict has no SymbolicTensor name: ${t}`);return this.id2Value[e]}}getMask(t){if(t instanceof Jt){if(null==this.id2Value[t.id])throw new o(`Nonexistent key: ${t.name}`);return this.id2Mask[t.id]}{const e=this.name2Id[t];if(null==e)throw new o(`Feed dict has no SymbolicTensor name: ${t}`);return this.id2Mask[e]}}disposeMasks(){null!=this.id2Mask&&e.dispose(this.id2Mask)}}const ie=new h,re=new h;function ae(t,n,s,i){const r=null!=s&&s.training,a=Array.isArray(t),o=a?t:[t],l=o.map((t=>t.name)),u=[],h=n.names();for(const t of l)-1!==h.indexOf(t)?u.push(n.getValue(t)):u.push(null);null!=i&&(i.maxNumTensors=-1/0,i.minNumTensors=1/0);const c=l.join(",")+"|"+n.names().sort().join(",");let p,d=ie.get(c);if(null==d){const t=function(t,n){e.util.assert(null!=t&&t.length>0,(()=>"Expected at least one fetch, got none"));let s=[],i={};if(1===t.length){const e=le(t[0],n);s=e.sorted,i=e.recipientMap}else{const e=new Set;for(const r of t){const{sorted:t,recipientMap:a}=le(r,n);for(const n of t)e.has(n.name)||(s.push(n),e.add(n.name));for(const t in a)null==i[t]&&(i[t]=new Set),a[t].forEach((e=>i[t].add(e)))}}return{sorted:s,recipientCounts:oe(i)}}(o,n);d=t.sorted,p=t.recipientCounts,ie.put(c,d),re.put(c,p)}p={},r||Object.assign(p,re.get(c));const f=new se(n);for(let t=0;t<d.length;++t){if(null!=i){const t=e.memory().numTensors;t>i.maxNumTensors&&(i.maxNumTensors=t),t<i.minNumTensors&&(i.minNumTensors=t)}const a=d[t],o=a.sourceLayer;if(o instanceof ee)continue;const h=[],c=[],m=[];let y=!1;for(const t of a.inputs){const e=f.getValue(t),s=f.getMask(t);h.push(e),c.push(s),null!=s&&(y=!0),r||(p[t.name]--,0!==p[t.name]||n.hasKey(t)||-1!==l.indexOf(t.name)||e.isDisposed||!0===t.sourceLayer.stateful||m.push(e))}y&&((s=s||{}).mask=c[0]);const b=g(o.apply(h,s));let w=null;o.supportsMasking&&(w=o.computeMask(h,c));const k=ue(a),v=Array.isArray(k)?k:[k];for(let t=0;t<v.length;++t){f.hasKey(v[t])||f.add(v[t],b[t],Array.isArray(w)?w[0]:w);const e=l.indexOf(v[t].name);-1!==e&&(u[e]=b[t])}r||e.dispose(m)}return f.disposeMasks(),a?u:u[0]}function oe(t){const e={};for(const n in t)e[n]=t[n].size;return e}function le(t,e){const n=new Set,s=[],i={};for(const t of e.names())n.add(t);const r=[],a=[];for(r.push(t);r.length>0;){const t=r[r.length-1];if(n.has(t.name)){r.pop();continue}const e=a[a.length-1]===r.length-1;if(0===t.inputs.length||e)r.pop(),s.push(t),n.add(t.name),e&&a.pop();else{a.push(r.length-1);for(const e of t.inputs)null==i[e.name]&&(i[e.name]=new Set),i[e.name].add(t.name),n.has(e.name)||r.push(e)}}return{sorted:s,recipientMap:i}}function ue(t){let e;if(1===t.sourceLayer.inboundNodes.length)e=t.sourceLayer.output;else{let n=null;for(let e=0;e<t.sourceLayer.inboundNodes.length;++e)for(const s of t.sourceLayer.inboundNodes[e].outputTensors)if(s.id===t.id){n=e;break}e=t.sourceLayer.getOutputAt(n)}return e}e.env().registerFlag("TOPOLOGICAL_SORT_CACHE_MAX_ENTRIES",(()=>100),(function(t){null!=ie&&ie.setMaxEntries(t),null!=re&&re.setMaxEntries(t)}));const he="Add",ce="BatchMatMul",pe="BatchToSpaceND",de="Cast",fe="Concat",ge="Conv2D",me="Conv2DBackpropInput",ye="Cosh",be="Cumsum",we="RealDiv",ke="ExpandDims",ve="Floor",Se="FloorDiv",xe="GatherV2",Ne="GreaterEqual",Ie="Identity",Ae="Maximum",ze="Multiply",Ee="Pack",Te="PadV2",Ce="Reshape",$e="Reverse",Fe="Rsqrt",De="Select",Le="Slice",_e="Sinh",Re="Sigmoid",Oe="Sqrt",Me="SpaceToBatchND",Be="SplitV",Pe="Tile",Ue="Transpose",We="Unpack",je="UnsortedSegmentSum",qe="ZerosLike",Ve="Step";function Ke(t){throw new Error(`'${t}' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen`)}function Ge(t,e){if(!t)throw new Error("string"==typeof e?e:e())}function He(t){if(0===t.length)return 1;let e=t[0];for(let n=1;n<t.length;n++)e*=t[n];return e}function Je(t,e){if(t===e)return!0;if(null==t||null==e)return!1;if(t.length!==e.length)return!1;for(let n=0;n<t.length;n++)if(t[n]!==e[n])return!1;return!0}function Ze(t){return t%1==0}function Ye(t,e){return e<=t.length?t:t+" ".repeat(e-t.length)}function Xe(t,e){const n=e.length;return Ge((t=null==t?e.map(((t,e)=>e)):[].concat(t)).every((t=>t>=-n&&t<n)),(()=>`All values in axis param must be in range [-${n}, ${n}) but got axis ${t}`)),Ge(t.every((t=>Ze(t))),(()=>`All values in axis param must be integers but got axis ${t}`)),t.map((t=>t<0?n+t:t))}function Qe(t){if("float32"===t||"int32"===t)return 4;if("complex64"===t)return 8;if("bool"===t)return 1;throw new Error(`Unknown dtype ${t}`)}function tn(t){return"string"==typeof t||t instanceof String}function en(t){return Array.isArray(t)?en(t[0]):t instanceof Float32Array?"float32":t instanceof Int32Array||t instanceof Uint8Array||t instanceof Uint8ClampedArray?"int32":"number"==typeof t?"float32":tn(t)?"string":function(t){return"boolean"==typeof t}(t)?"bool":"float32"}function nn(t){return!!(t&&t.constructor&&t.call&&t.apply)}function sn(t){const e=t.length;if(e<2)return[];const n=new Array(e-1);n[e-2]=t[e-1];for(let s=e-3;s>=0;--s)n[s]=n[s+1]*t[s+1];return n}function rn(t,e,n,s=!1){const i=new Array;if(1===e.length){const r=e[0]*(s?2:1);for(let e=0;e<r;e++)i[e]=n[t+e]}else{const r=e[0],a=e.slice(1),o=a.reduce(((t,e)=>t*e))*(s?2:1);for(let e=0;e<r;e++)i[e]=rn(t+e*o,a,n,s)}return i}function an(t,e,n=!1){if(0===t.length)return e[0];const s=t.reduce(((t,e)=>t*e))*(n?2:1);if(0===s)return[];if(s!==e.length)throw new Error(`[${t}] does not match the input size ${e.length}${n?" for a complex tensor":""}.`);return rn(0,t,e,n)}function on(t,e){const n=ln(t,e);for(let t=0;t<n.length;t++)n[t]=1;return n}function ln(t,e){if(null==e||"float32"===e||"complex64"===e)return new Float32Array(t);if("int32"===e)return new Int32Array(t);if("bool"===e)return new Uint8Array(t);throw new Error(`Unknown data type ${e}`)}function un(t){t.forEach((e=>{Ge(Number.isInteger(e)&&e>=0,(()=>`Tensor must have a shape comprised of positive integers but got shape [${t}].`))}))}function hn(t){return t&&t.then&&"function"==typeof t.then}const cn="tfjsflags";class pn{constructor(t){this.global=t,this.flags={},this.flagRegistry={},this.urlFlags={},this.getQueryParams=dn,this.populateURLFlags()}setPlatform(t,e){null!=this.platform&&(fn().getBool("IS_TEST")||fn().getBool("PROD")||console.warn(`Platform ${this.platformName} has already been set. Overwriting the platform with ${t}.`)),this.platformName=t,this.platform=e}registerFlag(t,e,n){if(this.flagRegistry[t]={evaluationFn:e,setHook:n},null!=this.urlFlags[t]){const e=this.urlFlags[t];fn().getBool("IS_TEST")||fn().getBool("PROD")||console.warn(`Setting feature override from URL ${t}: ${e}.`),this.set(t,e)}}async getAsync(t){return t in this.flags||(this.flags[t]=await this.evaluateFlag(t)),this.flags[t]}get(t){if(t in this.flags)return this.flags[t];const e=this.evaluateFlag(t);if(hn(e))throw new Error(`Flag ${t} cannot be synchronously evaluated. Please use getAsync() instead.`);return this.flags[t]=e,this.flags[t]}getNumber(t){return this.get(t)}getBool(t){return this.get(t)}getString(t){return this.get(t)}getFlags(){return this.flags}get features(){return this.flags}set(t,e){if(null==this.flagRegistry[t])throw new Error(`Cannot set flag ${t} as it has not been registered.`);this.flags[t]=e,null!=this.flagRegistry[t].setHook&&this.flagRegistry[t].setHook(e)}evaluateFlag(t){if(null==this.flagRegistry[t])throw new Error(`Cannot evaluate flag '${t}': no evaluation function found.`);return this.flagRegistry[t].evaluationFn()}setFlags(t){this.flags=Object.assign({},t)}reset(){this.flags={},this.urlFlags={},this.populateURLFlags()}populateURLFlags(){if("undefined"==typeof this.global||"undefined"==typeof this.global.location||"undefined"==typeof this.global.location.search)return;const t=this.getQueryParams(this.global.location.search);if(cn in t){t.tfjsflags.split(",").forEach((t=>{const[e,n]=t.split(":");this.urlFlags[e]=function(t,e){const n=e.toLowerCase();return"true"===n||"false"===n?"true"===n:""+ +n===n?+n:e}(0,n)}))}}}function dn(t){const e={};return t.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,((t,...n)=>(function(t,e,n){t[decodeURIComponent(e)]=decodeURIComponent(n||"")}(e,n[0],n[1]),n.join("=")))),e}function fn(){return mn}let gn,mn=null;function yn(){if(null==gn){let t;if("undefined"!=typeof window)t=window;else if("undefined"!=typeof global)t=global;else if("undefined"!=typeof process)t=process;else{if("undefined"==typeof self)throw new Error("Could not find a global object");t=self}gn=t}return gn}function bn(t,e){const n=function(){const t=yn();return null==t._tfGlobals&&(t._tfGlobals=new Map),t._tfGlobals}();if(n.has(t))return n.get(t);{const s=e();return n.set(t,s),n.get(t)}}function wn(...t){fn().getBool("IS_TEST")||fn().getBool("PROD")||console.warn(...t)}const kn=bn("kernelRegistry",(()=>new Map)),vn=bn("gradRegistry",(()=>new Map));function Sn(t,e){const n=function(t,e){return`${e}_${t}`}(t,e);return kn.get(n)}function xn(t){return vn.get(t)}function Nn(t){const e=kn.entries(),n=[];for(;;){const{done:s,value:i}=e.next();if(s)break;const[r,a]=i,[o]=r.split("_");o===t&&n.push(a)}return n}function In(t){const{kernelName:e}=t;vn.has(e)&&fn().getBool("DEBUG")&&wn(`Overriding the gradient for '${e}'`),vn.set(e,t)}var An="undefined"!=typeof globalThis?globalThis:"undefined"!=typeof window?window:"undefined"!=typeof global?global:"undefined"!=typeof self?self:{};function zn(t){return t&&t.__esModule&&Object.prototype.hasOwnProperty.call(t,"default")?t.default:t}function En(t){if(t.__esModule)return t;var e=t.default;if("function"==typeof e){var n=function t(){if(this instanceof t){var n=[null];n.push.apply(n,arguments);var s=Function.bind.apply(e,n);return new s}return e.apply(this,arguments)};n.prototype=e.prototype}else n={};return Object.defineProperty(n,"__esModule",{value:!0}),Object.keys(t).forEach((function(e){var s=Object.getOwnPropertyDescriptor(t,e);Object.defineProperty(n,e,s.get?s:{enumerable:!0,get:function(){return t[e]}})})),n}var Tn=$n,Cn=null;try{Cn=new WebAssembly.Instance(new WebAssembly.Module(new Uint8Array([0,97,115,109,1,0,0,0,1,13,2,96,0,1,127,96,4,127,127,127,127,1,127,3,7,6,0,1,1,1,1,1,6,6,1,127,1,65,0,11,7,50,6,3,109,117,108,0,1,5,100,105,118,95,115,0,2,5,100,105,118,95,117,0,3,5,114,101,109,95,115,0,4,5,114,101,109,95,117,0,5,8,103,101,116,95,104,105,103,104,0,0,10,191,1,6,4,0,35,0,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,126,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,127,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,128,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,129,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,130,34,4,66,32,135,167,36,0,32,4,167,11])),{}).exports}catch(t){}function $n(t,e,n){this.low=0|t,this.high=0|e,this.unsigned=!!n}function Fn(t){return!0===(t&&t.__isLong__)}$n.prototype.__isLong__,Object.defineProperty($n.prototype,"__isLong__",{value:!0}),$n.isLong=Fn;var Dn={},Ln={};function _n(t,e){var n,s,i;return e?(i=0<=(t>>>=0)&&t<256)&&(s=Ln[t])?s:(n=On(t,(0|t)<0?-1:0,!0),i&&(Ln[t]=n),n):(i=-128<=(t|=0)&&t<128)&&(s=Dn[t])?s:(n=On(t,t<0?-1:0,!1),i&&(Dn[t]=n),n)}function Rn(t,e){if(isNaN(t))return e?Kn:Vn;if(e){if(t<0)return Kn;if(t>=Wn)return Yn}else{if(t<=-jn)return Xn;if(t+1>=jn)return Zn}return t<0?Rn(-t,e).neg():On(t%Un|0,t/Un|0,e)}function On(t,e,n){return new $n(t,e,n)}$n.fromInt=_n,$n.fromNumber=Rn,$n.fromBits=On;var Mn=Math.pow;function Bn(t,e,n){if(0===t.length)throw Error("empty string");if("NaN"===t||"Infinity"===t||"+Infinity"===t||"-Infinity"===t)return Vn;if("number"==typeof e?(n=e,e=!1):e=!!e,(n=n||10)<2||36<n)throw RangeError("radix");var s;if((s=t.indexOf("-"))>0)throw Error("interior hyphen");if(0===s)return Bn(t.substring(1),e,n).neg();for(var i=Rn(Mn(n,8)),r=Vn,a=0;a<t.length;a+=8){var o=Math.min(8,t.length-a),l=parseInt(t.substring(a,a+o),n);if(o<8){var u=Rn(Mn(n,o));r=r.mul(u).add(Rn(l))}else r=(r=r.mul(i)).add(Rn(l))}return r.unsigned=e,r}function Pn(t,e){return"number"==typeof t?Rn(t,e):"string"==typeof t?Bn(t,e):On(t.low,t.high,"boolean"==typeof e?e:t.unsigned)}$n.fromString=Bn,$n.fromValue=Pn;var Un=4294967296,Wn=Un*Un,jn=Wn/2,qn=_n(1<<24),Vn=_n(0);$n.ZERO=Vn;var Kn=_n(0,!0);$n.UZERO=Kn;var Gn=_n(1);$n.ONE=Gn;var Hn=_n(1,!0);$n.UONE=Hn;var Jn=_n(-1);$n.NEG_ONE=Jn;var Zn=On(-1,2147483647,!1);$n.MAX_VALUE=Zn;var Yn=On(-1,-1,!0);$n.MAX_UNSIGNED_VALUE=Yn;var Xn=On(0,-2147483648,!1);$n.MIN_VALUE=Xn;var Qn=$n.prototype;Qn.toInt=function(){return this.unsigned?this.low>>>0:this.low},Qn.toNumber=function(){return this.unsigned?(this.high>>>0)*Un+(this.low>>>0):this.high*Un+(this.low>>>0)},Qn.toString=function(t){if((t=t||10)<2||36<t)throw RangeError("radix");if(this.isZero())return"0";if(this.isNegative()){if(this.eq(Xn)){var e=Rn(t),n=this.div(e),s=n.mul(e).sub(this);return n.toString(t)+s.toInt().toString(t)}return"-"+this.neg().toString(t)}for(var i=Rn(Mn(t,6),this.unsigned),r=this,a="";;){var o=r.div(i),l=(r.sub(o.mul(i)).toInt()>>>0).toString(t);if((r=o).isZero())return l+a;for(;l.length<6;)l="0"+l;a=""+l+a}},Qn.getHighBits=function(){return this.high},Qn.getHighBitsUnsigned=function(){return this.high>>>0},Qn.getLowBits=function(){return this.low},Qn.getLowBitsUnsigned=function(){return this.low>>>0},Qn.getNumBitsAbs=function(){if(this.isNegative())return this.eq(Xn)?64:this.neg().getNumBitsAbs();for(var t=0!=this.high?this.high:this.low,e=31;e>0&&0==(t&1<<e);e--);return 0!=this.high?e+33:e+1},Qn.isZero=function(){return 0===this.high&&0===this.low},Qn.eqz=Qn.isZero,Qn.isNegative=function(){return!this.unsigned&&this.high<0},Qn.isPositive=function(){return this.unsigned||this.high>=0},Qn.isOdd=function(){return 1==(1&this.low)},Qn.isEven=function(){return 0==(1&this.low)},Qn.equals=function(t){return Fn(t)||(t=Pn(t)),(this.unsigned===t.unsigned||this.high>>>31!=1||t.high>>>31!=1)&&(this.high===t.high&&this.low===t.low)},Qn.eq=Qn.equals,Qn.notEquals=function(t){return!this.eq(t)},Qn.neq=Qn.notEquals,Qn.ne=Qn.notEquals,Qn.lessThan=function(t){return this.comp(t)<0},Qn.lt=Qn.lessThan,Qn.lessThanOrEqual=function(t){return this.comp(t)<=0},Qn.lte=Qn.lessThanOrEqual,Qn.le=Qn.lessThanOrEqual,Qn.greaterThan=function(t){return this.comp(t)>0},Qn.gt=Qn.greaterThan,Qn.greaterThanOrEqual=function(t){return this.comp(t)>=0},Qn.gte=Qn.greaterThanOrEqual,Qn.ge=Qn.greaterThanOrEqual,Qn.compare=function(t){if(Fn(t)||(t=Pn(t)),this.eq(t))return 0;var e=this.isNegative(),n=t.isNegative();return e&&!n?-1:!e&&n?1:this.unsigned?t.high>>>0>this.high>>>0||t.high===this.high&&t.low>>>0>this.low>>>0?-1:1:this.sub(t).isNegative()?-1:1},Qn.comp=Qn.compare,Qn.negate=function(){return!this.unsigned&&this.eq(Xn)?Xn:this.not().add(Gn)},Qn.neg=Qn.negate,Qn.add=function(t){Fn(t)||(t=Pn(t));var e=this.high>>>16,n=65535&this.high,s=this.low>>>16,i=65535&this.low,r=t.high>>>16,a=65535&t.high,o=t.low>>>16,l=0,u=0,h=0,c=0;return h+=(c+=i+(65535&t.low))>>>16,u+=(h+=s+o)>>>16,l+=(u+=n+a)>>>16,l+=e+r,On((h&=65535)<<16|(c&=65535),(l&=65535)<<16|(u&=65535),this.unsigned)},Qn.subtract=function(t){return Fn(t)||(t=Pn(t)),this.add(t.neg())},Qn.sub=Qn.subtract,Qn.multiply=function(t){if(this.isZero())return Vn;if(Fn(t)||(t=Pn(t)),Cn)return On(Cn.mul(this.low,this.high,t.low,t.high),Cn.get_high(),this.unsigned);if(t.isZero())return Vn;if(this.eq(Xn))return t.isOdd()?Xn:Vn;if(t.eq(Xn))return this.isOdd()?Xn:Vn;if(this.isNegative())return t.isNegative()?this.neg().mul(t.neg()):this.neg().mul(t).neg();if(t.isNegative())return this.mul(t.neg()).neg();if(this.lt(qn)&&t.lt(qn))return Rn(this.toNumber()*t.toNumber(),this.unsigned);var e=this.high>>>16,n=65535&this.high,s=this.low>>>16,i=65535&this.low,r=t.high>>>16,a=65535&t.high,o=t.low>>>16,l=65535&t.low,u=0,h=0,c=0,p=0;return c+=(p+=i*l)>>>16,h+=(c+=s*l)>>>16,c&=65535,h+=(c+=i*o)>>>16,u+=(h+=n*l)>>>16,h&=65535,u+=(h+=s*o)>>>16,h&=65535,u+=(h+=i*a)>>>16,u+=e*l+n*o+s*a+i*r,On((c&=65535)<<16|(p&=65535),(u&=65535)<<16|(h&=65535),this.unsigned)},Qn.mul=Qn.multiply,Qn.divide=function(t){if(Fn(t)||(t=Pn(t)),t.isZero())throw Error("division by zero");var e,n,s;if(Cn)return this.unsigned||-2147483648!==this.high||-1!==t.low||-1!==t.high?On((this.unsigned?Cn.div_u:Cn.div_s)(this.low,this.high,t.low,t.high),Cn.get_high(),this.unsigned):this;if(this.isZero())return this.unsigned?Kn:Vn;if(this.unsigned){if(t.unsigned||(t=t.toUnsigned()),t.gt(this))return Kn;if(t.gt(this.shru(1)))return Hn;s=Kn}else{if(this.eq(Xn))return t.eq(Gn)||t.eq(Jn)?Xn:t.eq(Xn)?Gn:(e=this.shr(1).div(t).shl(1)).eq(Vn)?t.isNegative()?Gn:Jn:(n=this.sub(t.mul(e)),s=e.add(n.div(t)));if(t.eq(Xn))return this.unsigned?Kn:Vn;if(this.isNegative())return t.isNegative()?this.neg().div(t.neg()):this.neg().div(t).neg();if(t.isNegative())return this.div(t.neg()).neg();s=Vn}for(n=this;n.gte(t);){e=Math.max(1,Math.floor(n.toNumber()/t.toNumber()));for(var i=Math.ceil(Math.log(e)/Math.LN2),r=i<=48?1:Mn(2,i-48),a=Rn(e),o=a.mul(t);o.isNegative()||o.gt(n);)o=(a=Rn(e-=r,this.unsigned)).mul(t);a.isZero()&&(a=Gn),s=s.add(a),n=n.sub(o)}return s},Qn.div=Qn.divide,Qn.modulo=function(t){return Fn(t)||(t=Pn(t)),Cn?On((this.unsigned?Cn.rem_u:Cn.rem_s)(this.low,this.high,t.low,t.high),Cn.get_high(),this.unsigned):this.sub(this.div(t).mul(t))},Qn.mod=Qn.modulo,Qn.rem=Qn.modulo,Qn.not=function(){return On(~this.low,~this.high,this.unsigned)},Qn.and=function(t){return Fn(t)||(t=Pn(t)),On(this.low&t.low,this.high&t.high,this.unsigned)},Qn.or=function(t){return Fn(t)||(t=Pn(t)),On(this.low|t.low,this.high|t.high,this.unsigned)},Qn.xor=function(t){return Fn(t)||(t=Pn(t)),On(this.low^t.low,this.high^t.high,this.unsigned)},Qn.shiftLeft=function(t){return Fn(t)&&(t=t.toInt()),0==(t&=63)?this:t<32?On(this.low<<t,this.high<<t|this.low>>>32-t,this.unsigned):On(0,this.low<<t-32,this.unsigned)},Qn.shl=Qn.shiftLeft,Qn.shiftRight=function(t){return Fn(t)&&(t=t.toInt()),0==(t&=63)?this:t<32?On(this.low>>>t|this.high<<32-t,this.high>>t,this.unsigned):On(this.high>>t-32,this.high>=0?0:-1,this.unsigned)},Qn.shr=Qn.shiftRight,Qn.shiftRightUnsigned=function(t){if(Fn(t)&&(t=t.toInt()),0===(t&=63))return this;var e=this.high;return t<32?On(this.low>>>t|e<<32-t,e>>>t,this.unsigned):On(32===t?e:e>>>t-32,0,this.unsigned)},Qn.shru=Qn.shiftRightUnsigned,Qn.shr_u=Qn.shiftRightUnsigned,Qn.toSigned=function(){return this.unsigned?On(this.low,this.high,!1):this},Qn.toUnsigned=function(){return this.unsigned?this:On(this.low,this.high,!0)},Qn.toBytes=function(t){return t?this.toBytesLE():this.toBytesBE()},Qn.toBytesLE=function(){var t=this.high,e=this.low;return[255&e,e>>>8&255,e>>>16&255,e>>>24,255&t,t>>>8&255,t>>>16&255,t>>>24]},Qn.toBytesBE=function(){var t=this.high,e=this.low;return[t>>>24,t>>>16&255,t>>>8&255,255&t,e>>>24,e>>>16&255,e>>>8&255,255&e]},$n.fromBytes=function(t,e,n){return n?$n.fromBytesLE(t,e):$n.fromBytesBE(t,e)},$n.fromBytesLE=function(t,e){return new $n(t[0]|t[1]<<8|t[2]<<16|t[3]<<24,t[4]|t[5]<<8|t[6]<<16|t[7]<<24,e)},$n.fromBytesBE=function(t,e){return new $n(t[4]<<24|t[5]<<16|t[6]<<8|t[7],t[0]<<24|t[1]<<16|t[2]<<8|t[3],e)};var ts=zn(Tn);const es=ts||s({__proto__:null,default:ts},[Tn]);function ns(t){return es.fromString(t,!0,16)}function ss(t,e){if("string"===e)throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(t)&&(t=os(t)),fn().getBool("DEBUG")&&function(t,e){for(let n=0;n<t.length;n++){const s=t[n];if(isNaN(s)||!isFinite(s))throw Error(`A tensor of type ${e} being uploaded contains ${s}.`)}}(t,e),function(t,e){return t instanceof Float32Array&&"float32"===e||t instanceof Int32Array&&"int32"===e||t instanceof Uint8Array&&"bool"===e}(t,e))return t;if(null==e||"float32"===e||"complex64"===e)return new Float32Array(t);if("int32"===e)return new Int32Array(t);if("bool"===e){const e=new Uint8Array(t.length);for(let n=0;n<e.length;++n)0!==Math.round(t[n])&&(e[n]=1);return e}throw new Error(`Unknown data type ${e}`)}function is(){return fn().platform.now()}function rs(t,e="utf-8"){return e=e||"utf-8",fn().platform.decode(t,e)}function as(t){return null!=fn().platform.isTypedArray?fn().platform.isTypedArray(t):function(t){return t instanceof Float32Array||t instanceof Int32Array||t instanceof Uint8Array||t instanceof Uint8ClampedArray}(t)}function os(t,e=[],n=!1){if(null==e&&(e=[]),"boolean"==typeof t||"number"==typeof t||"string"==typeof t||hn(t)||null==t||as(t)&&n)e.push(t);else if(Array.isArray(t)||as(t))for(let s=0;s<t.length;++s)os(t[s],e,n);else{let s=-1;for(const e of Object.keys(t))/^([1-9]+[0-9]*|0)$/.test(e)&&(s=Math.max(s,Number(e)));for(let i=0;i<=s;i++)os(t[i],e,n)}return e}ns("c3a5c85c97cb3127"),ns("b492b66fbe98f273"),ns("9ae16a3b2f90404f");class ls{constructor(t,e){this.backendTimer=t,this.logger=e,null==e&&(this.logger=new hs)}profileKernel(t,e,n){let s;const i=()=>{s=n()};let r;const a=is();if(this.backendTimer.timerAvailable())r=this.backendTimer.time(i);else{i();for(const t of s)t.dataSync();r=Promise.resolve({kernelMs:is()-a})}if(fn().getBool("CHECK_COMPUTATION_FOR_ERRORS"))for(let e=0;e<s.length;e++){const n=s[e];n.data().then((e=>{us(e,n.dtype,t)}))}return{kernelName:t,outputs:s,inputs:e,timeMs:r.then((t=>t.kernelMs)),extraInfo:r.then((t=>null!=t.getExtraProfileInfo?t.getExtraProfileInfo():""))}}logKernelProfile(t){const{kernelName:e,outputs:n,timeMs:s,inputs:i,extraInfo:r}=t;n.forEach((t=>{Promise.all([t.data(),s,r]).then((n=>{this.logger.logKernelProfile(e,t,n[0],n[1],i,n[2])}))}))}}function us(t,e,n){if("float32"!==e)return!1;for(let e=0;e<t.length;e++){const s=t[e];if(isNaN(s)||!isFinite(s))return console.warn(`Found ${s} in the result of '${n}'`),!0}return!1}class hs{logKernelProfile(t,e,n,s,i,r){const a="number"==typeof s?Ye(`${s}ms`,9):s.error,o=Ye(t,25),l=e.rank,u=e.size,h=Ye(e.shape.toString(),14);let c="";for(const t in i){const n=i[t];if(null!=n){const s=n.shape||e.shape,i=s.length;c+=`${t}: ${i}D ${i>0?s:""} `}}console.log(`%c${o}\t%c${a}\t%c${l}D ${h}\t%c${u}\t%c${c}\t%c${r}`,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")}}function cs(t,e,n,s){const i=sn(e),r=function(t,e,n,s){const i=He(e),r=s[s.length-1],a=new Array(r).fill(0),o=e.length,l="complex64"===n?gs(t):t;if(o>1)for(let t=0;t<i/r;t++){const e=t*r;for(let t=0;t<r;t++)a[t]=Math.max(a[t],ps(l[e+t],0,n).length)}return a}(t,e,n,i),a=e.length,o=fs(t,e,n,i,r),l=["Tensor"];return s&&(l.push(`  dtype: ${n}`),l.push(`  rank: ${a}`),l.push(`  shape: [${e}]`),l.push("  values:")),l.push(o.map((t=>"    "+t)).join("\n")),l.join("\n")}function ps(t,e,n){let s;return s=Array.isArray(t)?`${parseFloat(t[0].toFixed(7))} + ${parseFloat(t[1].toFixed(7))}j`:tn(t)?`'${t}'`:"bool"===n?ds(t):parseFloat(t.toFixed(7)).toString(),Ye(s,e)}function ds(t){return 0===t?"false":"true"}function fs(t,e,n,s,i,r=!0){const a="complex64"===n?2:1,o=e[0],l=e.length;if(0===l){if("complex64"===n){return[ps(gs(t)[0],0,n)]}return"bool"===n?[ds(t[0])]:[t[0].toString()]}if(1===l){if(o>20){const e=3*a;let s=Array.from(t.slice(0,e)),r=Array.from(t.slice((o-3)*a,o*a));return"complex64"===n&&(s=gs(s),r=gs(r)),["["+s.map(((t,e)=>ps(t,i[e],n))).join(", ")+", ..., "+r.map(((t,e)=>ps(t,i[o-3+e],n))).join(", ")+"]"]}return["["+("complex64"===n?gs(t):Array.from(t)).map(((t,e)=>ps(t,i[e],n))).join(", ")+"]"]}const u=e.slice(1),h=s.slice(1),c=s[0]*a,p=[];if(o>20){for(let e=0;e<3;e++){const s=e*c,r=s+c;p.push(...fs(t.slice(s,r),u,n,h,i,!1))}p.push("...");for(let e=o-3;e<o;e++){const s=e*c,r=s+c;p.push(...fs(t.slice(s,r),u,n,h,i,e===o-1))}}else for(let e=0;e<o;e++){const s=e*c,r=s+c;p.push(...fs(t.slice(s,r),u,n,h,i,e===o-1))}const d=2===l?",":"";p[0]="["+(o>0?p[0]+d:"");for(let t=1;t<p.length-1;t++)p[t]=" "+p[t]+d;let f=",\n";for(let t=2;t<l;t++)f+="\n";return p[p.length-1]=" "+p[p.length-1]+"]"+(r?"":f),p}function gs(t){const e=[];for(let n=0;n<t.length;n+=2)e.push([t[n],t[n+1]]);return e}let ms=null,ys=null;class bs{constructor(t,e,n,s){this.kept=!1,this.isDisposedInternal=!1,this.shape=t.slice(),this.dtype=e||"float32",this.size=He(t),this.strides=sn(t),this.dataId=n,this.id=s,this.rankType=this.rank<5?this.rank.toString():"higher"}get rank(){return this.shape.length}async buffer(){const t=await this.data();return ys.buffer(this.shape,this.dtype,t)}bufferSync(){return ys.buffer(this.shape,this.dtype,this.dataSync())}async array(){const t=await this.data();return an(this.shape,t,"complex64"===this.dtype)}arraySync(){return an(this.shape,this.dataSync(),"complex64"===this.dtype)}async data(){this.throwIfDisposed();const t=ms().read(this.dataId);if("string"===this.dtype){const e=await t;try{return e.map((t=>rs(t)))}catch(t){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}}return t}dataToGPU(t){return this.throwIfDisposed(),ms().readToGPU(this.dataId,t)}dataSync(){this.throwIfDisposed();const t=ms().readSync(this.dataId);if("string"===this.dtype)try{return t.map((t=>rs(t)))}catch(t){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return t}async bytes(){this.throwIfDisposed();const t=await ms().read(this.dataId);return"string"===this.dtype?t:new Uint8Array(t.buffer)}dispose(){this.isDisposed||(this.kerasMask&&this.kerasMask.dispose(),ms().disposeTensor(this),this.isDisposedInternal=!0)}get isDisposed(){return this.isDisposedInternal}throwIfDisposed(){if(this.isDisposed)throw new Error("Tensor is disposed.")}print(t=!1){return ys.print(this,t)}clone(){return this.throwIfDisposed(),ys.clone(this)}toString(t=!1){return cs(this.dataSync(),this.shape,this.dtype,t)}cast(t){return this.throwIfDisposed(),ys.cast(this,t)}variable(t=!0,e,n){return this.throwIfDisposed(),ms().makeVariable(this,t,e,n)}}function ws(){return bn("Tensor",(()=>bs))}Object.defineProperty(bs,Symbol.hasInstance,{value:t=>!!t&&null!=t.data&&null!=t.dataSync&&null!=t.throwIfDisposed}),ws();class ks extends bs{constructor(t,e,n,s){super(t.shape,t.dtype,t.dataId,s),this.trainable=e,this.name=n}assign(t){if(t.dtype!==this.dtype)throw new Error(`dtype of the new value (${t.dtype}) and previous value (${this.dtype}) must match`);if(!Je(t.shape,this.shape))throw new Error(`shape of the new value (${t.shape}) and previous value (${this.shape}) must match`);ms().disposeTensor(this),this.dataId=t.dataId,ms().incRef(this,null)}dispose(){ms().disposeVariable(this),this.isDisposedInternal=!0}}var vs,Ss,xs,Ns,Is;Object.defineProperty(ks,Symbol.hasInstance,{value:t=>t instanceof bs&&null!=t.assign&&t.assign instanceof Function}),function(t){t.R0="R0",t.R1="R1",t.R2="R2",t.R3="R3",t.R4="R4",t.R5="R5",t.R6="R6"}(vs||(vs={})),function(t){t.float32="float32",t.int32="int32",t.bool="int32",t.complex64="complex64"}(Ss||(Ss={})),function(t){t.float32="float32",t.int32="int32",t.bool="bool",t.complex64="complex64"}(xs||(xs={})),function(t){t.float32="float32",t.int32="float32",t.bool="float32",t.complex64="complex64"}(Ns||(Ns={})),function(t){t.float32="complex64",t.int32="complex64",t.bool="complex64",t.complex64="complex64"}(Is||(Is={}));const As={float32:Ns,int32:Ss,bool:xs,complex64:Is};function zs(t){return null!=t&&"object"==typeof t&&"texture"in t&&t.texture instanceof WebGLTexture}function Es(t){return"undefined"!=typeof GPUBuffer&&null!=t&&"object"==typeof t&&"buffer"in t&&t.buffer instanceof GPUBuffer}function Ts(t,e){if(t.dtype===e.dtype)return[t,e];const n=function(t,e){if("string"===t||"string"===e){if("string"===t&&"string"===e)return"string";throw new Error(`Can not upcast ${t} with ${e}`)}return As[t][e]}(t.dtype,e.dtype);return[t.cast(n),e.cast(n)]}function Cs(t){const e=[];return $s(t,e,new Set),e}function $s(t,e,n){if(null==t)return;if(t instanceof bs)return void e.push(t);if(s=t,!Array.isArray(s)&&"object"!=typeof s)return;var s;const i=t;for(const t in i){const s=i[t];n.has(s)||(n.add(s),$s(s,e,n))}}function Fs(t){return null!=t.kernelName}class Ds{constructor(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null,get kernelNames(){return Array.from(new Set(this.kernels.map((t=>t.name))))}}}dispose(){for(const t in this.registeredVariables)this.registeredVariables[t].dispose()}}class Ls{constructor(t){this.ENV=t,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new Ds}async ready(){if(null!=this.pendingBackendInit)return this.pendingBackendInit.then((()=>{}));if(null!=this.backendInstance)return;const t=this.getSortedBackends();for(let e=0;e<t.length;e++){const n=t[e];if(await this.initializeBackend(n).success)return void await this.setBackend(n)}throw new Error("Could not initialize any backends, all backend initializations failed.")}get backend(){if(null!=this.pendingBackendInit)throw new Error(`Backend '${this.backendName}' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods`);if(null==this.backendInstance){const{name:t,asyncInit:e}=this.initializeBackendsAndReturnBest();if(e)throw new Error(`The highest priority backend '${t}' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods`);this.setBackend(t)}return this.backendInstance}backendNames(){return Object.keys(this.registryFactory)}findBackend(t){if(!(t in this.registry)){if(!(t in this.registryFactory))return null;{const{asyncInit:e}=this.initializeBackend(t);if(e)return null}}return this.registry[t]}findBackendFactory(t){return t in this.registryFactory?this.registryFactory[t].factory:null}registerBackend(t,e,n=1){return t in this.registryFactory?(wn(`${t} backend was already registered. Reusing existing backend factory.`),!1):(this.registryFactory[t]={factory:e,priority:n},!0)}async setBackend(t){if(null==this.registryFactory[t])throw new Error(`Backend name '${t}' not found in registry`);if(this.backendName=t,null==this.registry[t]){this.backendInstance=null;const{success:e,asyncInit:n}=this.initializeBackend(t);if(!(n?await e:e))return!1}return this.backendInstance=this.registry[t],this.setupRegisteredKernels(),this.profiler=new ls(this.backendInstance),!0}setupRegisteredKernels(){Nn(this.backendName).forEach((t=>{null!=t.setupFunc&&t.setupFunc(this.backendInstance)}))}disposeRegisteredKernels(t){Nn(t).forEach((e=>{null!=e.disposeFunc&&e.disposeFunc(this.registry[t])}))}initializeBackend(t){const e=this.registryFactory[t];if(null==e)throw new Error(`Cannot initialize backend ${t}, no registration found.`);try{const n=e.factory();if(!n||n instanceof class{refCount(t){return Ke("refCount")}incRef(t){return Ke("incRef")}timerAvailable(){return!0}time(t){return Ke("time")}read(t){return Ke("read")}readSync(t){return Ke("readSync")}readToGPU(t,e){return Ke("readToGPU")}numDataIds(){return Ke("numDataIds")}disposeData(t,e){return Ke("disposeData")}write(t,e,n){return Ke("write")}move(t,e,n,s,i){return Ke("move")}createTensorFromGPUData(t,e,n){return Ke("createTensorFromGPUData")}memory(){return Ke("memory")}floatPrecision(){return Ke("floatPrecision")}epsilon(){return 32===this.floatPrecision()?1e-7:1e-4}dispose(){return Ke("dispose")}}||"function"!=typeof n.then)return this.registry[t]=n,{success:!0,asyncInit:!1};{const e=++this.pendingBackendInitId,s=n.then((n=>!(e<this.pendingBackendInitId)&&(this.registry[t]=n,this.pendingBackendInit=null,!0))).catch((n=>(e<this.pendingBackendInitId||(this.pendingBackendInit=null,wn(`Initialization of backend ${t} failed`),wn(n.stack||n.message)),!1)));return this.pendingBackendInit=s,{success:s,asyncInit:!0}}}catch(e){return wn(`Initialization of backend ${t} failed`),wn(e.stack||e.message),{success:!1,asyncInit:!1}}}removeBackend(t){if(!(t in this.registryFactory))throw new Error(`${t} backend not found in registry`);this.backendName===t&&null!=this.pendingBackendInit&&this.pendingBackendInitId++,t in this.registry&&(this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t]),delete this.registryFactory[t],this.backendName===t&&(this.pendingBackendInit=null,this.backendName=null,this.backendInstance=null)}getSortedBackends(){if(0===Object.keys(this.registryFactory).length)throw new Error("No backend found in registry.");return Object.keys(this.registryFactory).sort(((t,e)=>this.registryFactory[e].priority-this.registryFactory[t].priority))}initializeBackendsAndReturnBest(){const t=this.getSortedBackends();for(let e=0;e<t.length;e++){const n=t[e],{success:s,asyncInit:i}=this.initializeBackend(n);if(i||s)return{name:n,asyncInit:i}}throw new Error("Could not initialize any backends, all backend initializations failed.")}moveData(t,e){const n=this.state.tensorInfo.get(e),s=n.backend,i=this.readSync(e),r=s.refCount(e);s.disposeData(e,!0),n.backend=t,t.move(e,i,n.shape,n.dtype,r),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++}tidy(t,e){let n,s=null;if(null==e){if("function"!=typeof t)throw new Error("Please provide a function to tidy()");e=t}else{if("string"!=typeof t&&!(t instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if("function"!=typeof e)throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");s=t}return this.scopedRun((()=>this.startScope(s)),(()=>this.endScope(n)),(()=>(n=e(),n instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),n)))}scopedRun(t,e,n){t();try{const t=n();return e(),t}catch(t){throw e(),t}}nextTensorId(){return Ls.nextTensorId++}nextVariableId(){return Ls.nextVariableId++}clone(t){const e=_s.runKernel(Ie,{x:t}),n={x:t};return this.addTapeNode(this.state.activeScope.name,n,[e],(t=>({x:()=>{const e={x:t},n={dtype:"float32"};return _s.runKernel(de,e,n)}})),[],{}),e}runKernel(t,e,n){null==this.backendName&&this.backend;if(!(null!=Sn(t,this.backendName)))throw new Error(`Kernel '${t}' not registered for backend '${this.backendName}'`);return this.runKernelFunc({kernelName:t,inputs:e,attrs:n})}shouldCheckForMemLeaks(){return this.ENV.getBool("IS_TEST")}checkKernelForMemLeak(t,e,n){const s=this.backend.numDataIds();let i=0;n.forEach((t=>{i+="complex64"===t.dtype?3:1}));const r=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],a=s-e-i-r;if(a>0)throw new Error(`Backend '${this.backendName}' has an internal memory leak (${a} data ids) after running '${t}'`)}runKernelFunc(t){let e,n=[];const s=this.isTapeOn(),i=this.state.numBytes,r=this.state.numTensors;let a,o;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0),null==this.backendName&&this.backend;const l=Fs(t)?t.kernelName:null!=this.state.activeScope?this.state.activeScope.name:"";if(Fs(t)){const{kernelName:e,inputs:i,attrs:r}=t;null==this.backendName&&this.backend;const l=Sn(e,this.backendName);Ge(null!=l,(()=>`Cannot find registered kernel '${e}' for backend '${this.backendName}'`)),a=()=>{const t=this.backend.numDataIds();o=l.kernelFunc({inputs:i,attrs:r,backend:this.backend});const a=Array.isArray(o)?o:[o];this.shouldCheckForMemLeaks()&&this.checkKernelForMemLeak(e,t,a);const u=a.map((t=>null!=t.rank?t:this.makeTensorFromTensorInfo(t)));if(s){const t=this.getTensorsForGradient(e,i,u);n=this.saveTensorsForBackwardMode(t)}return u}}else{const{forwardFunc:e}=t,i=t=>{s&&(n=t.map((t=>this.keep(this.clone(t)))))};a=()=>{const t=this.backend.numDataIds();o=this.tidy((()=>e(this.backend,i)));const n=Array.isArray(o)?o:[o];return this.shouldCheckForMemLeaks()&&this.checkKernelForMemLeak(l,t,n),n}}const{inputs:u,attrs:h}=t,c=Fs(t)?null:t.backwardsFunc;let p;return this.scopedRun((()=>this.state.kernelDepth++),(()=>this.state.kernelDepth--),(()=>{this.ENV.getBool("DEBUG")||this.state.profiling?(p=this.profiler.profileKernel(l,u,(()=>a())),this.ENV.getBool("DEBUG")&&this.profiler.logKernelProfile(p),e=p.outputs):e=a()})),s&&this.addTapeNode(l,u,e,c,n,h),this.state.profiling&&this.state.activeProfile.kernels.push({name:l,bytesAdded:this.state.numBytes-i,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-r,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(u).map((t=>null!=u[t]?u[t].shape:null)),outputShapes:e.map((t=>t.shape)),kernelTimeMs:p.timeMs,extraInfo:p.extraInfo}),Array.isArray(o)?e:e[0]}saveTensorsForBackwardMode(t){return t.map((t=>this.keep(this.clone(t))))}getTensorsForGradient(t,e,n){const s=xn(t);if(null!=s){const t=s.inputsToSave||[],i=s.outputsToSave||[];let r;s.saveAllInputs?(Ge(Array.isArray(e),(()=>"saveAllInputs is true, expected inputs to be an array.")),r=Object.keys(e).map((t=>e[t]))):r=t.map((t=>e[t]));const a=n.filter(((t,e)=>i[e]));return r.concat(a)}return[]}makeTensor(t,e,n,s){if(null==t)throw new Error("Values passed to engine.makeTensor() are null");n=n||"float32",s=s||this.backend;let i=t;"string"===n&&tn(t[0])&&(i=t.map((t=>function(t,e="utf-8"){return e=e||"utf-8",fn().platform.encode(t,e)}(t))));const r=s.write(i,e,n),a=new bs(e,n,r,this.nextTensorId());if(this.trackTensor(a,s),"string"===n){const t=this.state.tensorInfo.get(r),e=function(t){if(null==t)return 0;let e=0;return t.forEach((t=>e+=t.length)),e}(i);this.state.numBytes+=e-t.bytes,t.bytes=e}return a}makeTensorFromDataId(t,e,n,s){const i={dataId:t,shape:e,dtype:n=n||"float32"};return this.makeTensorFromTensorInfo(i,s)}makeTensorFromTensorInfo(t,e){const{dataId:n,shape:s,dtype:i}=t,r=new bs(s,i,n,this.nextTensorId());return this.trackTensor(r,e),r}makeVariable(t,e=!0,n,s){n=n||this.nextVariableId().toString(),null!=s&&s!==t.dtype&&(t=t.cast(s));const i=new ks(t,e,n,this.nextTensorId());if(null!=this.state.registeredVariables[i.name])throw new Error(`Variable with name ${i.name} was already registered`);return this.state.registeredVariables[i.name]=i,this.incRef(i,this.backend),i}trackTensor(t,e){this.state.numTensors++,"string"===t.dtype&&this.state.numStringTensors++;let n=0;"complex64"!==t.dtype&&"string"!==t.dtype&&(n=t.size*Qe(t.dtype)),this.state.numBytes+=n,this.state.tensorInfo.has(t.dataId)||(this.state.numDataBuffers++,this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:n})),t instanceof ks||this.track(t)}incRef(t,e){this.trackTensor(t,e),this.backend.incRef(t.dataId)}removeDataId(t,e){this.state.tensorInfo.has(t)&&this.state.tensorInfo.get(t).backend===e&&(this.state.tensorInfo.delete(t),this.state.numDataBuffers--)}disposeTensor(t){if(!this.state.tensorInfo.has(t.dataId))return;const e=this.state.tensorInfo.get(t.dataId);if(this.state.numTensors--,"string"===t.dtype&&(this.state.numStringTensors--,this.state.numBytes-=e.bytes),"complex64"!==t.dtype&&"string"!==t.dtype){const e=t.size*Qe(t.dtype);this.state.numBytes-=e}e.backend.disposeData(t.dataId)&&this.removeDataId(t.dataId,e.backend)}disposeVariables(){for(const t in this.state.registeredVariables){const e=this.state.registeredVariables[t];this.disposeVariable(e)}}disposeVariable(t){this.disposeTensor(t),null!=this.state.registeredVariables[t.name]&&delete this.state.registeredVariables[t.name]}memory(){const 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,null==t.reasons&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t}async profile(t){this.state.profiling=!0;const e=this.state.numBytes,n=this.state.numTensors;this.state.activeProfile.kernels=[],this.state.activeProfile.result=await t(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max(...this.state.activeProfile.kernels.map((t=>t.totalBytesSnapshot))),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-n;for(const t of this.state.activeProfile.kernels)t.kernelTimeMs=await t.kernelTimeMs,t.extraInfo=await t.extraInfo;return this.state.activeProfile}isTapeOn(){return this.state.gradientDepth>0&&0===this.state.kernelDepth}addTapeNode(t,e,n,s,i,r){const a={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:n,saved:i},o=xn(t);null!=o&&(s=o.gradFunc),null!=s&&(a.gradient=t=>(t=t.map(((t,e)=>{if(null==t){const t=n[e],s=ln(t.size,t.dtype);return this.makeTensor(s,t.shape,t.dtype)}return t})),s(t.length>1?t:t[0],i,r))),this.state.activeTape.push(a)}keep(t){return t.kept=!0,t}startTape(){0===this.state.gradientDepth&&(this.state.activeTape=[]),this.state.gradientDepth++}endTape(){this.state.gradientDepth--}startScope(t){const e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e}endScope(t){const e=Cs(t),n=new Set(e.map((t=>t.id)));for(let t=0;t<this.state.activeScope.track.length;t++){const e=this.state.activeScope.track[t];e.kept||n.has(e.id)||e.dispose()}const s=this.state.scopeStack.pop();this.state.activeScope=0===this.state.scopeStack.length?null:this.state.scopeStack[this.state.scopeStack.length-1],e.forEach((t=>{t.kept||t.scopeId!==s.id||this.track(t)}))}gradients(t,e,n,s=!1){if(Ge(e.length>0,(()=>"gradients() received an empty list of xs.")),null!=n&&"float32"!==n.dtype)throw new Error(`dy must have 'float32' dtype, but has '${n.dtype}'`);const i=this.scopedRun((()=>this.startTape()),(()=>this.endTape()),(()=>this.tidy("forward",t)));Ge(i instanceof bs,(()=>"The result y returned by f() must be a tensor."));const r=function(t,e,n){const s={},i={};for(let t=0;t<e.length;t++)s[e[t].id]=!0;for(let n=0;n<t.length;n++){const r=t[n],a=r.inputs;for(const t in a){const n=a[t];let o=!1;for(let t=0;t<e.length;t++)if(s[n.id]){r.outputs.forEach((t=>s[t.id]=!0)),o=!0,i[r.id]=!0;break}if(o)break}}const r={};r[n.id]=!0;const a={};for(let e=t.length-1;e>=0;e--){const n=t[e],s=n.inputs;for(let t=0;t<n.outputs.length;t++)if(r[n.outputs[t].id]){for(const t in s)r[s[t].id]=!0,a[n.id]=!0;break}}const o=[];for(let e=0;e<t.length;e++){const n=t[e];if(i[n.id]&&a[n.id]){const t={};for(const e in n.inputs){const i=n.inputs[e];s[i.id]&&(t[e]=i)}const e=Object.assign({},n);e.inputs=t,e.outputs=n.outputs,o.push(e)}}return o}(this.state.activeTape,e,i);if(!s&&0===r.length&&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",(()=>{const t={};t[i.id]=null==n?function(t){const e=on(He(t),"float32");return _s.makeTensor(e,t,"float32")}(i.shape):n,function(t,e,n,s){for(let i=e.length-1;i>=0;i--){const r=e[i],a=[];if(r.outputs.forEach((e=>{const n=t[e.id];null!=n?a.push(n):a.push(null)})),null==r.gradient)throw new Error(`Cannot compute gradient: gradient function not found for ${r.kernelName}.`);const o=r.gradient(a);for(const e in r.inputs){if(!(e in o))throw new Error(`Cannot backprop through input ${e}. Available gradients found: ${Object.keys(o)}.`);const i=n((()=>o[e]()));if("float32"!==i.dtype)throw new Error(`Error in gradient for op ${r.kernelName}. The gradient of input ${e} must have 'float32' dtype, but has '${i.dtype}'`);const a=r.inputs[e];if(!Je(i.shape,a.shape))throw new Error(`Error in gradient for op ${r.kernelName}. The gradient of input '${e}' has shape '${i.shape}', which does not match the shape of the input '${a.shape}'`);if(null==t[a.id])t[a.id]=i;else{const e=t[a.id];t[a.id]=s(e,i),e.dispose()}}}}(t,r,(t=>this.tidy(t)),Rs);const s=e.map((e=>t[e.id]));return 0===this.state.gradientDepth&&(this.state.activeTape.forEach((t=>{for(const e of t.saved)e.dispose()})),this.state.activeTape=null),{value:i,grads:s}}))}customGrad(t){return Ge(nn(t),(()=>"The f passed in customGrad(f) must be a function.")),(...e)=>{let n;Ge(e.every((t=>t instanceof bs)),(()=>"The args passed in customGrad(f)(x1, x2,...) must all be tensors"));const s={};e.forEach(((t,e)=>{s[e]=t}));return this.runKernelFunc({forwardFunc:(s,i)=>(n=t(...e,i),Ge(n.value instanceof bs,(()=>"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor")),Ge(nn(n.gradFunc),(()=>"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function.")),n.value),backwardsFunc:(t,s)=>{const i=n.gradFunc(t,s),r=Array.isArray(i)?i:[i];Ge(r.length===e.length,(()=>"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns the same number of tensors as inputs passed to f(...).")),Ge(r.every((t=>t instanceof bs)),(()=>"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns a list of only tensors."));const a={};return r.forEach(((t,e)=>{a[e]=()=>t})),a},inputs:s})}}readSync(t){return this.state.tensorInfo.get(t).backend.readSync(t)}read(t){return this.state.tensorInfo.get(t).backend.read(t)}readToGPU(t,e){return this.state.tensorInfo.get(t).backend.readToGPU(t,e)}async time(t){const e=is(),n=await this.backend.time(t);return n.wallMs=is()-e,n}track(t){return null!=this.state.activeScope&&(t.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(t)),t}get registeredVariables(){return this.state.registeredVariables}reset(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new Ds;for(const t in this.registry)this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null}}Ls.nextTensorId=0,Ls.nextVariableId=0;const _s=function(){const t=yn();if(null==t._tfengine){const e=new pn(t);t._tfengine=new Ls(e)}var e;return e=t._tfengine.ENV,mn=e,ms=()=>t._tfengine,t._tfengine}();function Rs(t,e){const n={a:t,b:e};return _s.runKernel(he,n)}function Os(t,e){let n=t;if(as(t))return"string"===e?[]:[t.length];if(zs(t)){const e=t.channels||"RGBA";return[t.height,t.width*e.length]}if(Es(t))return[t.buffer.size/(null==e?4:Qe(e))];if(!Array.isArray(t))return[];const s=[];for(;Array.isArray(n)||as(n)&&"string"!==e;)s.push(n.length),n=n[0];return Array.isArray(t)&&fn().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&Ms(t,s,[]),s}function Ms(t,e,n){if(n=n||[],!Array.isArray(t)&&!as(t))return void Ge(0===e.length,(()=>`Element arr[${n.join("][")}] is a primitive, but should be an array/TypedArray of ${e[0]} elements`));Ge(e.length>0,(()=>`Element arr[${n.join("][")}] should be a primitive, but is an array of ${t.length} elements`)),Ge(t.length===e[0],(()=>`Element arr[${n.join("][")}] should have ${e[0]} elements, but has ${t.length} elements`));const s=e.slice(1);for(let e=0;e<t.length;++e)Ms(t[e],s,n.concat(e))}function Bs(t,e,n,s){if("string_or_numeric"!==t){if(null==t)throw new Error("Expected dtype cannot be null.");if("numeric"!==t&&t!==e||"numeric"===t&&"string"===e)throw new Error(`Argument '${n}' passed to '${s}' must be ${t} tensor, but got ${e} tensor`)}}function Ps(t,e,n,s="numeric"){if(t instanceof ws())return Bs(s,t.dtype,e,n),t;let i=en(t);if("string"!==i&&["bool","int32","float32"].indexOf(s)>=0&&(i=s),Bs(s,i,e,n),null==t||!as(t)&&!Array.isArray(t)&&"number"!=typeof t&&"boolean"!=typeof t&&"string"!=typeof t){const s=null==t?"null":t.constructor.name;throw new Error(`Argument '${e}' passed to '${n}' must be a Tensor or TensorLike, but got '${s}'`)}const r=Os(t,i);as(t)||Array.isArray(t)||(t=[t]);const a="string"!==i?ss(t,i):os(t,[],!0);return _s.makeTensor(a,r,i)}function Us(t,e,n,s="numeric"){if(!Array.isArray(t))throw new Error(`Argument ${e} passed to ${n} must be a \`Tensor[]\` or \`TensorLike[]\``);return t.map(((t,i)=>Ps(t,`${e}[${i}]`,n,s)))}function Ws(t){const e=Object.keys(t);if(1!==e.length)throw new Error(`Please provide an object with a single key (operation name) mapping to a function. Got an object with ${e.length} keys.`);let n=e[0];const s=t[n];n.endsWith("_")&&(n=n.substring(0,n.length-1)),n+="__op";const i=(...t)=>{_s.startScope(n);try{const e=s(...t);return hn(e)&&console.error("Cannot return a Promise inside of tidy."),_s.endScope(e),e}catch(t){throw _s.endScope(null),t}};return Object.defineProperty(i,"name",{value:n,configurable:!0}),i}const js=Ws({cast_:function(t,e){const n=Ps(t,"x","cast");if(!function(t){return"bool"===t||"complex64"===t||"float32"===t||"int32"===t||"string"===t}(e))throw new Error(`Failed to cast to unknown dtype ${e}`);if("string"===e&&"string"!==n.dtype||"string"!==e&&"string"===n.dtype)throw new Error("Only strings can be casted to strings");const s={x:n},i={dtype:e};return _s.runKernel(de,s,i)}});const qs=Ws({mul_:function(t,e){let n=Ps(t,"a","mul"),s=Ps(e,"b","mul");[n,s]=Ts(n,s);const i={a:n,b:s};return _s.runKernel(ze,i)}});const Vs=Ws({step_:function(t,e=0){const n={x:Ps(t,"x","step")},s={alpha:e};return _s.runKernel(Ve,n,s)}}),Ks={kernelName:"Abs",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(t,Vs(js(n,"float32"),-1))}}};const Gs=Ws({floorDiv_:function(t,e){let n=Ps(t,"a","floorDiv"),s=Ps(e,"b","floorDiv");[n,s]=Ts(n,s);const i={a:n,b:s};return _s.runKernel(Se,i)}});const Hs=Ws({div_:function(t,e){let n=Ps(t,"a","div"),s=Ps(e,"b","div");if([n,s]=Ts(n,s),"int32"===n.dtype&&"int32"===s.dtype)return Gs(n,s);const i={a:n,b:s};return _s.runKernel(we,i,{})}});const Js=Ws({neg_:function(t){const e={x:Ps(t,"x","neg")};return _s.runKernel("Neg",e)}});function Zs(t,e){if((as(t)&&"string"!==e||Array.isArray(t))&&"complex64"!==e)throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if("string"===e&&as(t)&&!(t instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");return function(t,e,n,s){if(null==s)s=en(t);else if("complex64"===s)throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(Es(t)||zs(t)){if("float32"!==s&&"int32"!==s)throw new Error(`Creating tensor from GPU data only supports 'float32'|'int32' dtype, while the dtype is ${s}.`);return _s.backend.createTensorFromGPUData(t,e||n,s)}if(!as(t)&&!Array.isArray(t)&&"number"!=typeof t&&"boolean"!=typeof t&&"string"!=typeof t)throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(null!=e){un(e);const t=He(e),s=He(n);Ge(t===s,(()=>`Based on the provided shape, [${e}], the tensor should have ${t} values but has ${s}`));for(let t=0;t<n.length;++t){const s=n[t],i=t!==n.length-1||s!==He(e.slice(t));Ge(n[t]===e[t]||!i,(()=>`Error creating a new Tensor. Inferred shape (${n}) does not match the provided shape (${e}). `))}}return as(t)||Array.isArray(t)||(t=[t]),e=e||n,t="string"!==s?ss(t,s):os(t,[],!0),_s.makeTensor(t,e,s)}(t,[],[],e)}const Ys=Ws({sqrt_:function(t){const e={x:Ps(t,"x","sqrt","float32")};return _s.runKernel(Oe,e)}});const Xs=Ws({square_:function(t){const e=Ps(t,"x","square");return _s.runKernel("Square",{x:e},{})}});const Qs=Ws({sub_:function(t,e){let n=Ps(t,"a","sub"),s=Ps(e,"b","sub");[n,s]=Ts(n,s);const i={a:n,b:s};return _s.runKernel("Sub",i)}}),ti={kernelName:"Acos",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>{const e=Xs(js(n,"float32")),s=Ys(Qs(Zs(1),e));return Js(Hs(t,s))}}}},ei={kernelName:"Acosh",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>{const e=Ys(Qs(Xs(js(n,"float32")),1));return Hs(t,e)}}}};function ni(t,e){const n=[];for(let s=0;s<e.length;s++){const i=t[t.length-s-1],r=e.length-s-1,a=e[r];(null==i||1===i&&a>1)&&n.unshift(r)}return n}function si(t,e){const n=Math.max(t.length,e.length),s=new Array(n);for(let i=0;i<n;i++){let r=t[t.length-i-1];null==r&&(r=1);let a=e[e.length-i-1];if(null==a&&(a=1),1===r)s[n-i-1]=a;else if(1===a)s[n-i-1]=r;else{if(r!==a){throw Error(`Operands could not be broadcast together with shapes ${t} and ${e}.`)}s[n-i-1]=r}}return s}const ii=Ws({reshape_:function(t,e){const n={x:Ps(t,"x","reshape","string_or_numeric")},s={shape:e};return _s.runKernel(Ce,n,s)}});const ri=Ws({sum_:function(t,e=null,n=!1){let s=Ps(t,"x","sum");"bool"===s.dtype&&(s=js(s,"int32"));const i={x:s},r={axis:e,keepDims:n};return _s.runKernel("Sum",i,r)}}),ai={kernelName:he,inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e,i=si(n.shape,s.shape);return{a:()=>{let e=t;const s=ni(n.shape,i);return s.length>0&&(e=ri(e,s)),ii(e,n.shape)},b:()=>{let e=t;const n=ni(s.shape,i);return n.length>0&&(e=ri(e,n)),ii(e,s.shape)}}}},oi={kernelName:"AddN",saveAllInputs:!0,gradFunc:(t,e)=>{const n={};return e.forEach(((e,s)=>{n[s]=()=>t.clone()})),n}};const li=Ws({zerosLike_:function(t){const e={x:Ps(t,"x","zerosLike")};return _s.runKernel(qe,e)}}),ui={kernelName:"ArgMax",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>li(n)}}},hi={kernelName:"ArgMin",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>li(n)}}},ci={kernelName:"Asin",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Hs(t,Ys(Qs(Zs(1),Xs(js(n,"float32")))))}}};const pi=Ws({add_:function(t,e){let n=Ps(t,"a","add"),s=Ps(e,"b","add");[n,s]=Ts(n,s);const i={a:n,b:s};return _s.runKernel(he,i)}}),di={kernelName:"Asinh",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>{const e=Ys(pi(Zs(1),Xs(js(n,"float32"))));return Hs(t,e)}}}},fi={kernelName:"Atan2",inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e,i=si(n.shape,s.shape);return{a:()=>{const e=pi(Xs(n),Xs(s));let r=qs(t,Hs(s,e));const a=ni(n.shape,i);return a.length>0&&(r=ri(r,a)),ii(r,n.shape)},b:()=>{const e=pi(Xs(n),Xs(s));let r=Js(qs(t,Hs(n,e)));const a=ni(s.shape,i);return a.length>0&&(r=ri(r,a)),ii(r,s.shape)}}}},gi={kernelName:"Atan",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Hs(t,pi(Xs(js(n,"float32")),1))}}},mi={kernelName:"Atanh",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Hs(t,Qs(Zs(1),Xs(js(n,"float32"))))}}};function yi(t){return"number"==typeof t?[t,t,t]:2===t.length?[t[0],t[1],1]:t}function bi(t){const[e,n,s]=yi(t);return 1===e&&1===n&&1===s}function wi(t,e){return bi(t)||bi(e)}function ki(t){return yi(t).every((t=>t>0))}function vi(t,e,n){if(null!=n){if("string"==typeof e)throw Error(`Error in ${t}: pad must be an integer when using dimRoundingMode ${n} but got pad ${e}.`);if("number"==typeof e)Ge(Ze(e),(()=>`Error in ${t}: pad must be an integer when using dimRoundingMode ${n} but got pad ${e}.`));else{if("object"!=typeof e)throw Error(`Error in ${t}: Unknown padding parameter: ${e}`);e.forEach((e=>{e.forEach((e=>{Ge(Ze(e),(()=>`Error in ${t}: pad must be an integer when using dimRoundingMode ${n} but got pad ${e}.`))}))}))}}}const Si=Ws({avgPool3dGrad_:function(t,e,n,s,i,r){const a=Ps(t,"dy","avgPool3dGrad"),o=Ps(e,"input","avgPool3dGrad");let l=a,u=o,h=!1;4===o.rank&&(h=!0,l=ii(a,[1,a.shape[0],a.shape[1],a.shape[2],a.shape[3]]),u=ii(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),Ge(5===l.rank,(()=>`Error in avgPool3dGrad: dy must be rank 5 but got rank ${l.rank}.`)),Ge(5===u.rank,(()=>`Error in avgPool3dGrad: input must be rank 5 but got rank ${u.rank}.`)),vi("avgPool3dGrad",i,r);const c={dy:l,input:u},p={filterSize:n,strides:s,pad:i,dimRoundingMode:r},d=_s.runKernel("AvgPool3DGrad",c,p);return h?ii(d,[d.shape[1],d.shape[2],d.shape[3],d.shape[4]]):d}}),xi={kernelName:"AvgPool3D",inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{filterSize:i,strides:r,pad:a,dimRoundingMode:o}=n;return{x:()=>Si(t,s,i,r,a,o)}}};const Ni=Ws({avgPoolGrad_:function(t,e,n,s,i){const r=Ps(t,"dy","avgPoolGrad"),a=Ps(e,"input","avgPoolGrad");Ge(a.rank===r.rank,(()=>`Rank of input (${a.rank}) does not match rank of dy (${r.rank})`));let o=a,l=r,u=!1;3===a.rank&&(u=!0,o=ii(a,[1,a.shape[0],a.shape[1],a.shape[2]]),l=ii(r,[1,r.shape[0],r.shape[1],r.shape[2]])),Ge(4===l.rank,(()=>`Error in avgPoolGrad: dy must be rank 4 but got rank ${l.rank}.`)),Ge(4===o.rank,(()=>`Error in avgPoolGrad: input must be rank 4 but got rank ${o.rank}.`));const h={dy:l,input:o},c={filterSize:n,strides:s,pad:i},p=_s.runKernel("AvgPoolGrad",h,c);return u?ii(p,[p.shape[1],p.shape[2],p.shape[3]]):p}}),Ii={kernelName:"AvgPool",inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{filterSize:i,strides:r,pad:a}=n;return{x:()=>Ni(t,s,i,r,a)}}};const Ai=Ws({matMul_:function(t,e,n=!1,s=!1){let i=Ps(t,"a","matMul"),r=Ps(e,"b","matMul");[i,r]=Ts(i,r);const a={a:i,b:r},o={transposeA:n,transposeB:s};return _s.runKernel(ce,a,o)}}),zi={kernelName:ce,inputsToSave:["a","b"],gradFunc:(t,e,n)=>{const[s,i]=e,{transposeA:r,transposeB:a}=n;return r||a?!r&&a?{a:()=>Ai(t,i,!1,!1),b:()=>Ai(t,s,!0,!1)}:r&&!a?{a:()=>Ai(i,t,!1,!0),b:()=>Ai(s,t,!1,!1)}:{a:()=>Ai(i,t,!0,!0),b:()=>Ai(t,s,!0,!0)}:{a:()=>Ai(t,i,!1,!0),b:()=>Ai(s,t,!0,!1)}}};const Ei=Ws({spaceToBatchND_:function(t,e,n){const s=Ps(t,"x","spaceToBatchND");Ge(s.rank>=1+e.length,(()=>`input rank ${s.rank} should be > than [blockShape] ${e.length}`)),Ge(n.length===e.length,(()=>`paddings.shape[0] ${n.length} must be equal to [blockShape] ${e.length}`)),Ge(s.shape.reduce(((t,s,i)=>i>0&&i<=e.length?t&&(s+n[i-1][0]+n[i-1][1])%e[i-1]==0:t),!0),(()=>`input spatial dimensions ${s.shape.slice(1)} with paddings ${n.toString()} must be divisible by blockShapes ${e.toString()}`));const i={x:s},r={blockShape:e,paddings:n};return _s.runKernel(Me,i,r)}}),Ti={kernelName:pe,gradFunc:(t,e,n)=>{const{blockShape:s,crops:i}=n;return{x:()=>Ei(t,s,i)}}},Ci={kernelName:"BroadcastTo",gradFunc:(t,e,n)=>{const s=n,i=s.inputShape,r=s.shape,a=Array.from(r);for(let t=i.length-1;t>=0;t--)if(i[t]===r[t])a[t]=1;else if(1!==i[t])throw new Error(`broadcastTo(): [${i}] cannot be broadcast to [${r}].`);const o=[];for(let t=0;t<a.length;t++)a[t]>1&&o.push(t);return{x:()=>ri(t,o,!0)}}},$i={kernelName:de,gradFunc:t=>({x:()=>t.clone()})},Fi={kernelName:"Ceil",gradFunc:t=>({x:()=>li(t)})};const Di=Ws({greaterEqual_:function(t,e){let n=Ps(t,"a","greaterEqual","string_or_numeric"),s=Ps(e,"b","greaterEqual","string_or_numeric");[n,s]=Ts(n,s),si(n.shape,s.shape);const i={a:n,b:s};return _s.runKernel(Ne,i)}});const Li=Ws({lessEqual_:function(t,e){let n=Ps(t,"a","lessEqual","string_or_numeric"),s=Ps(e,"b","lessEqual","string_or_numeric");[n,s]=Ts(n,s),si(n.shape,s.shape);const i={a:n,b:s};return _s.runKernel("LessEqual",i)}});const _i=Ws({logicalAnd_:function(t,e){const n=Ps(t,"a","logicalAnd","bool"),s=Ps(e,"b","logicalAnd","bool");si(n.shape,s.shape);const i={a:n,b:s};return _s.runKernel("LogicalAnd",i)}});const Ri=Ws({clone_:function(t){const e={x:Ps(t,"x","clone","string_or_numeric")};return _s.runKernel(Ie,e)}});const Oi=Ws({broadcastTo_:function(t,e){let n=Ps(t,"broadcastTo","x");const s=n.shape;if(un(e),e.length<n.rank)throw new Error(`broadcastTo(): shape.length=${e.length} < input.rank=${n.rank}.`);if(e.length>n.rank){const t=n.shape.slice();for(;t.length<e.length;)t.unshift(1);n=ii(n,t)}const i=n.shape,r=Array.from(e);for(let t=e.length-1;t>=0;t--)if(i[t]===e[t])r[t]=1;else if(1!==n.shape[t])throw new Error(`broadcastTo(): [${s}] cannot be broadcast to [${e}].`);if(0===r.map(((t,e)=>t>1?e:-1)).filter((t=>t>=0)).length)return Ri(n);const a={x:n},o={reps:r};return _s.runKernel(Pe,a,o)}});const Mi=Ws({where_:function(t,e,n){const s=Ps(e,"a","where"),i=Ps(n,"b","where"),r=Ps(t,"condition","where","bool"),a=si(si(r.shape,s.shape),i.shape),o={condition:Oi(r,a),t:Oi(s,a),e:Oi(i,a)};return _s.runKernel(De,o)}}),Bi={kernelName:"ClipByValue",inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{clipValueMin:i,clipValueMax:r}=n;return{x:()=>Mi(_i(Di(s,i),Li(s,r)),t,li(t))}}},Pi={kernelName:"ComplexAbs",inputsToSave:["x"],gradFunc:Ks.gradFunc};const Ui=Ws({split_:function(t,e,n=0){const s={x:Ps(t,"x","split")},i={numOrSizeSplits:e,axis:n};return _s.runKernel(Be,s,i)}}),Wi={kernelName:fe,saveAllInputs:!0,gradFunc:(t,e,n)=>{const s=e.map((t=>t.shape)),{axis:i}=n,r=Xe(i,e[0].shape)[0],a=s.map((t=>t[r]));return Ui(t,a,r).map((t=>()=>t))}};const ji=Ws({conv2DBackpropFilter_:function(t,e,n,s,i,r="NHWC",a){let o=t;3===t.rank&&(o=ii(t,[1,t.shape[0],t.shape[1],t.shape[2]]));let l=e;3===l.rank&&(l=ii(e,[1,e.shape[0],e.shape[1],e.shape[2]])),Ge(4===o.rank,(()=>`Error in conv2dDerFilter: input must be rank 4, but got shape ${o.shape}.`)),Ge(4===l.rank,(()=>`Error in conv2dDerFilter: dy must be rank 4, but got shape ${l.shape}.`)),Ge(4===n.length,(()=>`Error in conv2dDerFilter: filterShape must be length 4, but got ${n}.`));const u="NHWC"===r?o.shape[3]:o.shape[1],h="NHWC"===r?l.shape[3]:l.shape[1];Ge(u===n[2],(()=>`Error in conv2dDerFilter: depth of input ${u}) must match input depth in filter (${n[2]}.`)),Ge(h===n[3],(()=>`Error in conv2dDerFilter: depth of dy (${h}) must match output depth for filter (${n[3]}).`)),vi("conv2dDerFilter",i,a);const c={x:o,dy:l},p={strides:s,pad:i,dataFormat:r,dimRoundingMode:a,filterShape:n};return _s.runKernel("Conv2DBackpropFilter",c,p)}});const qi=Ws({conv2DBackpropInput_:function(t,e,n,s,i,r="NHWC",a){Ge(t.length===e.rank,(()=>`Length of inShape (${t.length}) and rank of dy (${e.rank}) must match`));let o=t,l=e,u=!1;3===e.rank&&(u=!0,l=ii(e,[1,e.shape[0],e.shape[1],e.shape[2]]),o=[1,t[0],t[1],t[2]]),Ge(4===o.length,(()=>`Error in conv2dDerInput: inShape must be length 4, but got length ${o.length}.`)),Ge(4===l.rank,(()=>`Error in conv2dDerInput: dy must be rank 4, but got rank ${l.rank}`)),Ge(4===n.rank,(()=>`Error in conv2dDerInput: filter must be rank 4, but got rank ${n.rank}`));const h="NHWC"===r?o[3]:o[1],c="NHWC"===r?l.shape[3]:l.shape[1];Ge(h===n.shape[2],(()=>`Error in conv2dDerInput: depth of input (${h}) must match input depth for filter ${n.shape[2]}.`)),Ge(c===n.shape[3],(()=>`Error in conv2dDerInput: depth of output (${c}) must match output depth for filter ${n.shape[3]}.`)),vi("conv2dDerInput",i,a);const p={dy:l,filter:n},d={strides:s,pad:i,dataFormat:r,dimRoundingMode:a,inputShape:o},f=_s.runKernel(me,p,d);return u?ii(f,[f.shape[1],f.shape[2],f.shape[3]]):f}}),Vi={kernelName:ge,inputsToSave:["x","filter"],gradFunc:(t,e,n)=>{const[s,i]=e,{dilations:r,strides:a,pad:o,dataFormat:l}=n;return Ge(bi(r),(()=>`Error in gradient of conv2D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${r}'`)),{x:()=>qi(s.shape,t,i,a,o,l),filter:()=>ji(s,t,i.shape,a,o,l)}}};const Ki=Ws({conv2d_:function(t,e,n,s,i="NHWC",r=[1,1],a){const o=Ps(t,"x","conv2d","float32"),l=Ps(e,"filter","conv2d","float32");let u=o,h=!1;3===o.rank&&(h=!0,u=ii(o,[1,o.shape[0],o.shape[1],o.shape[2]])),Ge(4===u.rank,(()=>`Error in conv2d: input must be rank 4, but got rank ${u.rank}.`)),Ge(4===l.rank,(()=>`Error in conv2d: filter must be rank 4, but got rank ${l.rank}.`)),vi("conv2d",s,a);const c="NHWC"===i?u.shape[3]:u.shape[1];Ge(c===l.shape[2],(()=>`Error in conv2d: depth of input (${c}) must match input depth for filter ${l.shape[2]}.`)),Ge(wi(n,r),(()=>`Error in conv2D: Either strides or dilations must be 1. Got strides ${n} and dilations '${r}'`)),Ge(ki(r),(()=>"Error in conv2D: Dilated rates should be larger than 0.")),Ge(ki(n),(()=>"Error in conv2D: Strides should be larger than 0."));const p={x:u,filter:l},d={strides:n,pad:s,dataFormat:i,dilations:r,dimRoundingMode:a},f=_s.runKernel(ge,p,d);return h?ii(f,[f.shape[1],f.shape[2],f.shape[3]]):f}}),Gi={kernelName:me,inputsToSave:["dy","filter"],gradFunc:(t,e,n)=>{const[s,i]=e,{strides:r,pad:a,dataFormat:o,dimRoundingMode:l}=n;return{dy:()=>Ki(t,i,r,a,o,1,l),filter:()=>ji(t,s,i.shape,r,a,o,l)}}};const Hi=Ws({conv3DBackpropFilter_:function(t,e,n,s,i){let r=t;4===t.rank&&(r=ii(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]));let a=e;4===a.rank&&(a=ii(e,[1,e.shape[0],e.shape[1],e.shape[2],e.shape[3]])),Ge(5===r.rank,(()=>`Error in conv3dDerFilter: input must be rank 5, but got shape ${r.shape}.`)),Ge(5===a.rank,(()=>`Error in conv3dDerFilter: dy must be rank 5, but got shape ${a.shape}.`)),Ge(5===n.length,(()=>`Error in conv3dDerFilter: filterShape must be length 5, but got ${n}.`)),Ge(r.shape[4]===n[3],(()=>`Error in conv3dDerFilter: depth of input ${r.shape[4]}) must match input depth in filter (${n[3]}.`)),Ge(a.shape[4]===n[4],(()=>`Error in conv3dDerFilter: depth of dy (${a.shape[4]}) must match output depth for filter (${n[4]}).`));const o={x:r,dy:a},l={strides:s,pad:i,filterShape:n};return _s.runKernel("Conv3DBackpropFilterV2",o,l)}});const Ji=Ws({conv3DBackpropInput_:function(t,e,n,s,i){Ge(t.length===e.rank,(()=>`Length of inShape (${t.length}) and rank of dy (${e.rank}) must match`));let r=t,a=e,o=!1;4===e.rank&&(o=!0,a=ii(e,[1,e.shape[0],e.shape[1],e.shape[2],e.shape[3]]),r=[1,t[0],t[1],t[2],t[3]]);const l=r[4],u=a.shape[4];Ge(5===r.length,(()=>`Error in conv3dDerInput: inShape must be length 5, but got length ${r.length}.`)),Ge(5===a.rank,(()=>`Error in conv3dDerInput: dy must be rank 5, but got rank ${a.rank}`)),Ge(5===n.rank,(()=>`Error in conv3dDerInput: filter must be rank 5, but got rank ${n.rank}`)),Ge(l===n.shape[3],(()=>`Error in conv3dDerInput: depth of input (${l}) must match input depth for filter ${n.shape[3]}.`)),Ge(u===n.shape[4],(()=>`Error in conv3dDerInput: depth of output (${u}) must match output depth for filter ${n.shape[4]}.`));const h={dy:a,filter:n},c={pad:i,strides:s,inputShape:r},p=_s.runKernel("Conv3DBackpropInputV2",h,c);return o?ii(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}}),Zi={kernelName:"Conv3D",inputsToSave:["x","filter"],gradFunc:(t,e,n)=>{const{dilations:s,strides:i,pad:r}=n;Ge(bi(s),(()=>`Error in gradient of conv3D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${s}'`));const[a,o]=e;return{x:()=>Ji(a.shape,t,o,i,r),filter:()=>Hi(a,t,o.shape,i,r)}}};const Yi=Ws({sin_:function(t){const e={x:Ps(t,"x","sin","float32")};return _s.runKernel("Sin",e)}}),Xi={kernelName:"Cos",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(Js(Yi(js(n,"float32"))),t)}}};const Qi=Ws({sinh_:function(t){const e={x:Ps(t,"x","sinh")};return _s.runKernel(_e,e)}}),tr={kernelName:ye,inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(Qi(js(n,"float32")),t)}}};function er(t,e){return function(t,e,n){const s=t.length+e.length,i=[];let r=0,a=0;for(let o=0;o<s;o++)-1===n.indexOf(o)?i.push(t[r++]):i.push(e[a++]);return i}(t,e.map((t=>1)),e)}function nr(t,e){if(function(t,e){for(let n=0;n<t.length;++n)if(t[t.length-n-1]!==e-1-n)return!1;return!0}(t,e))return null;const n=[];for(let s=0;s<e;++s)-1===t.indexOf(s)&&n.push(s);return t.forEach((t=>n.push(t))),n}function sr(t){return t.map(((t,e)=>[e,t])).sort(((t,e)=>t[1]-e[1])).map((t=>t[0]))}const ir=Ws({cumsum_:function(t,e=0,n=!1,s=!1){const i={x:Ps(t,"x","cumsum")},r={axis:e,exclusive:n,reverse:s};return _s.runKernel(be,i,r)}});const rr=Ws({complex_:function(t,e){const n=Ps(t,"real","complex"),s=Ps(e,"imag","complex");!function(t,e,n=""){Ge(Je(t,e),(()=>n+` Shapes ${t} and ${e} must match`))}(n.shape,s.shape,`real and imag shapes, ${n.shape} and ${s.shape}, must match in call to tf.complex().`);const i={real:n,imag:s};return _s.runKernel("Complex",i)}});const ar=Ws({imag_:function(t){const e={input:Ps(t,"input","imag")};return _s.runKernel("Imag",e)}});const or=Ws({real_:function(t){const e={input:Ps(t,"input","real")};return _s.runKernel("Real",e)}});const lr=Ws({transpose_:function(t,e,n){const s=Ps(t,"x","transpose");if(null==e&&(e=s.shape.map(((t,e)=>e)).reverse()),Ge(s.rank===e.length,(()=>`Error in transpose: rank of input ${s.rank} must match length of perm ${e}.`)),e.forEach((t=>{Ge(t>=0&&t<s.rank,(()=>"All entries in 'perm' must be between 0 and "+(s.rank-1)+` but got ${e}`))})),s.rank<=1)return s.clone();const i={x:s},r={perm:e};return"complex64"===s.dtype?(a=()=>{let t=or(s),e=ar(s);return t=_s.runKernel(Ue,{x:t},r),e=_s.runKernel(Ue,{x:e},r),n&&(e=Js(e)),rr(t,e)},_s.tidy(a,o)):_s.runKernel(Ue,i,r);var a,o}}),ur={kernelName:be,inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{axis:i,exclusive:r,reverse:a}=n;return{x:()=>{const e=nr([i],s.rank);let n=ir(t,i,r,!a);return null!=e&&(n=lr(n,e)),n}}}};const hr=Ws({depthwiseConv2dNativeBackpropFilter_:function(t,e,n,s,i,r=[1,1],a){let o=t;3===t.rank&&(o=ii(t,[1,t.shape[0],t.shape[1],t.shape[2]]));let l=e;3===l.rank&&(l=ii(e,[1,e.shape[0],e.shape[1],e.shape[2]]));const u={x:o,dy:l},h={strides:s,pad:i,dimRoundingMode:a,dilations:r,filterShape:n};return _s.runKernel("DepthwiseConv2dNativeBackpropFilter",u,h)}});const cr=Ws({depthwiseConv2dNativeBackpropInput_:function(t,e,n,s,i,r=[1,1],a){let o=e,l=!1;3===e.rank&&(l=!0,o=ii(e,[1,e.shape[0],e.shape[1],e.shape[2]]));const u={dy:o,filter:n},h={strides:s,pad:i,dimRoundingMode:a,dilations:r,inputShape:t},c=_s.runKernel("DepthwiseConv2dNativeBackpropInput",u,h);return l?ii(c,[c.shape[1],c.shape[2],c.shape[3]]):c}}),pr={kernelName:"DepthwiseConv2dNative",inputsToSave:["x","filter"],gradFunc:(t,e,n)=>{const{dilations:s,strides:i,pad:r,dimRoundingMode:a}=n,o=null==s?[1,1]:s;Ge(bi(o),(()=>`Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations '${o}'`));const[l,u]=e;return Ge(4===l.rank,(()=>`Error in gradient of depthwiseConv2dNative: input must be rank 4, but got rank ${l.rank}.`)),Ge(4===u.rank,(()=>`Error in gradient of depthwiseConv2dNative: filter must be rank 4, but got rank ${u.rank}.`)),Ge(l.shape[3]===u.shape[2],(()=>`Error in gradient of depthwiseConv2d: number of input channels (${l.shape[3]}) must match the inChannels dimension in filter ${u.shape[2]}.`)),Ge(wi(i,o),(()=>`Error in gradient of depthwiseConv2d: Either strides or dilations must be  1. Got strides ${i} and dilations '${o}'.`)),vi("depthwiseConv2d",r,a),{x:()=>cr(l.shape,t,u,i,r,o,a),filter:()=>hr(l,t,u.shape,i,r,o,a)}}},dr={kernelName:"Dilation2D",inputsToSave:["x","filter"],gradFunc:(t,e,n)=>{const[s,i]=e,r={x:s,filter:i,dy:t},a={x:s,filter:i,dy:t};return{x:()=>_s.runKernel("Dilation2DBackpropInput",r,n),filter:()=>_s.runKernel("Dilation2DBackpropFilter",a,n)}}},fr={kernelName:"Elu",outputsToSave:[!0],gradFunc:(t,e)=>{const[n]=e,s={dy:t,y:n};return{x:()=>_s.runKernel("EluGrad",s)}}};const gr=Ws({exp_:function(t){const e={x:Ps(t,"x","exp")};return _s.runKernel("Exp",e)}}),mr={kernelName:"Erf",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e,s=qs(gr(Js(Xs(n))),2/Math.sqrt(Math.PI));return{x:()=>qs(t,s)}}},yr={kernelName:"Exp",outputsToSave:[!0],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(t,n)}}},br={kernelName:ke,inputsToSave:["input"],gradFunc:(t,e)=>{const[n]=e;return{input:()=>ii(t,n.shape)}}},wr={kernelName:"Expm1",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(t,gr(n))}}},kr={kernelName:ve,gradFunc:t=>({x:()=>li(t)})},vr={kernelName:Se,inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e,i=si(n.shape,s.shape);return{a:()=>{const e=Hs(t,js(s,"float32")),r=ni(n.shape,i);return r.length>0?ii(ri(e,r),n.shape):e},b:()=>{let e=qs(t,js(n,"float32"));const r=ni(s.shape,i);r.length>0&&(e=ii(ri(e,r),s.shape));const a=Xs(s);return Js(Hs(e,js(a,"float32")))}}}};const Sr=Ws({rsqrt_:function(t){const e={x:Ps(t,"x","rsqrt","float32")};return _s.runKernel(Fe,e)}});const xr=Ws({tile_:function(t,e){const n=Ps(t,"x","tile","string_or_numeric");Ge(n.rank===e.length,(()=>`Error in transpose: rank of input ${n.rank} must match length of reps ${e}.`));const s={x:n},i={reps:e};return _s.runKernel(Pe,s,i)}}),Nr={kernelName:"FusedBatchNorm",inputsToSave:["x","mean","variance","scale"],gradFunc:(t,e,n)=>{const{varianceEpsilon:s}=n,[i,r,a,o]=e,l=null==o?Zs(1):o,u=ni(r.shape,i.shape),h=[];if(1===r.rank){for(let t=0;t<i.shape.length-1;++t)h.push(i.shape[t]);h.push(1)}const c=Qs(i,r),p=qs(t,l),d=Sr(pi(a,Zs(s))),f=qs(qs(qs(d,d),d),Zs(-.5));return{x:()=>1===r.rank?ii(qs(qs(t,xr(ii(d,[1,1,1,r.shape[0]]),h)),l),i.shape):ii(qs(qs(t,d),l),i.shape),mean:()=>{let t=qs(qs(d,Zs(-1)),p);return 1===r.rank&&(t=ri(t,u)),ii(t,r.shape)},variance:()=>{let t=qs(qs(f,c),p);return 1===r.rank&&(t=ri(t,u)),ii(t,r.shape)},scale:()=>{const e=qs(c,d);let n=qs(t,e);return 1===r.rank&&(n=ri(n,u)),ii(n,r.shape)},offset:()=>{let e=t;return 1===r.rank&&(e=ri(e,u)),ii(e,r.shape)}}}};const Ir=Ws({stack_:function(t,e=0){const n=Us(t,"tensors","stack","string_or_numeric");Ge(n.length>=1,(()=>"Pass at least one tensor to tf.stack")),n.length>0&&Ge(e<=n[0].rank,(()=>"Axis must be <= rank of the tensor"));const s=n,i={axis:e};return _s.runKernel(Ee,s,i)}});const Ar=Ws({unsortedSegmentSum_:function(t,e,n){const s=Ps(t,"x","unsortedSegmentSum"),i=Ps(e,"segmentIds","unsortedSegmentSum","int32");Ge(Ze(n),(()=>"numSegments must be of dtype int"));const r={x:s,segmentIds:i},a={numSegments:n};return _s.runKernel(je,r,a)}}),zr={kernelName:xe,inputsToSave:["x","indices"],gradFunc:(t,e,n)=>{const[s,i]=e,{axis:r,batchDims:a}=n,o=Xe(r,s.shape)[0],l=(t,e,n)=>()=>{const s=t.shape,i=e.size,a=s.slice(0,o),l=a.length,u=s.slice(r,s.length).slice(1),h=u.length,c=Er(0,l),p=Er(l+1,l+1+h),d=Tr([a,[i],u]),f=ii(n,d),g=ii(e,[i]),m=Tr([[l],c,p]),y=lr(f,m);let b=Ar(y,g,t.shape[o]);const w=sr(m);return b=lr(b,w),b};if(1===a){const e=s.shape[0],n=s.split(e,0);return{x:()=>{const e=Ir(n.map(((e,n)=>l(e,i.slice(n,1),t.slice(n,1))())));return e.reshape(s.shape)},indices:()=>i}}return{x:l(s,i,t),indices:()=>i}}};function Er(t,e){const n=[];for(let s=t;s<e;++s)n.push(s);return n}function Tr(t){const e=[];for(let n=0;n<t.length;++n)for(let s=0;s<t[n].length;++s)e.push(t[n][s]);return e}const Cr={kernelName:Ne,inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e;return{a:()=>li(n),b:()=>li(s)}}},$r={kernelName:Ie,gradFunc:t=>({x:()=>js(t,"float32")})},Fr={kernelName:"IsFinite",gradFunc:t=>({x:()=>li(t)})},Dr={kernelName:"IsInf",gradFunc:t=>({x:()=>li(t)})},Lr={kernelName:"IsNan",gradFunc:t=>({x:()=>li(t)})};const _r=Ws({greater_:function(t,e){let n=Ps(t,"a","greater","string_or_numeric"),s=Ps(e,"b","greater","string_or_numeric");[n,s]=Ts(n,s),si(n.shape,s.shape);const i={a:n,b:s};return _s.runKernel("Greater",i)}}),Rr={kernelName:"LeakyRelu",inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{alpha:i}=n,r=_r(s,0);return{x:()=>Mi(r,t,qs(t,i))}}},Or={kernelName:"Log1p",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Hs(t,pi(n,1))}}},Mr={kernelName:"Log",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Hs(t,js(n,"float32"))}}},Br={kernelName:"LogSoftmax",inputsToSave:[],outputsToSave:[!0],gradFunc:(t,e,n)=>{const[s]=e,{axis:i}=n;return{logits:()=>{const e=gr(s);return Qs(t,qs(ri(t,i,!0),e))}}}};const Pr=Ws({localResponseNormalizationBackprop_:function(t,e,n,s=5,i=1,r=1,a=.5){const o={x:t,y:e,dy:n},l={depthRadius:s,bias:i,alpha:r,beta:a};return _s.runKernel("LRNGrad",o,l)}}),Ur={kernelName:"LRN",inputsToSave:["x"],outputsToSave:[!0],gradFunc:(t,e,n)=>{const[s,i]=e,{depthRadius:r,bias:a,alpha:o,beta:l}=n;return{x:()=>Pr(s,i,t,r,a,o,l)}}};const Wr=Ws({equal_:function(t,e){let n=Ps(t,"a","equal","string_or_numeric"),s=Ps(e,"b","equal","string_or_numeric");[n,s]=Ts(n,s),si(n.shape,s.shape);const i={a:n,b:s};return _s.runKernel("Equal",i)}});function jr(t,e,n,s){return e.rank<n.rank&&(e=ii(e,er(e.shape,s))),t.rank<n.rank&&(t=ii(t,er(t.shape,s))),{x:()=>qs(t,js(Wr(n,e),t.dtype))}}const qr={kernelName:"Max",inputsToSave:["x"],outputsToSave:[!0],gradFunc:(t,e,n)=>{const s=n,{reductionIndices:i}=s,r=e[0],a=jr(t,e[1],r,Xe(i,r.shape));return{x:()=>a.x()}}};const Vr=Ws({less_:function(t,e){let n=Ps(t,"a","less","string_or_numeric"),s=Ps(e,"b","less","string_or_numeric");[n,s]=Ts(n,s),si(n.shape,s.shape);const i={a:n,b:s};return _s.runKernel("Less",i)}}),Kr={kernelName:Ae,inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e;return{a:()=>qs(t,js(Di(n,s),"float32")),b:()=>qs(t,js(Vr(n,s),"float32"))}}};const Gr=Ws({maxPool3dGrad_:function(t,e,n,s,i,r,a){const o=Ps(t,"dy","maxPool3dGrad"),l=Ps(e,"input","maxPool3dGrad"),u=Ps(n,"output","maxPool3dGrad");let h=o,c=l,p=u,d=!1;4===l.rank&&(d=!0,h=ii(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=ii(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),p=ii(u,[1,u.shape[0],u.shape[1],u.shape[2],u.shape[3]])),Ge(5===h.rank,(()=>`Error in maxPool3dGrad: dy must be rank 5 but got rank ${h.rank}.`)),Ge(5===c.rank,(()=>`Error in maxPool3dGrad: input must be rank 5 but got rank ${c.rank}.`)),Ge(5===p.rank,(()=>`Error in maxPool3dGrad: output must be rank 5 but got rank ${p.rank}.`)),vi("maxPool3dGrad",r,a);const f={dy:h,input:c,output:p},g={filterSize:s,strides:i,pad:r,dimRoundingMode:a},m=_s.runKernel("MaxPool3DGrad",f,g);return d?ii(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}}),Hr={kernelName:"MaxPool3D",inputsToSave:["x"],outputsToSave:[!0],gradFunc:(t,e,n)=>{const[s,i]=e,{filterSize:r,strides:a,pad:o,dimRoundingMode:l}=n;return{x:()=>Gr(t,s,i,r,a,o,l)}}};const Jr=Ws({maxPoolGrad_:function(t,e,n,s,i,r,a){const o=Ps(t,"dy","maxPoolGrad"),l=Ps(e,"input","maxPoolGrad"),u=Ps(n,"output","maxPoolGrad");Ge(l.rank===o.rank,(()=>`Rank of input (${l.rank}) does not match rank of dy (${o.rank})`)),Ge(4===o.rank,(()=>`Error in maxPoolGrad: dy must be rank 4 but got rank ${o.rank}.`)),Ge(4===l.rank,(()=>`Error in maxPoolGrad: input must be rank 4 but got rank ${l.rank}.`)),vi("maxPoolGrad",r,a);const h={dy:o,input:l,output:u},c={filterSize:s,strides:i,pad:r,dimRoundingMode:a};return _s.runKernel("MaxPoolGrad",h,c)}}),Zr={kernelName:"MaxPool",inputsToSave:["x"],outputsToSave:[!0],gradFunc:(t,e,n)=>{const[s,i]=e,{filterSize:r,strides:a,pad:o}=n;return{x:()=>Jr(t,s,i,r,a,o)}}};function Yr(t,e="float32"){if(un(t),"complex64"===e){const e=Yr(t,"float32"),n=Yr(t,"float32");return rr(e,n)}const n=ln(He(t),e);return _s.makeTensor(n,t,e)}function Xr(t,e="float32"){if(un(t),"complex64"===e){const e=Xr(t,"float32"),n=Yr(t,"float32");return rr(e,n)}const n=on(He(t),e);return _s.makeTensor(n,t,e)}const Qr={kernelName:"Mean",inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{axis:i}=n,r=Xe(i,s.shape),a=function(t,e){const n=[],s=t.length;for(let i=0;i<s;i++)-1===e.indexOf(i)&&n.push(t[i]);return[n,e.map((e=>t[e]))]}(s.shape,r),o=He(a[1]);return{x:()=>{const e=s.shape.slice();r.forEach((t=>{e[t]=1}));const n=ii(t,e);return Hs(qs(n,Xr(s.shape,"float32")),o)}}}},ta={kernelName:"Min",inputsToSave:["x"],outputsToSave:[!0],gradFunc:(t,e,n)=>{const s=n,{axis:i}=s,[r,a]=e,o=jr(t,a,r,Xe(i,r.shape));return{x:()=>o.x()}}},ea={kernelName:"Minimum",inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e;return{a:()=>qs(t,js(Li(n,s),"float32")),b:()=>qs(t,js(_r(n,s),"float32"))}}};const na=Ws({slice_:function(t,e,n){const s=Ps(t,"x","slice","string_or_numeric");if(0===s.rank)throw new Error("Slicing scalar is not possible");const i={x:s},r={begin:e,size:n};return _s.runKernel(Le,i,r)}}),sa={kernelName:"MirrorPad",inputsToSave:["x"],gradFunc:(t,e,n)=>{const s=e[0],{paddings:i}=n,r=i.map((t=>t[0]));return{x:()=>na(t,r,s.shape)}}};const ia=Ws({floor_:function(t){const e={x:Ps(t,"x","floor","float32")};return _s.runKernel(ve,e)}}),ra={kernelName:"Mod",inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e,i=si(n.shape,s.shape);return{a:()=>{const e=ni(n.shape,i);return e.length>0?ii(ri(t,e),n.shape):t},b:()=>{const e=qs(t,Js(ia(Hs(n,s)))),r=ni(s.shape,i);return r.length>0?ii(ri(e,r),s.shape):e}}}},aa={kernelName:ze,inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e,i=si(n.shape,s.shape);return{a:()=>{const e=qs(t,js(s,"float32")),r=ni(n.shape,i);return r.length>0?ii(ri(e,r),n.shape):e},b:()=>{const e=qs(t,js(n,"float32")),r=ni(s.shape,i);return r.length>0?ii(ri(e,r),s.shape):e}}}},oa={kernelName:"Neg",gradFunc:t=>({x:()=>Js(t)})},la={kernelName:"OneHot",inputsToSave:["indices"],gradFunc:(t,e)=>{const n=e[0];return{indices:()=>Yr(n.shape,"float32")}}},ua={kernelName:"OnesLike",gradFunc:t=>({x:()=>li(t)})};const ha=Ws({unstack_:function(t,e=0){const n=Ps(t,"x","unstack","string_or_numeric");Ge(e>=-n.shape.length&&e<n.shape.length,(()=>`Axis = ${e} is not in [-${n.shape.length}, ${n.shape.length})`));const s={value:n},i={axis:e};return _s.runKernel(We,s,i)}}),ca={kernelName:Ee,saveAllInputs:!0,gradFunc:(t,e,n)=>{const{axis:s}=n;return ha(t,s).map((t=>()=>t))}},pa={kernelName:Te,inputsToSave:["x"],gradFunc:(t,e,n)=>{const s=e[0],{paddings:i}=n,r=i.map((t=>t[0]));return{x:()=>na(t,r,s.shape)}}};const da=Ws({log_:function(t){const e={x:Ps(t,"x","log","float32")};return _s.runKernel("Log",e)}});const fa=Ws({pow_:function(t,e){let n=Ps(t,"base","pow"),s=Ps(e,"exp","pow");[n,s]=Ts(n,s);const i={a:n,b:s};return _s.runKernel("Pow",i)}}),ga={kernelName:"Pow",inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:(t,e)=>{const[n,s,i]=e,r=n,a=s,o=si(r.shape,a.shape);return{a:()=>{const e=js(a,"float32");let n=qs(t,qs(e,fa(r,Qs(e,Zs(1)))));const s=ni(r.shape,o);return s.length>0&&(n=ri(n,s)),ii(n,r.shape)},b:()=>{const e=_r(r,0),n=Mi(e,da(r),li(r));let s=qs(t,qs(i,n));const l=ni(a.shape,o);return l.length>0&&(s=ri(s,l)),ii(s,a.shape)}}}},ma={kernelName:"Prelu",inputsToSave:["x","alpha"],gradFunc:(t,e)=>{const[n,s]=e,i=_r(n,0);return{x:()=>Mi(i,t,qs(t,s)),alpha:()=>{let e=Mi(i,li(t),qs(t,n));const r=ni(s.shape,t.shape);return r.length>0&&(e=ri(e,r)),ii(e,s.shape)}}}};const ya=fn();ya.registerFlag("DEBUG",(()=>!1),(t=>{t&&console.warn("Debugging mode is ON. The output of every math call will be downloaded to CPU and checked for NaNs. This significantly impacts performance.")})),ya.registerFlag("IS_BROWSER",(()=>"undefined"!=typeof window&&null!=window.document||"undefined"!=typeof WorkerGlobalScope)),ya.registerFlag("IS_NODE",(()=>"undefined"!=typeof process&&"undefined"!=typeof process.versions&&"undefined"!=typeof process.versions.node)),ya.registerFlag("IS_CHROME",(()=>"undefined"!=typeof navigator&&null!=navigator&&null!=navigator.userAgent&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor))),ya.registerFlag("IS_SAFARI",(()=>"undefined"!=typeof navigator&&null!=navigator&&null!=navigator.userAgent&&/Safari/.test(navigator.userAgent)&&/Apple/.test(navigator.vendor))),ya.registerFlag("PROD",(()=>!1)),ya.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",(()=>ya.getBool("DEBUG"))),ya.registerFlag("DEPRECATION_WARNINGS_ENABLED",(()=>!0)),ya.registerFlag("IS_TEST",(()=>!1)),ya.registerFlag("CHECK_COMPUTATION_FOR_ERRORS",(()=>ya.getBool("DEBUG"))),ya.registerFlag("WRAP_TO_IMAGEBITMAP",(()=>!1)),ya.registerFlag("CANVAS2D_WILL_READ_FREQUENTLY_FOR_GPU",(()=>!1)),ya.registerFlag("USE_SETTIMEOUTCUSTOM",(()=>!1));class ba{static join(t){return new ba(t).slice()}constructor(t){if(this.shards=[],this.previousShardIndex=0,null==t)return;if(t instanceof Array||(t=[t]),0===(t=t.map((t=>as(t)?t.buffer:t))).length)return;this.bufferUniformSize=t[0].byteLength;let e=0;for(let n=0;n<t.length;n++){const s=t[n];n!==t.length-1&&s.byteLength!==this.bufferUniformSize&&(this.bufferUniformSize=void 0);const i=e+s.byteLength;this.shards.push({buffer:s,start:e,end:i}),e=i}0===this.shards.length&&(this.byteLength=0),this.byteLength=this.shards[this.shards.length-1].end}slice(t=0,e=this.byteLength){if(0===this.shards.length)return new ArrayBuffer(0);if(t=isNaN(Number(t))?0:t,e=isNaN(Number(e))?0:e,t=Math.max(0,t),(e=Math.min(this.byteLength,e))<=t)return new ArrayBuffer(0);const n=this.findShardForByte(t);if(-1===n)throw new Error(`Could not find start shard for byte ${t}`);const s=new ArrayBuffer(e-t),i=new Uint8Array(s);let r=0;for(let s=n;s<this.shards.length;s++){const n=this.shards[s],a=t+r-n.start,o=r,l=Math.min(e,n.end)-n.start,u=new Uint8Array(n.buffer,a,l-a);if(i.set(u,o),r+=u.length,e<n.end)break}return s}findShardForByte(t){if(0===this.shards.length||t<0||t>=this.byteLength)return-1;if(null!=this.bufferUniformSize)return this.previousShardIndex=Math.floor(t/this.bufferUniformSize),this.previousShardIndex;function e(e){return t<e.start?-1:t>=e.end?1:0}if(0===e(this.shards[this.previousShardIndex]))return this.previousShardIndex;const n=function(t,e){let n=0,s=t.length;for(;n<=s;){const i=Math.floor((s-n)/2)+n,r=e(t[i]);if(0===r)return i;r<0?s=i:n=i+1}return-1}(this.shards,e);return-1===n?-1:(this.previousShardIndex=n,this.previousShardIndex)}}const wa="undefined"!=typeof Buffer&&("undefined"==typeof Blob||"undefined"==typeof atob||"undefined"==typeof btoa);function ka(t){return wa?Buffer.byteLength(t,"utf8"):new Blob([t]).size}function va(t,e){const n={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,weightsManifest:e};return null!=t.signature&&(n.signature=t.signature),null!=t.userDefinedMetadata&&(n.userDefinedMetadata=t.userDefinedMetadata),null!=t.modelInitializer&&(n.modelInitializer=t.modelInitializer),null!=t.initializerSignature&&(n.initializerSignature=t.initializerSignature),null!=t.trainingConfig&&(n.trainingConfig=t.trainingConfig),n}async function Sa(t,e){let n,s;return null!=t.weightsManifest&&([n,s]=await e(t.weightsManifest)),function(t,e,n){const s={modelTopology:t.modelTopology,format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy};if(null!=t.trainingConfig&&(s.trainingConfig=t.trainingConfig),null!=t.weightsManifest){if(!e)throw new Error("modelJSON has weightsManifest but weightSpecs is null");if(!n)throw new Error("modelJSON has weightsManifest but weightData is null");s.weightSpecs=e,s.weightData=n}return null!=t.signature&&(s.signature=t.signature),null!=t.userDefinedMetadata&&(s.userDefinedMetadata=t.userDefinedMetadata),null!=t.modelInitializer&&(s.modelInitializer=t.modelInitializer),null!=t.initializerSignature&&(s.initializerSignature=t.initializerSignature),s}(t,n,s)}function xa(t){if(t.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:null==t.modelTopology?0:ka(JSON.stringify(t.modelTopology)),weightSpecsBytes:null==t.weightSpecs?0:ka(JSON.stringify(t.weightSpecs)),weightDataBytes:null==t.weightData?0:new ba(t.weightData).byteLength}}function Na(t){const e=[];for(const n of t)e.push(...n.weights);return e}class Ia{constructor(){this.saveRouters=[],this.loadRouters=[]}static getInstance(){return null==Ia.instance&&(Ia.instance=new Ia),Ia.instance}static registerSaveRouter(t){Ia.getInstance().saveRouters.push(t)}static registerLoadRouter(t){Ia.getInstance().loadRouters.push(t)}static getSaveHandlers(t){return Ia.getHandlers(t,"save")}static getLoadHandlers(t,e){return Ia.getHandlers(t,"load",e)}static getHandlers(t,e,n){const s=[];return("load"===e?Ia.getInstance().loadRouters:Ia.getInstance().saveRouters).forEach((e=>{const i=e(t,n);null!==i&&s.push(i)})),s}}const Aa="models_store",za="model_info_store";class Ea{constructor(t){if(this.indexedDB=function(){if(!fn().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");const t="undefined"==typeof window?self:window,e=t.indexedDB||t.mozIndexedDB||t.webkitIndexedDB||t.msIndexedDB||t.shimIndexedDB;if(null==e)throw new Error("The current browser does not appear to support IndexedDB.");return e}(),null==t||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}async save(t){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return this.databaseAction(this.modelPath,t)}async load(){return this.databaseAction(this.modelPath)}databaseAction(t,e){return new Promise(((t,n)=>{const s=this.indexedDB.open("tensorflowjs",1);s.onupgradeneeded=()=>function(t){const e=t.result;e.createObjectStore(Aa,{keyPath:"modelPath"}),e.createObjectStore(za,{keyPath:"modelPath"})}(s),s.onsuccess=()=>{const i=s.result;if(null==e){const e=i.transaction(Aa,"readonly"),s=e.objectStore(Aa).get(this.modelPath);s.onsuccess=()=>{if(null==s.result)return i.close(),n(new Error(`Cannot find model with path '${this.modelPath}' in IndexedDB.`));t(s.result.modelArtifacts)},s.onerror=t=>(i.close(),n(s.error)),e.oncomplete=()=>i.close()}else{e.weightData=ba.join(e.weightData);const s=xa(e),r=i.transaction(za,"readwrite");let a,o,l=r.objectStore(za);try{a=l.put({modelPath:this.modelPath,modelArtifactsInfo:s})}catch(t){return n(t)}a.onsuccess=()=>{o=i.transaction(Aa,"readwrite");const a=o.objectStore(Aa);let u;try{u=a.put({modelPath:this.modelPath,modelArtifacts:e,modelArtifactsInfo:s})}catch(t){return n(t)}u.onsuccess=()=>t({modelArtifactsInfo:s}),u.onerror=t=>{l=r.objectStore(za);const e=l.delete(this.modelPath);e.onsuccess=()=>(i.close(),n(u.error)),e.onerror=t=>(i.close(),n(u.error))}},a.onerror=t=>(i.close(),n(a.error)),r.oncomplete=()=>{null==o?i.close():o.oncomplete=()=>i.close()}}},s.onerror=t=>n(s.error)}))}}Ea.URL_SCHEME="indexeddb://";const Ta=t=>{return fn().getBool("IS_BROWSER")&&!Array.isArray(t)&&t.startsWith(Ea.URL_SCHEME)?(e=t.slice(Ea.URL_SCHEME.length),new Ea(e)):null;var e};Ia.registerSaveRouter(Ta),Ia.registerLoadRouter(Ta);const Ca="/",$a="tensorflowjs_models",Fa="info",Da="model_topology",La="weight_specs",_a="weight_data",Ra="model_metadata";class Oa{constructor(t){if(!fn().getBool("IS_BROWSER")||"undefined"==typeof window||"undefined"==typeof window.localStorage)throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,null==t||!t)throw new Error("For local storage, modelPath must not be null, undefined or empty.");var e;this.modelPath=t,this.keys=(e=this.modelPath,{info:[$a,e,Fa].join(Ca),topology:[$a,e,Da].join(Ca),weightSpecs:[$a,e,La].join(Ca),weightData:[$a,e,_a].join(Ca),modelMetadata:[$a,e,Ra].join(Ca)})}async save(t){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");{const e=JSON.stringify(t.modelTopology),n=JSON.stringify(t.weightSpecs),s=xa(t),i=ba.join(t.weightData);try{this.LS.setItem(this.keys.info,JSON.stringify(s)),this.LS.setItem(this.keys.topology,e),this.LS.setItem(this.keys.weightSpecs,n),this.LS.setItem(this.keys.weightData,function(t){if(wa)return Buffer.from(t).toString("base64");const e=new Uint8Array(t);let n="";for(let t=0,s=e.length;t<s;t++)n+=String.fromCharCode(e[t]);return btoa(n)}(i));const r={format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,signature:null!=t.signature?t.signature:void 0,userDefinedMetadata:null!=t.userDefinedMetadata?t.userDefinedMetadata:void 0,modelInitializer:null!=t.modelInitializer?t.modelInitializer:void 0,initializerSignature:null!=t.initializerSignature?t.initializerSignature:void 0,trainingConfig:null!=t.trainingConfig?t.trainingConfig:void 0};return this.LS.setItem(this.keys.modelMetadata,JSON.stringify(r)),{modelArtifactsInfo:s}}catch(t){throw function(t){for(const e of Object.values(t))window.localStorage.removeItem(e)}(this.keys),new Error(`Failed to save model '${this.modelPath}' to local storage: size quota being exceeded is a possible cause of this failure: modelTopologyBytes=${s.modelTopologyBytes}, weightSpecsBytes=${s.weightSpecsBytes}, weightDataBytes=${s.weightDataBytes}.`)}}}async load(){const t=JSON.parse(this.LS.getItem(this.keys.info));if(null==t)throw new Error(`In local storage, there is no model with name '${this.modelPath}'`);if("JSON"!==t.modelTopologyType)throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");const e={},n=JSON.parse(this.LS.getItem(this.keys.topology));if(null==n)throw new Error(`In local storage, the topology of model '${this.modelPath}' is missing.`);e.modelTopology=n;const s=JSON.parse(this.LS.getItem(this.keys.weightSpecs));if(null==s)throw new Error(`In local storage, the weight specs of model '${this.modelPath}' are missing.`);e.weightSpecs=s;const i=this.LS.getItem(this.keys.modelMetadata);if(null!=i){const t=JSON.parse(i);e.format=t.format,e.generatedBy=t.generatedBy,e.convertedBy=t.convertedBy,null!=t.signature&&(e.signature=t.signature),null!=t.userDefinedMetadata&&(e.userDefinedMetadata=t.userDefinedMetadata),null!=t.modelInitializer&&(e.modelInitializer=t.modelInitializer),null!=t.initializerSignature&&(e.initializerSignature=t.initializerSignature),null!=t.trainingConfig&&(e.trainingConfig=t.trainingConfig)}const r=this.LS.getItem(this.keys.weightData);if(null==r)throw new Error(`In local storage, the binary weight values of model '${this.modelPath}' are missing.`);return e.weightData=function(t){if(wa){const e=Buffer.from(t,"base64");return e.buffer.slice(e.byteOffset,e.byteOffset+e.byteLength)}const e=atob(t),n=new Uint8Array(e.length);for(let t=0;t<e.length;++t)n.set([e.charCodeAt(t)],t);return n.buffer}(r),e}}Oa.URL_SCHEME="localstorage://";const Ma=t=>{return fn().getBool("IS_BROWSER")&&!Array.isArray(t)&&t.startsWith(Oa.URL_SCHEME)?(e=t.slice(Oa.URL_SCHEME.length),new Oa(e)):null;var e};Ia.registerSaveRouter(Ma),Ia.registerLoadRouter(Ma);function Ba(t){return new Promise((t=>setTimeout(t))).then(t)}class Pa{constructor(t){if(!fn().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");t.startsWith(Pa.URL_SCHEME)&&(t=t.slice(Pa.URL_SCHEME.length)),null!=t&&0!==t.length||(t="model"),this.modelJsonFileName=t+".json",this.weightDataFileName=t+".weights.bin"}async save(t){if("undefined"==typeof document)throw new Error("Browser downloads are not supported in this environment since `document` is not present");const e=ba.join(t.weightData),n=window.URL.createObjectURL(new Blob([e],{type:"application/octet-stream"}));if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");{const e=va(t,[{paths:["./"+this.weightDataFileName],weights:t.weightSpecs}]),s=window.URL.createObjectURL(new Blob([JSON.stringify(e)],{type:"application/json"})),i=null==this.modelJsonAnchor?document.createElement("a"):this.modelJsonAnchor;if(i.download=this.modelJsonFileName,i.href=s,await Ba((()=>i.dispatchEvent(new MouseEvent("click")))),null!=t.weightData){const t=null==this.weightDataAnchor?document.createElement("a"):this.weightDataAnchor;t.download=this.weightDataFileName,t.href=n,await Ba((()=>t.dispatchEvent(new MouseEvent("click"))))}return{modelArtifactsInfo:xa(t)}}}}Pa.URL_SCHEME="downloads://";function Ua(t,e,n,s){!function(t){Ge(null!=t&&Array.isArray(t)&&t.length>0,(()=>"promises must be a none empty array"))}(t),function(t,e){Ge(t>=0&&t<=1,(()=>`Progress fraction must be in range [0, 1], but got startFraction ${t}`)),Ge(e>=0&&e<=1,(()=>`Progress fraction must be in range [0, 1], but got endFraction ${e}`)),Ge(e>=t,(()=>`startFraction must be no more than endFraction, but got startFraction ${t} and endFraction ${e}`))}(n=null==n?0:n,s=null==s?1:s);let i=0;return Promise.all(t.map((r=>(r.then((r=>{const a=n+ ++i/t.length*(s-n);return e(a),r})),r))))}Ia.registerSaveRouter((t=>fn().getBool("IS_BROWSER")&&!Array.isArray(t)&&t.startsWith(Pa.URL_SCHEME)?function(t="model"){return new Pa(t)}(t.slice(Pa.URL_SCHEME.length)):null));class Wa{constructor(t,e){if(this.DEFAULT_METHOD="POST",null==e&&(e={}),this.weightPathPrefix=e.weightPathPrefix,this.weightUrlConverter=e.weightUrlConverter,null!=e.fetchFunc?(Ge("function"==typeof e.fetchFunc,(()=>"Must pass a function that matches the signature of `fetch` (see https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)")),this.fetch=e.fetchFunc):this.fetch=fn().platform.fetch,Ge(null!=t&&t.length>0,(()=>"URL path for http must not be null, undefined or empty.")),Array.isArray(t)&&Ge(2===t.length,(()=>`URL paths for http must have a length of 2, (actual length is ${t.length}).`)),this.path=t,null!=e.requestInit&&null!=e.requestInit.body)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{},this.loadOptions=e}async save(t){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");const e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit);e.body=new FormData;const n=va(t,[{paths:["./model.weights.bin"],weights:t.weightSpecs}]);if(e.body.append("model.json",new Blob([JSON.stringify(n)],{type:"application/json"}),"model.json"),null!=t.weightData){const n=ba.join(t.weightData);e.body.append("model.weights.bin",new Blob([n],{type:"application/octet-stream"}),"model.weights.bin")}const s=await this.fetch(this.path,e);if(s.ok)return{modelArtifactsInfo:xa(t),responses:[s]};throw new Error(`BrowserHTTPRequest.save() failed due to HTTP response status ${s.status}.`)}async loadModelJSON(){const t=await this.fetch(this.path,this.requestInit);if(!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.`);let e;try{e=await t.json()}catch(t){let e=`Failed to parse model JSON of response from ${this.path}.`;throw this.path.endsWith(".pb")?e+=" 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.":e+=" Please make sure the server is serving valid JSON for this request.",new Error(e)}const n=e.modelTopology,s=e.weightsManifest;if(null==n&&null==s)throw new Error(`The JSON from HTTP path ${this.path} contains neither model topology or manifest for weights.`);return e}async load(){if(this.loadOptions.streamWeights)return this.loadStream();return Sa(await this.loadModelJSON(),(t=>this.loadWeights(t)))}async loadStream(){const t=await this.loadModelJSON(),e=await this.getWeightUrls(t.weightsManifest),n=Na(t.weightsManifest);return Object.assign(Object.assign({},t),{weightSpecs:n,getWeightStream:()=>function(t,e){var n;const s=null==e.fetchFunc?fn().platform.fetch:e.fetchFunc;let i,r=0;return null===(n=e.onProgress)||void 0===n||n.call(e,0),new ReadableStream({pull:async n=>{for(var a;r<t.length;){if(!i){const n=(await s(t[r],e.requestInit,{isBinary:!0})).body;i=n.getReader()}const{done:o,value:l}=await i.read();if(!o)return void n.enqueue(l);r++,i=void 0,null===(a=e.onProgress)||void 0===a||a.call(e,r/t.length)}n.close()}})}(e,this.loadOptions)})}async getWeightUrls(t){const e=Array.isArray(this.path)?this.path[1]:this.path,[n,s]=function(t){const e=t.lastIndexOf("/"),n=t.lastIndexOf("?"),s=t.substring(0,e),i=n>e?t.substring(n):"";return[s+"/",i]}(e),i=this.weightPathPrefix||n,r=[],a=[];for(const e of t)for(const t of e.paths)null!=this.weightUrlConverter?a.push(this.weightUrlConverter(t)):r.push(i+t+s);return this.weightUrlConverter&&r.push(...await Promise.all(a)),r}async loadWeights(t){const e=await this.getWeightUrls(t),n=Na(t),s=await async function(t,e){null==e&&(e={});const n=null==e.fetchFunc?fn().platform.fetch:e.fetchFunc,s=t.map((t=>n(t,e.requestInit,{isBinary:!0}))),i=(null==e.onProgress?await Promise.all(s):await Ua(s,e.onProgress,0,.5)).map((t=>t.arrayBuffer()));return null==e.onProgress?await Promise.all(i):await Ua(i,e.onProgress,.5,1)}(e,this.loadOptions);return[n,s]}}function ja(t){return null!=t.match(Wa.URL_SCHEME_REGEX)}Wa.URL_SCHEME_REGEX=/^https?:\/\//;const qa=(t,e)=>{if("undefined"==typeof fetch&&(null==e||null==e.fetchFunc))return null;{let n=!0;if(n=Array.isArray(t)?t.every((t=>ja(t))):ja(t),n)return function(t,e){return new Wa(t,e)}(t,e)}return null};function Va(t,e,n){if(e.rank<1)throw new Error(`tf.scatterND() expects the indices to be rank 1 or higher, but the rank was ${e.rank}.`);if(t.rank<1)throw new Error(`tf.scatterND() expects the updates to be rank 1 or higher, but the rank was ${t.rank}.`);if("int32"!==e.dtype)throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${e.dtype}`);if(n.length<1)throw new Error(`Output rank must be greater or equal to 1, but got shape: ${n}`);if(0===n.length){if(0===e.size)throw new Error(`Indices specified for empty output. indices shape: ${e.shape}`);if(0===t.size)throw new Error(`Updates specified for empty output. updates shape: ${t.shape}`)}!function(t,e,n){const s=e.rank>1?e.shape[e.rank-1]:1,i=e.rank>1?e.rank-1:1,r=`Must have updates.shape = indices.shape[:batchDim] + shape[sliceDim:], got updates.shape: ${n.shape}, indices.shape: ${e.shape}, shape: ${t}, sliceDim: ${s}, and batchDim: ${i}.`;if(n.rank<i)throw new Error(r+` update.rank < ${i}. `);if(t.length<s+(n.rank-i))throw new Error(r+` Output shape length < ${s+(n.rank-i)}`);if(n.rank!==i+t.length-s)throw new Error(r+" update.rank != "+(i+t.length-s));for(let t=0;t<i;++t)if(n.shape[t]!==e.shape[t])throw new Error(r+` updates.shape[${t}] (${n.shape[t]}) != indices.shape[${t}] (${e.shape[t]}).`);for(let e=0;e<n.rank-i;++e)if(n.shape[e+i]!==t[e+s])throw new Error(r+` updates.shape[${e+i}] (${n.shape[e+i]}) != shape[${e+i}] (${t[e+i]})`)}(n,e,t)}Ia.registerSaveRouter(qa),Ia.registerLoadRouter(qa);const Ka=Ws({concat_:function(t,e=0){Ge(t.length>=1,(()=>"Pass at least one tensor to concat"));const n=Us(t,"tensors","concat","string_or_numeric");if("complex64"===n[0].dtype&&n.forEach((t=>{if("complex64"!==t.dtype)throw new Error(`Cannot concatenate complex64 tensors with a tensor\n          with dtype ${t.dtype}. `)})),1===n.length)return Ri(n[0]);const s=n,i={axis:e};return _s.runKernel(fe,s,i)}});const Ga=Ws({sigmoid_:function(t){const e={x:Ps(t,"x","sigmoid","float32")};return _s.runKernel(Re,e)}});const Ha=Ws({batchToSpaceND_:function(t,e,n){const s=Ps(t,"x","batchToSpaceND"),i=e.reduce(((t,e)=>t*e));Ge(s.rank>=1+e.length,(()=>`input rank is ${s.rank} but should be > than blockShape.length ${e.length}`)),Ge(n.length===e.length,(()=>`crops.length is ${n.length} but should be equal to blockShape.length  ${e.length}`)),Ge(s.shape[0]%i==0,(()=>`input tensor batch is ${s.shape[0]} but is not divisible by the product of the elements of blockShape ${e.join(" * ")} === ${i}`));const r={x:s},a={blockShape:e,crops:n};return _s.runKernel(pe,r,a)}});const Ja=Ws({cos_:function(t){const e={x:Ps(t,"x","cos","float32")};return _s.runKernel("Cos",e)}});const Za=Ws({cosh_:function(t){const e={x:Ps(t,"x","cosh","float32")};return _s.runKernel(ye,e)}});const Ya=Ws({cumprod_:function(t,e=0,n=!1,s=!1){const i={x:Ps(t,"x","cumprod")},r={axis:e,exclusive:n,reverse:s};return _s.runKernel("Cumprod",i,r)}});const Xa=Ws({expandDims_:function(t,e=0){const n=Ps(t,"x","expandDims","string_or_numeric");Ge(e<=n.rank,(()=>"Axis must be <= rank of the tensor"));const s={input:n},i={dim:e};return _s.runKernel(ke,s,i)}});const Qa=Ws({gather_:function(t,e,n=0,s=0){const i={x:Ps(t,"x","gather"),indices:Ps(e,"indices","gather","int32")},r={axis:n,batchDims:s};return _s.runKernel(xe,i,r)}});const to=Ws({logicalNot_:function(t){const e={x:Ps(t,"x","logicalNot","bool")};return _s.runKernel("LogicalNot",e)}});const eo=Ws({maximum_:function(t,e){let n=Ps(t,"a","maximum"),s=Ps(e,"b","maximum");[n,s]=Ts(n,s),"bool"===n.dtype&&(n=js(n,"int32"),s=js(s,"int32")),si(n.shape,s.shape);const i={a:n,b:s};return _s.runKernel(Ae,i)}});const no=Ws({pad_:function(t,e,n=0){const s=Ps(t,"x","pad");if(0===s.rank)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");const i={paddings:e,constantValue:n},r={x:s};return _s.runKernel(Te,r,i)}});var so={exports:{}};!function(t){!function(t,e,n){function s(t){var e,n=this,s=(e=4022871197,function(t){t=String(t);for(var n=0;n<t.length;n++){var s=.02519603282416938*(e+=t.charCodeAt(n));s-=e=s>>>0,e=(s*=e)>>>0,e+=4294967296*(s-=e)}return 2.3283064365386963e-10*(e>>>0)});n.next=function(){var t=2091639*n.s0+2.3283064365386963e-10*n.c;return n.s0=n.s1,n.s1=n.s2,n.s2=t-(n.c=0|t)},n.c=1,n.s0=s(" "),n.s1=s(" "),n.s2=s(" "),n.s0-=s(t),n.s0<0&&(n.s0+=1),n.s1-=s(t),n.s1<0&&(n.s1+=1),n.s2-=s(t),n.s2<0&&(n.s2+=1),s=null}function i(t,e){return e.c=t.c,e.s0=t.s0,e.s1=t.s1,e.s2=t.s2,e}function r(t,e){var n=new s(t),r=e&&e.state,a=n.next;return a.int32=function(){return 4294967296*n.next()|0},a.double=function(){return a()+11102230246251565e-32*(2097152*a()|0)},a.quick=a,r&&("object"==typeof r&&i(r,n),a.state=function(){return i(n,{})}),a}e&&e.exports?e.exports=r:n&&n.amd?n((function(){return r})):this.alea=r}(0,t,!1)}(so);var io=so.exports,ro={exports:{}};!function(t){!function(t,e,n){function s(t){var e=this,n="";e.x=0,e.y=0,e.z=0,e.w=0,e.next=function(){var t=e.x^e.x<<11;return e.x=e.y,e.y=e.z,e.z=e.w,e.w^=e.w>>>19^t^t>>>8},t===(0|t)?e.x=t:n+=t;for(var s=0;s<n.length+64;s++)e.x^=0|n.charCodeAt(s),e.next()}function i(t,e){return e.x=t.x,e.y=t.y,e.z=t.z,e.w=t.w,e}function r(t,e){var n=new s(t),r=e&&e.state,a=function(){return(n.next()>>>0)/4294967296};return a.double=function(){do{var t=((n.next()>>>11)+(n.next()>>>0)/4294967296)/(1<<21)}while(0===t);return t},a.int32=n.next,a.quick=a,r&&("object"==typeof r&&i(r,n),a.state=function(){return i(n,{})}),a}e&&e.exports?e.exports=r:n&&n.amd?n((function(){return r})):this.xor128=r}(0,t,!1)}(ro);var ao=ro.exports,oo={exports:{}};!function(t){!function(t,e,n){function s(t){var e=this,n="";e.next=function(){var t=e.x^e.x>>>2;return e.x=e.y,e.y=e.z,e.z=e.w,e.w=e.v,(e.d=e.d+362437|0)+(e.v=e.v^e.v<<4^t^t<<1)|0},e.x=0,e.y=0,e.z=0,e.w=0,e.v=0,t===(0|t)?e.x=t:n+=t;for(var s=0;s<n.length+64;s++)e.x^=0|n.charCodeAt(s),s==n.length&&(e.d=e.x<<10^e.x>>>4),e.next()}function i(t,e){return e.x=t.x,e.y=t.y,e.z=t.z,e.w=t.w,e.v=t.v,e.d=t.d,e}function r(t,e){var n=new s(t),r=e&&e.state,a=function(){return(n.next()>>>0)/4294967296};return a.double=function(){do{var t=((n.next()>>>11)+(n.next()>>>0)/4294967296)/(1<<21)}while(0===t);return t},a.int32=n.next,a.quick=a,r&&("object"==typeof r&&i(r,n),a.state=function(){return i(n,{})}),a}e&&e.exports?e.exports=r:n&&n.amd?n((function(){return r})):this.xorwow=r}(0,t,!1)}(oo);var lo=oo.exports,uo={exports:{}};!function(t){!function(t,e,n){function s(t){var e=this;e.next=function(){var t,n,s=e.x,i=e.i;return t=s[i],n=(t^=t>>>7)^t<<24,n^=(t=s[i+1&7])^t>>>10,n^=(t=s[i+3&7])^t>>>3,n^=(t=s[i+4&7])^t<<7,t=s[i+7&7],n^=(t^=t<<13)^t<<9,s[i]=n,e.i=i+1&7,n},function(t,e){var n,s=[];if(e===(0|e))s[0]=e;else for(e=""+e,n=0;n<e.length;++n)s[7&n]=s[7&n]<<15^e.charCodeAt(n)+s[n+1&7]<<13;for(;s.length<8;)s.push(0);for(n=0;n<8&&0===s[n];++n);for(8==n?s[7]=-1:s[n],t.x=s,t.i=0,n=256;n>0;--n)t.next()}(e,t)}function i(t,e){return e.x=t.x.slice(),e.i=t.i,e}function r(t,e){null==t&&(t=+new Date);var n=new s(t),r=e&&e.state,a=function(){return(n.next()>>>0)/4294967296};return a.double=function(){do{var t=((n.next()>>>11)+(n.next()>>>0)/4294967296)/(1<<21)}while(0===t);return t},a.int32=n.next,a.quick=a,r&&(r.x&&i(r,n),a.state=function(){return i(n,{})}),a}e&&e.exports?e.exports=r:n&&n.amd?n((function(){return r})):this.xorshift7=r}(0,t,!1)}(uo);var ho=uo.exports,co={exports:{}};!function(t){!function(t,e,n){function s(t){var e=this;e.next=function(){var t,n,s=e.w,i=e.X,r=e.i;return e.w=s=s+1640531527|0,n=i[r+34&127],t=i[r=r+1&127],n^=n<<13,t^=t<<17,n^=n>>>15,t^=t>>>12,n=i[r]=n^t,e.i=r,n+(s^s>>>16)|0},function(t,e){var n,s,i,r,a,o=[],l=128;for(e===(0|e)?(s=e,e=null):(e+="\0",s=0,l=Math.max(l,e.length)),i=0,r=-32;r<l;++r)e&&(s^=e.charCodeAt((r+32)%e.length)),0===r&&(a=s),s^=s<<10,s^=s>>>15,s^=s<<4,s^=s>>>13,r>=0&&(a=a+1640531527|0,i=0==(n=o[127&r]^=s+a)?i+1:0);for(i>=128&&(o[127&(e&&e.length||0)]=-1),i=127,r=512;r>0;--r)s=o[i+34&127],n=o[i=i+1&127],s^=s<<13,n^=n<<17,s^=s>>>15,n^=n>>>12,o[i]=s^n;t.w=a,t.X=o,t.i=i}(e,t)}function i(t,e){return e.i=t.i,e.w=t.w,e.X=t.X.slice(),e}function r(t,e){null==t&&(t=+new Date);var n=new s(t),r=e&&e.state,a=function(){return(n.next()>>>0)/4294967296};return a.double=function(){do{var t=((n.next()>>>11)+(n.next()>>>0)/4294967296)/(1<<21)}while(0===t);return t},a.int32=n.next,a.quick=a,r&&(r.X&&i(r,n),a.state=function(){return i(n,{})}),a}e&&e.exports?e.exports=r:n&&n.amd?n((function(){return r})):this.xor4096=r}(0,t,!1)}(co);var po=co.exports,fo={exports:{}};!function(t){!function(t,e,n){function s(t){var e=this,n="";e.next=function(){var t=e.b,n=e.c,s=e.d,i=e.a;return t=t<<25^t>>>7^n,n=n-s|0,s=s<<24^s>>>8^i,i=i-t|0,e.b=t=t<<20^t>>>12^n,e.c=n=n-s|0,e.d=s<<16^n>>>16^i,e.a=i-t|0},e.a=0,e.b=0,e.c=-1640531527,e.d=1367130551,t===Math.floor(t)?(e.a=t/4294967296|0,e.b=0|t):n+=t;for(var s=0;s<n.length+20;s++)e.b^=0|n.charCodeAt(s),e.next()}function i(t,e){return e.a=t.a,e.b=t.b,e.c=t.c,e.d=t.d,e}function r(t,e){var n=new s(t),r=e&&e.state,a=function(){return(n.next()>>>0)/4294967296};return a.double=function(){do{var t=((n.next()>>>11)+(n.next()>>>0)/4294967296)/(1<<21)}while(0===t);return t},a.int32=n.next,a.quick=a,r&&("object"==typeof r&&i(r,n),a.state=function(){return i(n,{})}),a}e&&e.exports?e.exports=r:n&&n.amd?n((function(){return r})):this.tychei=r}(0,t,!1)}(fo);var go=fo.exports,mo={exports:{}},yo=En({__proto__:null,default:{}});!function(t){!function(e,n,s){var i,r=256,a=s.pow(r,6),o=s.pow(2,52),l=2*o,u=255;function h(t,u,h){var m=[],y=f(d((u=1==u?{entropy:!0}:u||{}).entropy?[t,g(n)]:null==t?function(){try{var t;return i&&(t=i.randomBytes)?t=t(r):(t=new Uint8Array(r),(e.crypto||e.msCrypto).getRandomValues(t)),g(t)}catch(t){var s=e.navigator,a=s&&s.plugins;return[+new Date,e,a,e.screen,g(n)]}}():t,3),m),b=new c(m),w=function(){for(var t=b.g(6),e=a,n=0;t<o;)t=(t+n)*r,e*=r,n=b.g(1);for(;t>=l;)t/=2,e/=2,n>>>=1;return(t+n)/e};return w.int32=function(){return 0|b.g(4)},w.quick=function(){return b.g(4)/4294967296},w.double=w,f(g(b.S),n),(u.pass||h||function(t,e,n,i){return i&&(i.S&&p(i,b),t.state=function(){return p(b,{})}),n?(s.random=t,e):t})(w,y,"global"in u?u.global:this==s,u.state)}function c(t){var e,n=t.length,s=this,i=0,a=s.i=s.j=0,o=s.S=[];for(n||(t=[n++]);i<r;)o[i]=i++;for(i=0;i<r;i++)o[i]=o[a=u&a+t[i%n]+(e=o[i])],o[a]=e;(s.g=function(t){for(var e,n=0,i=s.i,a=s.j,o=s.S;t--;)e=o[i=u&i+1],n=n*r+o[u&(o[i]=o[a=u&a+e])+(o[a]=e)];return s.i=i,s.j=a,n})(r)}function p(t,e){return e.i=t.i,e.j=t.j,e.S=t.S.slice(),e}function d(t,e){var n,s=[],i=typeof t;if(e&&"object"==i)for(n in t)try{s.push(d(t[n],e-1))}catch(t){}return s.length?s:"string"==i?t:t+"\0"}function f(t,e){for(var n,s=t+"",i=0;i<s.length;)e[u&i]=u&(n^=19*e[u&i])+s.charCodeAt(i++);return g(e)}function g(t){return String.fromCharCode.apply(0,t)}if(f(s.random(),n),t.exports){t.exports=h;try{i=yo}catch(t){}}else s.seedrandom=h}("undefined"!=typeof self?self:An,[],Math)}(mo);var bo=io,wo=ao,ko=lo,vo=ho,So=po,xo=go,No=mo.exports;No.alea=bo,No.xor128=wo,No.xorwow=ko,No.xorshift7=vo,No.xor4096=So,No.tychei=xo;const Io=Ws({reverse_:function(t,e){const n={x:Ps(t,"x","reverse")},s={dims:e};return _s.runKernel($e,n,s)}});var Ao,zo;Ws({tensorScatterUpdate_:function(t,e,n){const s=Ps(t,"tensor","tensorScatterupdate"),i=Ps(e,"indices","tensorScatterupdate","int32"),r=Ps(n,"updates","tensorScatterupdate");if(Va(r,i,s.shape),s.dtype!==r.dtype)throw new Error(`tensor and updates must have the same dtype, instead they are ${s.dtype} and ${r.dtype}.`);const a={tensor:s,indices:i,updates:r};return _s.runKernel("TensorScatterUpdate",a,{})}}),function(t){t[t.NONE=0]="NONE",t[t.MEAN=1]="MEAN",t[t.SUM=2]="SUM",t[t.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"}(Ao||(Ao={})),"undefined"!=typeof requestAnimationFrame?requestAnimationFrame:"undefined"!=typeof setImmediate&&setImmediate,function(t){t[t.FIRST_DIM_SIZE=0]="FIRST_DIM_SIZE",t[t.VALUE_ROWIDS=1]="VALUE_ROWIDS",t[t.ROW_LENGTHS=2]="ROW_LENGTHS",t[t.ROW_SPLITS=3]="ROW_SPLITS",t[t.ROW_LIMITS=4]="ROW_LIMITS",t[t.ROW_STARTS=5]="ROW_STARTS"}(zo||(zo={}));function Eo(t,e,n){const s=t.shape.length,i=s-n.length,r=nr(n,s);let a=t;null!=r&&(a=lr(t,r));const o=a.shape.slice(),l=o.splice(s-n.length,n.length).reduce(((t,e)=>t*e),1);o.push(l);let u=function(t,e,n){const s=t.shape.slice();s[n]=1;const i=ii(e,s),r=Ya(t,n,!0,!1),a=Ya(t,n,!0,!0),o=qs(r,a);return qs(i,o)}(a.reshape(o),e,i);if(u=u.reshape(a.shape),null!=r){const t=sr(r);u=lr(u,t)}return u}const To={kernelName:Me,gradFunc:(t,e,n)=>{const{blockShape:s,paddings:i}=n;return{x:()=>Ha(t,s,i)}}},Co={kernelName:Be,gradFunc:(t,e,n)=>{const{axis:s}=n;return{x:()=>Ka(t,s)}}};const $o=[Ks,ti,ei,ai,oi,ui,hi,ci,di,fi,gi,mi,xi,Ii,zi,Ti,Ci,$i,Fi,Bi,Pi,Wi,Gi,Vi,Zi,Xi,tr,ur,pr,dr,{kernelName:we,inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e,i=si(n.shape,s.shape);return{a:()=>{const e=Hs(t,js(s,"float32")),r=ni(n.shape,i);return r.length>0?ii(ri(e,r),n.shape):e},b:()=>{let e=qs(t,js(n,"float32"));const r=ni(s.shape,i);r.length>0&&(e=ii(ri(e,r),s.shape));const a=Xs(s);return Js(Hs(e,js(a,"float32")))}}}},fr,mr,yr,br,wr,vr,kr,Nr,zr,Cr,$r,Fr,Dr,Lr,Rr,Or,Mr,Br,Ur,qr,qr,Kr,Hr,Zr,Qr,ta,ea,sa,ra,aa,oa,la,ua,ca,pa,pa,ga,ma,{kernelName:"Prod",inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{axis:i}=n;let r=[];return r=null==i?s.shape.map(((t,e)=>e)):"number"==typeof i?[i]:i,{x:()=>Eo(s,t,r)}}},{kernelName:"Reciprocal",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Hs(t,Js(Xs(n)))}}},{kernelName:"Relu6",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e,s=qs(Li(n,6),Vs(n));return{x:()=>qs(t,js(s,"float32"))}}},{kernelName:"Relu",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(t,js(Vs(n),"float32"))}}},{kernelName:Ce,inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>ii(t,n.shape)}}},{kernelName:"ResizeBilinear",inputsToSave:["images"],gradFunc:(t,e,n)=>{const[s]=e,i={dy:t,images:s};return{images:()=>_s.runKernel("ResizeBilinearGrad",i,n)}}},{kernelName:"ResizeNearestNeighbor",inputsToSave:["images"],gradFunc:(t,e,n)=>{const[s]=e,i={dy:t,images:s};return{images:()=>_s.runKernel("ResizeNearestNeighborGrad",i,n)}}},{kernelName:$e,gradFunc:(t,e,n)=>{const{dims:s}=n,i=Xe(s,t.shape);return{x:()=>Io(t,i)}}},{kernelName:"Round",gradFunc:t=>({x:()=>li(t)})},{kernelName:Fe,inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Js(Hs(t,qs(fa(n,1.5),2)))}}},{kernelName:De,inputsToSave:["condition"],gradFunc:(t,e)=>{const[n]=e;return{condition:()=>js(li(n),"float32"),t:()=>qs(t,js(n,t.dtype)),e:()=>qs(t,js(to(n),t.dtype))}}},{kernelName:"Selu",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>{const e=_r(n,Zs(0)),s=Zs(1.7580993408473768),i=Zs(1.0507009873554805),r=qs(t,i),a=qs(qs(t,s),gr(js(n,"float32")));return Mi(e,r,a)}}}},{kernelName:Re,outputsToSave:[!0],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(t,qs(n,Qs(Zs(1),n)))}}},{kernelName:"Sign",gradFunc:t=>({x:()=>li(t)})},{kernelName:"Sin",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(Ja(js(n,"float32")),t)}}},{kernelName:_e,inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(Za(js(n,"float32")),t)}}},{kernelName:Le,inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{begin:i,size:r}=n,a=s.shape,[o,l]=function(t,e,n){let s;const i=t.shape.length;let r;return s="number"==typeof e?[e,...new Array(i-1).fill(0)]:e.length<i?e.concat(new Array(i-e.length).fill(0)):e.slice(),s.forEach((t=>{Ge(-1!==t,(()=>"slice() does not support negative begin indexing."))})),r=null==n?new Array(i).fill(-1):"number"==typeof n?[n,...new Array(i-1).fill(-1)]:n.length<i?n.concat(new Array(i-n.length).fill(-1)):n,r=r.map(((e,n)=>e>=0?e:(Ge(-1===e,(()=>`Negative size values should be exactly -1 but got ${e} for the slice() size at index ${n}.`)),t.shape[n]-s[n]))),[s,r]}(s,i,r),u=[];for(let e=0;e<t.rank;e++)u.push([o[e],a[e]-o[e]-l[e]]);return{x:()=>no(t,u)}}},{kernelName:"Softmax",outputsToSave:[!0],gradFunc:(t,e,n)=>{const[s]=e,{dim:i}=n,r=qs(t,s);return{logits:()=>Qs(r,qs(ri(r,[i],true),s))}}},{kernelName:"Softplus",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(t,Ga(n))}}},To,To,Co,Co,{kernelName:Oe,inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Hs(t,qs(Ys(js(n,"float32")),2))}}},{kernelName:"SquaredDifference",inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e,i=Zs(2);return{a:()=>qs(t,qs(i,Qs(n,s))),b:()=>qs(t,qs(i,Qs(s,n)))}}},{kernelName:"Square",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(t,qs(js(n,"float32"),2))}}},{kernelName:Ve,gradFunc:t=>({x:()=>li(t)})},{kernelName:"Sub",inputsToSave:["a","b"],gradFunc:(t,e)=>{const[n,s]=e,i=si(n.shape,s.shape);return{a:()=>{let e=t;const s=ni(n.shape,i);return s.length>0&&(e=ri(e,s)),ii(e,n.shape)},b:()=>{let e=t;const n=ni(s.shape,i);return n.length>0&&(e=ri(e,n)),ii(Js(e),s.shape)}}}},{kernelName:"Sum",inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,i=s.shape.slice(),{axis:r}=n;Xe(r,s.shape).forEach((t=>{i[t]=1}));const a=ii(t,i),o=qs(a,Xr(s.shape,"float32"));return{x:()=>o}}},{kernelName:"Tan",inputsToSave:["x"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>Hs(t,Xs(Ja(n)))}}},{kernelName:"Tanh",outputsToSave:[!0],gradFunc:(t,e)=>{const[n]=e;return{x:()=>qs(Qs(Zs(1),Xs(n)),t)}}},{kernelName:Pe,inputsToSave:["x"],gradFunc:(t,e,n)=>{const[s]=e,{reps:i}=n;return{x:()=>{let e=li(s);if(1===s.rank)for(let n=0;n<i[0];++n)e=pi(e,na(t,[n*s.shape[0]],[s.shape[0]]));else if(2===s.rank)for(let n=0;n<i[0];++n)for(let r=0;r<i[1];++r)e=pi(e,na(t,[n*s.shape[0],r*s.shape[1]],[s.shape[0],s.shape[1]]));else if(3===s.rank)for(let n=0;n<i[0];++n)for(let r=0;r<i[1];++r)for(let a=0;a<i[2];++a)e=pi(e,na(t,[n*s.shape[0],r*s.shape[1],a*s.shape[2]],[s.shape[0],s.shape[1],s.shape[2]]));else{if(4!==s.rank)throw new Error(`Gradient for tile operation is not implemented for rank-${s.rank} tensors yet.`);for(let n=0;n<i[0];++n)for(let r=0;r<i[1];++r)for(let a=0;a<i[2];++a)for(let o=0;o<i[3];++o)e=pi(e,na(t,[n*s.shape[0],r*s.shape[1],a*s.shape[2],o*s.shape[3]],[s.shape[0],s.shape[1],s.shape[2],s.shape[3]]))}return e}}}},{kernelName:Ue,gradFunc:(t,e,n)=>{const s=n,{perm:i}=s,r=sr(i);return{x:()=>lr(t,r)}}},{kernelName:We,gradFunc:(t,e,n)=>{const s=n,{axis:i}=s;return{value:()=>Ir(t,i)}}},{kernelName:je,inputsToSave:["segmentIds"],gradFunc:(t,e)=>{const[n]=e;return{x:()=>function(t,e){const n=eo(e,li(e)),s=Qa(t,n);let i=Di(e,Zs(0,"int32"));const r=s.rank-i.rank;for(let t=0;t<r;++t)i=Xa(i,t+1);i=_i(i,Xr(s.shape,"bool"));const a=li(s);return Mi(i,s,a)}(t,n)}}},{kernelName:qe,gradFunc:t=>({x:()=>li(t)})}];for(const t of $o)In(t);function Fo(t,n){return e.tidy((()=>i.sqrt(i.sum(i.mul(t,t),n,!0))))}class Do extends e.serialization.Serializable{getConfig(){return{}}}class Lo extends Do{constructor(t){super(),this.defaultMaxValue=2,this.defaultAxis=0,this.maxValue=null!=t.maxValue?t.maxValue:this.defaultMaxValue,this.axis=null!=t.axis?t.axis:this.defaultAxis}apply(t){return e.tidy((()=>{const e=Fo(t,this.axis),n=i.clipByValue(e,0,this.maxValue);return i.mul(t,i.div(n,i.add(tt(),e)))}))}getConfig(){return{maxValue:this.maxValue,axis:this.axis}}}Lo.className="MaxNorm",e.serialization.registerClass(Lo);class _o extends Do{constructor(t){super(),this.defaultAxis=0,this.axis=null!=t.axis?t.axis:this.defaultAxis}apply(t){return e.tidy((()=>i.div(t,i.add(tt(),Fo(t,this.axis)))))}getConfig(){return{axis:this.axis}}}_o.className="UnitNorm",e.serialization.registerClass(_o);class Ro extends Do{apply(t){return i.relu(t)}}Ro.className="NonNeg",e.serialization.registerClass(Ro);class Oo extends Do{constructor(t){super(),this.defaultMinValue=0,this.defaultMaxValue=1,this.defaultRate=1,this.defaultAxis=0,this.minValue=null!=t.minValue?t.minValue:this.defaultMinValue,this.maxValue=null!=t.maxValue?t.maxValue:this.defaultMaxValue,this.rate=null!=t.rate?t.rate:this.defaultRate,this.axis=null!=t.axis?t.axis:this.defaultAxis}apply(t){return e.tidy((()=>{const e=Fo(t,this.axis),n=i.add(i.mul(this.rate,i.clipByValue(e,this.minValue,this.maxValue)),i.mul(1-this.rate,e));return i.mul(t,i.div(n,i.add(tt(),e)))}))}getConfig(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}}}Oo.className="MinMaxNorm",e.serialization.registerClass(Oo);const Mo={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function Bo(t){return w(t)}function Po(t,n={}){return v(t,e.serialization.SerializationMap.getMap().classNameMap,n,"constraint")}function Uo(t){if(null==t)return null;if("string"==typeof t){return Po({className:t in Mo?Mo[t]:t,config:{}})}return t instanceof Do?t:Po(t)}var Wo={__proto__:null,maxNorm:function(t){return new Lo(t)},minMaxNorm:function(t){return new Oo(t)},nonNeg:function(){return new Ro},unitNorm:function(t){return new _o(t)}};var jo,qo={__proto__:null,constant:function(t){return new St(t)},glorotNormal:function(t){return new Tt(t)},glorotUniform:function(t){return new Et(t)},heNormal:function(t){return new Ct(t)},heUniform:function(t){return new $t(t)},identity:function(t){return new At(t)},leCunNormal:function(t){return new Ft(t)},leCunUniform:function(t){return new Dt(t)},ones:function(){return new vt},orthogonal:function(t){return new Lt(t)},randomNormal:function(t){return new Nt(t)},randomUniform:function(t){return new xt(t)},truncatedNormal:function(t){return new It(t)},varianceScaling:function(t){return new zt(t)},zeros:function(){return new kt}};async function Vo(t){if(null==t)return;const n=[],s=[],i=[];for(const e in t){const r=t[e];if("number"!=typeof r){const t=r;n.push(t.data()),s.push(e),i.push(t)}}if(n.length>0){const r=await Promise.all(n);for(let e=0;e<r.length;++e)t[s[e]]=r[e][0];e.dispose(i)}}function Ko(t){if(null!=t)for(const e in t){const n=t[e];"number"!=typeof n&&n.dispose()}}!function(t){t[t.SILENT=0]="SILENT",t[t.VERBOSE=1]="VERBOSE"}(jo||(jo={}));class Go{constructor(){this.validationData=null}setParams(t){this.params=t}async onEpochBegin(t,e){}async onEpochEnd(t,e){}async onBatchBegin(t,e){}async onBatchEnd(t,e){}async onTrainBegin(t){}async onTrainEnd(t){}setModel(t){}}class Ho{constructor(t,e=10){null==t&&(t=[]),this.callbacks=t,this.queueLength=e}append(t){this.callbacks.push(t)}setParams(t){for(const e of this.callbacks)e.setParams(t)}setModel(t){for(const e of this.callbacks)e.setModel(t)}async onEpochBegin(t,e){null==e&&(e={});for(const n of this.callbacks)await n.onEpochBegin(t,e)}async onEpochEnd(t,e){null==e&&(e={});for(const n of this.callbacks)await n.onEpochEnd(t,e)}async onBatchBegin(t,e){null==e&&(e={});for(const n of this.callbacks)await n.onBatchBegin(t,e)}async onBatchEnd(t,e){null==e&&(e={});for(const n of this.callbacks)await n.onBatchEnd(t,e)}async onTrainBegin(t){null==t&&(t={});for(const e of this.callbacks)await e.onTrainBegin(t)}async onTrainEnd(t){null==t&&(t={});for(const e of this.callbacks)await e.onTrainEnd(t)}}class Jo extends Go{constructor(){super()}async onEpochBegin(t){this.seen=0,this.totals={}}async onBatchEnd(t,n){null==n&&(n={});const s=null==n.size?0:n.size;this.seen+=s;for(const t in n){const i=n[t];if("number"==typeof i)this.totals.hasOwnProperty(t)||(this.totals[t]=0),this.totals[t]=this.totals[t]+i*s;else{let n;t in this.totals?n=this.totals[t]:this.totals[t]=0;const r=e.tidy((()=>e.add(this.totals[t],e.mul(i,s))));this.totals[t]=r,null!=n&&n.dispose()}}}async onEpochEnd(t,n){if(null!=n)for(const t of this.params.metrics)null!=this.totals[t]&&("number"==typeof this.totals[t]?n[t]=this.totals[t]/this.seen:e.tidy((()=>{const s=e.mul(e.div(1,this.seen),this.totals[t]);n[t]=s,this.totals[t].dispose(),e.keep(n[t])})))}}class Zo extends Go{async onTrainBegin(t){this.epoch=[],this.history={}}async onEpochEnd(t,e){null==e&&(e={}),this.epoch.push(t);for(const t in e)null==this.history[t]&&(this.history[t]=[]),this.history[t].push(e[t])}async syncData(){const t=[],e=[],n=[];for(const s in this.history){const i=this.history[s];for(let r=0;r<i.length;++r)if("number"!=typeof i[r]){const a=i[r];t.push(a.data()),e.push(s),n.push(r)}}const s=await Promise.all(t);for(let t=0;t<s.length;++t){this.history[e[t]][n[t]].dispose(),this.history[e[t]][n[t]]=s[t][0]}}}class Yo extends Go{constructor(t,n){if(super(),this.currentEpoch=0,this.nowFunc=t.nowFunc,this.nextFrameFunc=t.nextFrameFunc||e.nextFrame,this.yieldEvery=n||"auto","auto"===this.yieldEvery&&(this.yieldEvery=125),"never"===this.yieldEvery&&null!=t.onYield)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");e.util.isNumber(this.yieldEvery)&&(this.maybeWait=function(t,n,s){let i,r=null!=s?s():e.util.now();return(...a)=>{const o=null!=s?s():e.util.now();return o-r<n||(r=o,i=t(...a)),i}}(this.maybeWait.bind(this),this.yieldEvery,this.nowFunc)),this.trainBegin=t.onTrainBegin,this.trainEnd=t.onTrainEnd,this.epochBegin=t.onEpochBegin,this.epochEnd=t.onEpochEnd,this.batchBegin=t.onBatchBegin,this.batchEnd=t.onBatchEnd,this.yield=t.onYield}async maybeWait(t,e,n){const s=[];null!=this.yield&&(await Vo(n),s.push(this.yield(t,e,n))),s.push(this.nextFrameFunc()),await Promise.all(s)}async onEpochBegin(t,e){this.currentEpoch=t,null!=this.epochBegin&&(await Vo(e),await this.epochBegin(t,e))}async onEpochEnd(t,e){const n=[];null!=this.epochEnd&&(await Vo(e),n.push(this.epochEnd(t,e))),"epoch"===this.yieldEvery&&n.push(this.nextFrameFunc()),await Promise.all(n)}async onBatchBegin(t,e){null!=this.batchBegin&&(await Vo(e),await this.batchBegin(t,e))}async onBatchEnd(t,n){const s=[];null!=this.batchEnd&&(await Vo(n),s.push(this.batchEnd(t,n))),"batch"===this.yieldEvery?s.push(this.nextFrameFunc()):e.util.isNumber(this.yieldEvery)&&s.push(this.maybeWait(this.currentEpoch,t,n)),await Promise.all(s)}async onTrainBegin(t){null!=this.trainBegin&&(await Vo(t),await this.trainBegin(t))}async onTrainEnd(t){null!=this.trainEnd&&(await Vo(t),await this.trainEnd(t))}}function Xo(t,e){if(null==t&&(t={}),t instanceof Go)return[t];if(Array.isArray(t)&&t[0]instanceof Go)return t;return g(t).map((t=>new Yo(t,e)))}class Qo{constructor(){}static registerCallbackConstructor(t,n){e.util.assert(t>=0&&Number.isInteger(t),(()=>`Verbosity level is expected to be an integer >= 0, but got ${t}`)),Qo.checkForDuplicate(n),null==Qo.constructors[t]&&(Qo.constructors[t]=[]),Qo.constructors[t].push(n)}static checkForDuplicate(t){for(const e in Qo.constructors){Qo.constructors[+e].forEach((e=>{if(e===t)throw new o("Duplicate callback constructor.")}))}}static clear(){Qo.constructors={}}static createCallbacks(t){const e=[];for(const n in Qo.constructors){const s=+n;t>=s&&e.push(...Qo.constructors[s])}return e.map((t=>new t))}}function tl(t,e,n,s,i,r,a,o,l){const u=new Zo,h=[new Jo,...Qo.createCallbacks(e)];null!=t&&h.push(...t),h.push(u);const c=new Ho(h);return c.setParams({epochs:n,initialEpoch:s,samples:i,steps:r,batchSize:a,verbose:e,doValidation:o,metrics:l}),{callbackList:c,history:u}}function el(t,n={},s=!1){return v(t,e.serialization.SerializationMap.getMap().classNameMap,n,"layer",s)}function nl(t,n){return e.tidy((()=>{"float32"!==t.dtype&&(t=i.cast(t,"float32"));const e=i.sum(pt(t),n,!0),s=i.fill(e.shape,tt()),r=i.sqrt(i.maximum(e,s));return i.div(t,r)}))}function sl(t,n){return e.tidy((()=>i.mean(pt(i.sub(n,t)),-1)))}function il(t,n){return e.tidy((()=>i.mean(i.abs(i.sub(n,t)),-1)))}function rl(t,n){return e.tidy((()=>{const e=i.sub(t,n),s=i.clipByValue(i.abs(t),tt(),Number.MAX_VALUE),r=i.abs(i.div(e,s));return i.mul(100,i.mean(r,-1))}))}function al(t,n,s=!1){return e.tidy((()=>{if(s)n=i.softmax(n);else{const t=i.sum(n,n.shape.length-1,!0);n=i.div(n,t)}return n=i.clipByValue(n,tt(),1-tt()),i.neg(i.sum(i.mul(i.cast(t,"float32"),i.log(n)),n.shape.length-1))}))}function ol(t,n,s=!1){return e.tidy((()=>{const e=i.cast(i.floor(function(t){const e=[J(t.shape)];return i.reshape(t,e)}(t)),"int32"),r=(n=i.clipByValue(n,tt(),1-tt())).shape;return al(i.reshape(i.oneHot(e,r[r.length-1]),r),n,s)}))}function ll(t,n){return e.tidy((()=>{let s;return s=i.clipByValue(n,tt(),1-tt()),s=i.log(i.div(s,i.sub(1,s))),i.mean(function(t,n){if(!e.util.arraysEqual(t.shape,n.shape))throw new o(`logits and labels must have the same shape, but got shapes ${JSON.stringify(t.shape)} and ${JSON.stringify(n.shape)}`);return e.tidy((()=>{const e=i.relu(n),s=i.neg(i.abs(n));return i.add(i.sub(e,i.mul(n,t)),i.log1p(i.exp(s)))}))}(t,s),-1)}))}function ul(t,n){return e.tidy((()=>{const e=nl(t,-1),s=nl(n,-1),r=i.mul(e,s);return i.neg(i.sum(r,-1))}))}Qo.constructors={};const hl={meanSquaredError:sl,meanAbsoluteError:il,meanAbsolutePercentageError:rl,meanSquaredLogarithmicError:function(t,n){return e.tidy((()=>{const e=i.clipByValue(n,tt(),Number.MAX_VALUE),s=i.log(i.add(1,e)),r=i.clipByValue(t,tt(),Number.MAX_VALUE),a=i.log(i.add(1,r));return i.mean(pt(i.sub(s,a)),-1)}))},squaredHinge:function(t,n){return e.tidy((()=>{const e=i.maximum(0,i.sub(1,i.mul(t,n)));return i.mean(pt(e),-1)}))},hinge:function(t,n){return e.tidy((()=>{const e=i.maximum(0,i.sub(1,i.mul(t,n)));return i.mean(e,-1)}))},categoricalHinge:function(t,n){return e.tidy((()=>{const e=i.sum(i.mul(t,n),-1),s=i.max(i.mul(i.sub(1,t),n),-1);return i.maximum(0,i.add(1,i.sub(s,e)))}))},logcosh:function(t,n){return e.tidy((()=>{const e=Math.log(2),s=i.sub(n,t),r=i.sub(i.add(s,i.softplus(i.mul(-2,s))),e);return i.mean(r,-1)}))},categoricalCrossentropy:al,sparseCategoricalCrossentropy:ol,binaryCrossentropy:ll,kullbackLeiblerDivergence:function(t,n){return e.tidy((()=>{const e=i.clipByValue(t,tt(),1),s=i.clipByValue(n,tt(),1);return i.sum(i.mul(t,i.log(i.div(e,s))),-1)}))},poisson:function(t,n){return e.tidy((()=>{const e=i.log(i.add(tt(),n));return i.mean(i.sub(n,i.mul(t,e)),-1)}))},cosineProximity:ul};function cl(t){if("string"==typeof t){if(t in hl)return hl[t];let e=`Unknown loss ${t}`;throw t.toLowerCase().includes("softmaxcrossentropy")&&(e=`Unknown loss ${t}. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy`),new o(e)}return t}function pl(t,n){return e.tidy((()=>{const e=i.mul(.5,i.onesLike(n)),s=et(i.greater(n,e),t.dtype);return i.mean(i.equal(t,s),-1)}))}function dl(t,n){return e.tidy((()=>et(i.equal(i.argMax(t,-1),i.argMax(n,-1)),"float32")))}function fl(t,n){return e.tidy((()=>i.cast(i.sum(i.logicalAnd(i.equal(t,1),i.equal(n,1))),"float32")))}function gl(t,n){return e.tidy((()=>{const s=fl(t,n),r=function(t,n){return e.tidy((()=>i.cast(i.sum(i.logicalAnd(i.equal(t,0),i.equal(n,1))),"float32")))}(t,n),a=i.add(s,r);return i.cast(i.where(i.greater(a,0),i.div(s,a),0),"float32")}))}function ml(t,n){return e.tidy((()=>{const s=fl(t,n),r=function(t,n){return e.tidy((()=>i.cast(i.sum(i.logicalAnd(i.equal(t,1),i.equal(n,0))),"float32")))}(t,n),a=i.add(s,r);return i.cast(i.where(i.greater(a,0),i.div(s,a),0),"float32")}))}function yl(t,e){return ll(t,e)}function bl(t,e){return t.rank===e.rank&&(t=i.squeeze(t,[t.rank-1])),(e=i.argMax(e,-1)).dtype!==t.dtype&&(e=i.cast(e,t.dtype)),i.cast(i.equal(t,e),"float32")}const wl=al,kl=ol,vl={binaryAccuracy:pl,categoricalAccuracy:dl,precision:gl,categoricalCrossentropy:wl,sparseCategoricalCrossentropy:kl,mse:sl,MSE:sl,mae:il,MAE:il,mape:rl,MAPE:rl,cosine:ul};function Sl(t){if("string"==typeof t&&t in vl)return vl[t];if("string"!=typeof t&&null!=t)return t;throw new o(`Unknown metric ${t}`)}function xl(t){if(p(null!==t,`Unknown LossOrMetricFn ${t}`),"string"==typeof t)return t;{let e;for(const n of Object.keys(hl))if(hl[n]===t){e=n;break}if(void 0!==e)return e;for(const n of Object.keys(vl))if(vl[n]===t){e=n;break}return void 0!==e?e:t.name}}const Nl=1048576;function Il(t,e,n=!1){if(null==t||"object"!=typeof t||Object.getPrototypeOf(t)!==Object.prototype||!Al(t))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(n){const n=JSON.stringify(t);n.length>Nl&&console.warn(`User-defined metadata of model "${e}" is too large in size (length=${n.length} when serialized). It is not recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= 1048576.`)}}function Al(t){if(null===t)return!0;if("object"==typeof t){if(Object.getPrototypeOf(t)===Object.prototype){const e=Object.keys(t);for(const n of e){if("string"!=typeof n)return!1;if(!Al(t[n]))return!1}return!0}if(Array.isArray(t)){for(const e of t)if(!Al(e))return!1;return!0}return!1}{const e=typeof t;return"string"===e||"number"===e||"boolean"===e}}function zl(t,e,n,s=console.log){const i=function(t){let e=!0;const n=[],s=[];for(const e in t.nodesByDepth)n.push(t.nodesByDepth[e]);for(const t of n){if(t.length>1||1===t.length&&t[0].inboundLayers.length>1){e=!1;break}s.push(...t)}if(e)for(const n of t.layers){let t=!1;for(const i of n.inboundNodes)if(-1!==s.indexOf(i)){if(t){e=!1;break}t=!0}if(!e)break}return e}(t),r=["Layer (type)","Input Shape","Output shape","Param #"];let a;if(i?(e=e||90,n=n||[.32,.61,.89,1]):(e=e||115,n=n||[.24,.48,.7,.8,1]),n[n.length-1]<=1&&(n=n.map((t=>Math.floor(e*t)))),!i){r.push("Receives inputs"),a=[];for(const e in t.nodesByDepth)a.push(...t.nodesByDepth[e])}s("_".repeat(e)),El(r,n,s),s("=".repeat(e));const o=t.layers;for(let t=0;t<o.length;++t)i?Tl(o[t],n,s):Cl(o[t],n,a,s),s((t===o.length-1?"=":"_").repeat(e));t.checkTrainableWeightsConsistency();const l=function(t){let e;e=null!=t.collectedTrainableWeights?jt(t.collectedTrainableWeights):jt(t.trainableWeights);return e}(t),u=jt(t.nonTrainableWeights);s(`Total params: ${l+u}`),s(`Trainable params: ${l}`),s(`Non-trainable params: ${u}`),s("_".repeat(e))}function El(t,e,n=console.log){let s="";for(let n=0;n<t.length;++n)n>0&&(s=s.slice(0,s.length-1)+" "),s+=t[n],s=s.slice(0,e[n]),s+=" ".repeat(e[n]-s.length);n(s)}function Tl(t,e,n){let s,i;try{i=t.inboundNodes.map((t=>JSON.stringify(t.inputShapes))).join(",")}catch(t){i="multiple"}try{s=JSON.stringify(t.outputShape)}catch(t){s="multiple"}El([`${t.name} (${t.getClassName()})`,i,s,t.countParams().toString()],e,n)}function Cl(t,e,n,s){let i,r;try{r=t.inboundNodes.map((t=>JSON.stringify(t.inputShapes))).join(",")}catch(t){r="multiple"}try{i=JSON.stringify(t.outputShape)}catch(t){i="multiple"}const a=[];for(const e of t.inboundNodes)if(!(null!=n&&n.length>0&&-1===n.indexOf(e)))for(let t=0;t<e.inboundLayers.length;++t){const n=e.inboundLayers[t].name,s=e.nodeIndices[t],i=e.tensorIndices[t];a.push(`${n}[${s}][${i}]`)}const o=t.name,l=t.getClassName(),u=0===a.length?"":a[0];El([`${o} (${l})`,r,i,t.countParams().toString(),u],e,s);for(let t=1;t<a.length;++t)El(["","","","",a[t]],e,s)}function $l(t,e,n){return("inboundNodes"===t||"outputLayers"===t||"inputLayers"===t)&&0===e&&"string"==typeof n}function Fl(t,e){if(null===t)return null;if("string"==typeof t)return y(t);if("number"==typeof t||"boolean"==typeof t)return t;if(t instanceof Array){const n=[],s=t.length;for(let i=0;i<s;++i){const s=t[i];$l(e,i,s)?n.push(s):n.push(Fl(s,e))}return n}{const e={};for(const n of Object.keys(t)){const s=t[n];if("name"===n&&"string"==typeof s)e[n]=s;else{const t=y(n);e[t]=Fl(s,t)}}return e}}function Dl(t,e){if(null==t)return null;if("string"==typeof t)return m(t);if("number"==typeof t||"boolean"==typeof t)return t;if(t instanceof Array){const n=[],s=t.length;for(let i=0;i<s;++i){const s=t[i];$l(e,i,s)?n.push(s):n.push(Dl(s,e))}return n}{const e={};for(const n of Object.keys(t)){const s=t[n],i=m(n);e[i]="name"!==n&&"className"!==n||"string"!=typeof s?Dl(s,n):s}return e}}const Ll="4.15.0";class _l extends Qt{constructor(t){if(super({}),this.containerNodes=new Set,this.name=t.name,null==this.name){const t=this.getClassName().toLowerCase();this.name=D(t)}if(this.supportsMasking=!1,this.trainable_=!0,Array.isArray(t.inputs)?this.inputs=t.inputs.slice():this.inputs=[t.inputs],Array.isArray(t.outputs)?this.outputs=t.outputs.slice():this.outputs=[t.outputs],x(this.inputs).length!==this.inputs.length)throw new o(`The list of inputs passed to the model is redundant. All inputs should only appear once. Found: ${this.inputs.map((t=>t.name))}`);x(this.outputs).length!==this.outputs.length&&console.warn(`The list of outputs passed to the model is redundant. All outputs should only appear once. Found: ${this.outputs.map((t=>t.name))}`),this.inputLayers=[],this.inputLayersNodeIndices=[],this.inputLayersTensorIndices=[],this.outputLayers=[],this.outputLayersNodeIndices=[],this.outputLayersTensorIndices=[],this.layers=[],this.internalContainerRefs=[];for(const t of this.outputs){const e=t.sourceLayer,n=t.nodeIndex,s=t.tensorIndex;this.outputLayers.push(e),this.outputLayersNodeIndices.push(n),this.outputLayersTensorIndices.push(s)}for(const t of this.inputs){const e=t.sourceLayer,n=t.nodeIndex,s=t.tensorIndex;p(0===n,"input layer has >1 nodes"),p(0===s,"input layer has >1 tensors"),this.inputLayers.push(e),this.inputLayersNodeIndices.push(n),this.inputLayersTensorIndices.push(s)}this.inputNames=[],this.outputNames=[],this.feedInputShapes=[],this.feedInputNames=[],this.feedOutputNames=[];for(let e=0;e<this.inputLayers.length;e++){const n=this.inputLayers[e];if(!(n instanceof ee))throw new TypeError(`Input layers to a LayersModel must be InputLayer objects. Received inputs: ${t.inputs}. Input ${e} (0-based) originates from layer type ${n.getClassName()}.`);this.inputNames.push(n.name),this.feedInputShapes.push(n.batchInputShape),this.feedInputNames.push(n.name)}for(const t of this.outputLayers)this.outputNames.push(t.name);this.internalInputShapes=this.inputs.map((t=>t.shape)),this.internalOutputShapes=this.outputs.map((t=>t.shape));const e={},n={},s={},i={},r={},l=[],u=(t,e,n,s,i,o)=>{null!=s&&null!=i&&null!=o||(s=t.sourceLayer,i=t.nodeIndex,o=t.tensorIndex);const h=s.inboundNodes[i];if(-1!==n.indexOf(h))throw new a(`The tensor ${t.name} at layer "${s.name}" is part of a cycle.`);if(-1!==e.indexOf(h))return;this.containerNodes.add(_l.nodeKey(s,i)),s.id in r||(r[s.id]=Object.keys(r).length),-1===n.indexOf(h)&&n.push(h);const c=h.inboundLayers.length;for(let t=0;t<c;t++){const s=h.inputTensors[t],i=h.inboundLayers[t],r=h.nodeIndices[t],a=h.tensorIndices[t];u(s,e,n,i,r,a)}for(e.push(h);n.indexOf(h)>=0;)n.splice(n.indexOf(h),1);l.push(h)},h=[],c=[];for(const t of this.outputs)u(t,h,c);const d=l.slice().reverse();for(const t of d){n[t.id]=t,t.id in e||(e[t.id]=0);let r=e[t.id];const a=null==s[t.outboundLayer.id]?0:s[t.outboundLayer.id];r=Math.max(r,a),s[t.outboundLayer.id]=r,i[t.outboundLayer.id]=t.outboundLayer,e[t.id]=r;for(let s=0;s<t.inboundLayers.length;s++){const i=t.inboundLayers[s],a=t.nodeIndices[s],o=i.inboundNodes[a],l=null==e[o.id]?0:e[o.id];e[o.id]=Math.max(r+1,l),n[o.id]=o}}const f={};for(const t in e){const s=e[t];s in f||(f[s]=[]),f[s].push(n[t])}const g={};for(const t in s){const e=s[t];e in g||(g[e]=[]),g[e].push(i[t])}let m=Object.keys(g).map((t=>parseInt(t,10))).sort(S);this.layers=[];for(const t of m){const e=g[t];e.sort(((t,e)=>{const n=r[t.id],s=r[e.id];return n<s?-1:n>s?1:0}));for(const t of e)t instanceof _l&&this.internalContainerRefs.push(t),this.layers.push(t)}this.layersByDepth=g,m=Object.keys(f).map((t=>parseInt(t,10))).sort(S);const y=this.inputs.slice(),b=[];for(const t of m)for(const e of f[t]){const t=e.outboundLayer;if(null!=t){for(const n of e.inputTensors)if(-1===y.indexOf(n))throw new a(`Graph disconnected: cannot obtain value for tensor ${n} at layer "${t.name}". The following previous layers were accessed without issue: ${b}`);for(const t of e.outputTensors)y.push(t);b.push(t.name)}}this.nodesByDepth=f;const w=this.layers.map((t=>t.name));for(const t of w){const e=w.filter((e=>e===t)).length;if(1!==e)throw new a(`The name "${t}" is used ${e} times in the model. All layer names should be unique. Layer names: `+JSON.stringify(w))}this.outboundNodes=[],this.inboundNodes=[],new Yt({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:this.inputs.map((t=>null)),outputMasks:this.outputs.map((t=>null)),inputShapes:this.inputs.map((t=>t.shape)),outputShapes:this.outputs.map((t=>t.shape))}),this.built=!0,this._refCount=1}assertNotDisposed(){if(0===this._refCount)throw new Error(`Container '${this.name}' is already disposed.`)}dispose(){this.assertNotDisposed();const t={refCountAfterDispose:null,numDisposedVariables:0};if(0==--this._refCount){for(const e of this.layers)t.numDisposedVariables+=e.dispose().numDisposedVariables;for(const e of this.internalContainerRefs)t.numDisposedVariables+=e.dispose().numDisposedVariables}return t.refCountAfterDispose=this._refCount,t}get trainable(){return this.trainable_}set trainable(t){this.layers.forEach((e=>{e._trainableWeights.forEach((e=>e.trainable=t))})),this.trainable_=t}get trainableWeights(){if(this._trainableWeights.length>0)throw new o("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[];let t=[];for(const e of this.layers)t=t.concat(e.trainableWeights);return t}get nonTrainableWeights(){const t=[];for(const e of this.layers)t.push(...e.nonTrainableWeights);if(!this.trainable){const e=[];for(const t of this.layers)e.push(...t.trainableWeights);return e.concat(t)}return t}get weights(){return this.trainableWeights.concat(this.nonTrainableWeights)}loadWeights(t,e=!0){const n={};let s=0;const i=(t=>{const e=Object.keys(t);if(0===e.length)return!1;const n=e[0].split("/");return!isNaN(parseInt(n[n.length-1],10))})(t);i&&this.parseWeights(t);for(const t of this.layers)for(const[e,r]of t.weights.entries()){const t=i?`${r.name.split("/").slice(0,-1).join("/")+"/"}${e}`:r.originalName;if(null!=n[t])throw new o(`Duplicate weight name: ${t}`);n[t]=r,s++}const r=[];for(const s in t){let i=s;if(null==n[s]){const t=s.split("/");i=t.slice(0,-2).concat([t[t.length-1]]).join("/")}if(null!=n[i])r.push([n[i],t[s]]);else if(e)throw new o(`Provided weight data has no target variable: ${s}`);delete n[i]}if(e){const t=[];for(const e in n)t.push(e);if(t.length>0)throw new o(`${t.length} of ${s} weights are not set: ${t}`)}Gt(r)}parseWeights(t){for(const e in Object.keys(t)){const n=e.split("/"),s=["vars","layer_checkpoint_dependencies"],i=n.map((t=>t.startsWith("_")?t.slice(1):t)).filter((t=>!s.includes(t))).join("/");i!==e&&(t[i]=t[e],delete t[e])}}updatedConfig(){const t=this.getConfig(),e={};return e.className=this.getClassName(),e.config=t,e.kerasVersion="tfjs-layers 4.15.0",e.backend="TensorFlow.js",e}toJSON(t,e=!0){const n=Dl(this.updatedConfig());return e?JSON.stringify(n):n}call(t,n){return e.tidy((()=>{t=g(t);const e=new se;for(let n=0;n<this.inputs.length;++n)e.add(this.inputs[n],t[n]);return ae(this.outputs,e,n)}))}computeMask(t,n){return e.tidy((()=>{let e;return t=g(t),e=null==n?c(null,t.length):g(n),this.runInternalGraph(t,e)[1]}))}computeOutputShape(t){const e=Pt(t);if(e.length!==this.inputLayers.length)throw new o(`Invalid inputShape argument ${t}: model has ${this.inputLayers.length} tensor inputs.`);const n={};for(let t=0;t<e.length;t++){const s=this.inputLayers[t],i=e[t];n[s.name+"_0_0"]=i}const s=Object.keys(this.nodesByDepth).map((t=>parseInt(t,10))).sort(S);if(s.length>1)for(const t of s){const e=this.nodesByDepth[t];for(const t of e){const e=t.outboundLayer;if(-1!==this.inputLayers.map((t=>t.id)).indexOf(e.id))continue;const s=[];for(let e=0;e<t.inboundLayers.length;e++){const i=t.inboundLayers[e],r=t.nodeIndices[e],a=t.tensorIndices[e],o=n[`${i.name}_${r}_${a}`];s.push(o)}const i=Pt(e.computeOutputShape(f(s))),r=e.inboundNodes.indexOf(t);for(let t=0;t<i.length;t++){n[`${e.name}_${r}_${t}`]=i[t]}}}const i=[],r=[];for(let t=0;t<this.outputLayers.length;t++){const e=this.outputLayers[t],n=this.outputLayersNodeIndices[t],s=this.outputLayersTensorIndices[t],i=`${e.name}_${n}_${s}`;r.push(i)}for(let t=0;t<r.length;t++){const e=r[t];p(e in n),i.push(n[e])}return f(i)}runInternalGraph(t,e){null==e&&(e=c(null,t.length));const n={};for(let s=0;s<this.inputs.length;++s){const i=this.inputs[s],r=t[s],a=e[s];n[i.id]=[r,a]}const s=Object.keys(this.nodesByDepth).map((t=>parseInt(t,10))).sort(S);for(const t of s){const e=this.nodesByDepth[t];for(const t of e){const e=t.outboundLayer,s=t.inputTensors,i=t.outputTensors,r=new Array;for(const t of s)t.id in n&&r.push(n[t.id]);if(r.length===s.length){let s,a,o,u,h={};if(null!=t.callArgs&&(h=t.callArgs),1===r.length){const[t,n]=r[0];null==h.mask&&(h.mask=n),o=g(e.call(t,h)),u=g(e.computeMask(t,n)),s=[t],a=[n]}else s=r.map((t=>t[0])),a=r.map((t=>t[1])),null==h.mask&&(h.mask=a),o=g(e.call(s,h)),u=g(e.computeMask(s,a));if(e.activityRegularizer)throw new l("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(let t=0;t<i.length;++t){const e=i[t],s=o[t],r=u[t];n[e.id]=[s,r]}}}}const i=[],r=[],a=[];for(const t of this.outputs){p(t.id in n,`Could not compute output ${t.name} : ${t.id}`);const[e,s]=n[t.id];a.push(e.shape),i.push(e),r.push(s)}return[i,r,a]}buildNodeConversionMap(t){const e={};let n;for(const t of this.layers){n=t instanceof _l?1:0;for(let s=0;s<t.inboundNodes.length;s++){const i=_l.nodeKey(t,s);this.containerNodes.has(i)&&(e[i]=n,n+=1)}}return e}getLayer(t,e){if(null!=e)return this.findLayer(e);if(null==t)throw new o("Provide either a layer name or layer index");if("number"==typeof t)return this.findLayer(t);for(const e of this.layers)if(e.name===t)return e;throw new o(`No such layer: ${t}`)}findLayer(t){if(this.layers.length<=t)throw new o(`Was asked to retrieve layer at index ${t}, but model only has ${this.layers.length} layer(s).`);return this.layers[t]}calculateLosses(){return e.tidy((()=>{const t=[];for(const e of this.layers)for(let n=0;n<e.inboundNodes.length;++n){const s=_l.nodeKey(e,n);this.containerNodes.has(s)&&t.push(...e.calculateLosses())}return t}))}getConfig(){const t={name:this.name},e=this.buildNodeConversionMap(this.layers),n=[];for(const t of this.layers){const s=t.getClassName(),i=t.getConfig(),r=[];for(let n=0;n<t.inboundNodes.length;n++){const s=t.inboundNodes[n],i=_l.nodeKey(t,n);let a={};if(this.containerNodes.has(i)){if(s.callArgs)try{JSON.stringify(s.callArgs),a=s.callArgs}catch(e){console.warn(`Layer ${t.name} was passed non-serializable keyword arguments: ${s.callArgs}. They will not be included in the serialized model (and thus will be missing at deserialization time).`),a={}}if(s.inboundLayers.length>0){const t=[];for(let n=0;n<s.inboundLayers.length;n++){const i=s.inboundLayers[n],r=s.nodeIndices[n],o=s.tensorIndices[n];let l=e[_l.nodeKey(i,r)];null==l&&(l=0),t.push([i.name,l,o,a])}r.push(t)}}}const a={};a.name=t.name,a.className=s,a.config=i,a.inboundNodes=r,n.push(a)}t.layers=n;const s=[];for(let t=0;t<this.inputLayers.length;t++){const n=this.inputLayers[t],i=this.inputLayersNodeIndices[t],r=_l.nodeKey(n,i);if(!this.containerNodes.has(r))continue;let a=e[r];null==a&&(a=0);const o=this.inputLayersTensorIndices[t];s.push([n.name,a,o])}t.inputLayers=s;const i=[];for(let t=0;t<this.outputLayers.length;t++){const n=this.outputLayers[t],s=this.outputLayersNodeIndices[t],r=_l.nodeKey(n,s);if(!this.containerNodes.has(r))continue;let a=e[r];null==a&&(a=0);const o=this.outputLayersTensorIndices[t];i.push([n.name,a,o])}return t.outputLayers=i,t}static fromConfig(t,e,n={},s=!1){const i={},r={};function a(t,e){t.name in r?r[t.name].push(e):r[t.name]=[e]}function l(t,e){const n=[];let s;for(const r of e){const o=r[0],l=r[1],u=r[2];if(s=null==r[3]?{}:r[3],!(o in i))return void a(t,e);const h=i[o];if(h.inboundNodes.length<=l)return void a(t,e);const c=h.inboundNodes[l];n.push(c.outputTensors[u])}n.length>0&&t.apply(f(n),s)}function u(t){const n=t.name,r=el(t,null!=e.customObjects?e.customObjects:{});r.setFastWeightInitDuringBuild(s),i[n]=r;t.inboundNodes.forEach((t=>{if(!(t instanceof Array))throw new o(`Corrupted configuration, expected array for nodeData: ${t}`);a(r,t)}))}const h=e.name,c=e.layers;for(const t of c)u(t);for(;!N(r);)for(const t of c){const e=i[t.name];if(e.name in r){const t=r[e.name];delete r[e.name];for(const n of t)l(e,n)}}const d=[],g=[],m=e.inputLayers;for(const t of m){const e=t[0],n=t[1],s=t[2];p(e in i);const r=i[e].inboundNodes[n].outputTensors;d.push(r[s])}const y=e.outputLayers;for(const t of y){const e=t[0],n=t[1],s=t[2];p(e in i);const r=i[e].inboundNodes[n].outputTensors;g.push(r[s])}return new t({inputs:d,outputs:g,name:h})}get stateful(){if(this._stateful)throw new o("Container instance unexpectedly has _stateful = true. The statefulness of a Container is determined by the Layers it contains. Its _stateful property must remain the default false.");for(const t of this.layers)if(t.stateful)return!0;return!1}resetStates(){e.tidy((()=>{this.layers.forEach((t=>{t.stateful&&t.resetStates()}))}))}}function Rl(t,e){return function(t,e,n){const s=e.length;if(null==t||Array.isArray(t)&&0===t.length)return e.map((t=>null));if(1===s)return Array.isArray(t)&&1===t.length?t:"object"==typeof t&&e[0]in t?[t[e[0]]]:[t];if(Array.isArray(t)){if(t.length!==s)throw new Error(`Provided ${n} is an array of ${t.length} element(s), but the model has ${s} outputs. Make sure a set of weights is provided for each model output.`);return t}if("object"==typeof t&&Object.keys(t).length>0&&"object"==typeof t[Object.keys(t)[0]]){const n=[];return e.forEach((e=>{e in t?n.push(t[e]):n.push(null)})),n}throw new Error(`The model has multiple (${s}) outputs, so ${n} must be either an array with ${s} elements or an object with ${e} keys. Provided ${n} not understood: ${JSON.stringify(t)}`)}(t,e,"classWeight")}async function Ol(t,n,s,i){if(null!=n||null!=i)throw new Error("Support sampleWeight is not implemented yet");if(null!=s){const n=e.tidy((()=>{if(1===t.shape.length)return e.clone(t);if(2===t.shape.length){if(t.shape[1]>1){const n=1;return e.argMax(t,n)}if(1===t.shape[1])return e.reshape(t,[t.shape[0]]);throw new Error(`Encountered unexpected last-dimension size (${t.shape[1]}) during handling of class weights. The size is expected to be >= 1.`)}throw new Error(`Unexpected rank of target (y) tensor (${t.rank}) during handling of class weights. The rank is expected to be 1 or 2.`)})),i=Array.from(await n.data());e.dispose(n);const r=[];return i.forEach((t=>{if(null==s[t])throw new Error(`classWeight must contain all classes in the training data. The class ${t} exists in the data but not in classWeight`);r.push(s[t])})),e.tensor1d(r,"float32")}return null}function Ml(t,n){return e.mul(t,n)}function Bl(t,e){let n,s;const r=e;n=r.xs,s=r.ys,i.util.assert(null!=n&&null!=s,(()=>`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 ${e}`));const a=Pl("input",t.inputNames,n),o=Pl("output",t.outputNames,s),l=a[0].shape[0];i.util.assert(a.length===t.inputs.length,(()=>`LayersModel has ${t.inputs.length} inputs, but the dataset provides ${a.length} inputs.  (Expected input keys: ${JSON.stringify(t.inputNames)})`)),i.util.assert(o.length===t.outputs.length,(()=>`LayersModel has ${t.outputs.length} outputs, but the dataset provides ${o.length} outputs.  (Expected output keys: ${JSON.stringify(t.outputNames)})`));for(let e=0;e<a.length;e++)i.util.assert(a[e].shape[0]===l,(()=>`Batch size mismatch: input ${t.inputNames[e]} has ${a[e].shape[0]}; expected  ${l} based on input ${t.inputNames[0]}.`));for(let e=0;e<o.length;e++)i.util.assert(o[e].shape[0]===l,(()=>`Batch size mismatch: output ${t.outputNames[e]} has ${o[e].shape[0]}; expected  ${l} based on input ${t.inputNames[0]}.`));return{xs:a,ys:o}}function Pl(t,e,n){if(n instanceof i.Tensor)return[n];if(Array.isArray(n))return i.util.assert(n.length===e.length,(()=>`Received an array of ${n.length} Tensors, but expected ${e.length} to match the ${t} keys ${e}.`)),n;{const s=[];for(const i of e){if(null==n[i])throw new o(`The feature data generated by the dataset lacks the required ${t} key '${i}'.`);s.push(n[i])}return s}}async function Ul(t,e,n){const s=null!=n.batchesPerEpoch;if(i.util.assert(null!=t.optimizer,(()=>"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig).")),i.util.assert(null!=n,(()=>"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call.")),i.util.assert(null!=n.epochs&&n.epochs>0&&Number.isInteger(n.epochs),(()=>`For fitDataset(), config.epochs is expected to be a positive integer, but got ${n.epochs}`)),i.util.assert(!s||n.batchesPerEpoch>0&&Number.isInteger(n.batchesPerEpoch),(()=>`For fitDataset(), config.batchesPerEpoch is expected to be a positive integer if specified, but got ${n.batchesPerEpoch}`)),i.util.assert(null==n.validationSplit,(()=>"`validationSplit` is not supported by `fitDataset()`. Use validationData instead.")),t.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");t.isTraining=!0;try{const r=null!=n.validationData;let a,o;if(r)if(Wl(n.validationData))i.util.assert(null==n.validationBatches||n.validationBatches>0&&Number.isInteger(n.validationBatches),(()=>`For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, but got ${n.validationBatches}`));else{const t=function(t){if(3===t.length)throw new l("Validation with sample weights is not implemented yet.");return{xs:t[0],ys:t[1]}}(n.validationData);a=t.xs,o=t.ys}const u=t.makeTrainFunction(),h=t.getDedupedMetricsNames();let c;c=r?h.slice().concat(h.map((t=>"val_"+t))):h.slice();const p=Xo(n.callbacks,n.yieldEvery),d=null==n.verbose?1:n.verbose,{callbackList:f,history:m}=tl(p,d,n.epochs,null,null,function(t,e){let n=null;null!=e.batchesPerEpoch?n=e.batchesPerEpoch:Number.isFinite(t.size)&&(n=t.size);return n}(e,n),null,r,c);f.setModel(t),t.history=m,await f.onTrainBegin(),t.stopTraining_=!1;let y=null==n.initialEpoch?0:n.initialEpoch,b=await e.iterator();for(;y<n.epochs;){const l={};await f.onEpochBegin(y);let c=0,p=0;for(s||(b=await e.iterator());!s||c<n.batchesPerEpoch;){const e=await b.next();if(s&&e.done){console.warn(`You provided \`batchesPerEpoch\` as ${n.batchesPerEpoch}, but your dataset iterator ran out of data after ${c} batches; interrupting training. Make sure that your dataset can generate at least \`batchesPerEpoch * epochs\` batches (in this case, `+n.batchesPerEpoch*n.epochs+" batches). You may need to use the repeat() function when building your dataset.");break}if(null!=e.value){const{xs:s,ys:r}=Bl(t,e.value),a={};a.batch=p,a.size=s[0].shape[0],await f.onBatchBegin(p,a);const o=[];if(null!=n.classWeight){const e=Rl(n.classWeight,t.outputNames);for(let t=0;t<e.length;++t)o.push(await Ol(r[t],null,e[t]))}const l=s.concat(r).concat(o),d=u(l);i.dispose(l);for(let t=0;t<h.length;++t){const e=h[t],n=d[t];a[e]=n,i.keep(n)}await f.onBatchEnd(p,a),Ko(a),p++,c++}if(s?c>=n.batchesPerEpoch:e.done){if(r){let e;e=Wl(n.validationData)?g(await t.evaluateDataset(n.validationData,{batches:n.validationBatches})):g(t.evaluate(a,o,{batchSize:null==n.validationBatchSize?32:n.validationBatchSize,verbose:0}));for(let n=0;n<t.metricsNames.length;++n)l[`val_${t.metricsNames[n]}`]=e[n]}break}if(t.stopTraining_)break}if(await f.onEpochEnd(y,l),y++,t.stopTraining_)break}return await f.onTrainEnd(),await t.history.syncData(),t.history}finally{t.isTraining=!1}}function Wl(t){return"function"==typeof t.iterator}function jl(t){i.util.assert(t>0&&Number.isInteger(t),(()=>`batchSize is required to be a positive integer, but got ${t}`))}function ql(t,e,n){return null==t?[null]:Array.isArray(t)?t.map((t=>st(t,e,n-e))):st(t,e,n-e)}function Vl(t,e){return i.tidy((()=>null==t?null:Array.isArray(t)?t.map((t=>Vl(t,e))):ct(t,"int32"===e.dtype?e:i.cast(e,"int32"))))}function Kl(t,e){const n=[];let s=0,i=null;for(;s<t;)i=s+e,i>=t&&(i=t),n.push([s,i]),s=i;return n}function Gl(t){const n=[];t instanceof e.Tensor&&(t=[t]);for(let e=0;e<t.length;++e){const s=t[e];if(1===s.rank)n.push(nt(s,1));else{if(0===s.rank)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");n.push(s)}}return n}function Hl(t,n){if(null==t)return;const s=[];if(n instanceof e.Tensor)s.push(n.id);else if(Array.isArray(n))n.forEach((t=>s.push(t.id)));else if(null!=n)for(const t in n){const e=n[t];s.push(e.id)}const i=[];if(t instanceof e.Tensor)-1===s.indexOf(t.id)&&i.push(t);else if(Array.isArray(t))t.forEach((t=>{-1===s.indexOf(t.id)&&i.push(t)}));else if(null!=t)for(const e in t){const n=t[e];-1===s.indexOf(n.id)&&i.push(n)}i.forEach((t=>{t.isDisposed||t.dispose()}))}function Jl(t){return Array.isArray(t)}function Zl(t){return!function(t){return t instanceof e.Tensor}(t)&&!Jl(t)}function Yl(t,e,n,s=!0,i=""){if(null==e||0===e.length){if(null!=t){let e=!1;if(Jl(t)&&t.length>0)e=!0;else if(Zl(t)){for(const n in t)if(t.hasOwnProperty(n)){e=!0;break}}else e=!0;if(e)throw new o(`Error when checking model ${i} expected no data, but got ${t}`)}return[]}if(null==t)return e.map((t=>null));let r;if(Zl(t)){r=[];for(const n of e){if(null==t[n])throw new o(`No data provided for "${n}". Need data for each key in: ${e}`);r.push(t[n])}}else if(Jl(t)){if(t.length!==e.length)throw new o(`Error when checking model ${i}: the Array of Tensors that you are passing to your model is not the size the model expected. Expected to see ${e.length} Tensor(s), but instead got the following list of Tensor(s): ${t}`);r=t}else{if(e.length>1)throw new o(`The model ${i} expects ${e.length} Tensor(s), but only received one Tensor. Found: Tensor with shape ${t.shape}`);r=[t]}if(r=Gl(r),null!=n)for(let t=0;t<e.length;++t){if(null==n[t])continue;const a=r[t];if(a.shape.length!==n[t].length)throw new o(`Error when checking ${i}: expected ${e[t]} to have ${n[t].length} dimension(s). but got array with shape ${a.shape}`);for(let e=0;e<n[t].length;++e){if(0===e&&!s)continue;const r=a.shape[e],l=n[t][e];if(null!=l&&l>=0&&r!==l)throw new o(`${i} expected a batch of elements where each example has shape [${n[t].slice(1,n[t].length)}] (i.e.,tensor shape [*,${n[t].slice(1,n[t].length)}]) but the ${i} received an input with ${a.shape[0]} examples, each with shape [${a.shape.slice(1,a.shape.length)}] (tensor shape [${a.shape}])`)}}return r}function Xl(t,e,n,s=!0,i=""){let r;if(Array.isArray(t)){if(t.length!==e.length)throw new o(`Error when checking model ${i}: the Array of Tensors that you are passing to your model is not the size the the model expected. Expected to see ${e.length} Tensor(s), but instead got ${t.length} Tensors(s).`);r=t}else{if(e.length>1)throw new o(`The model expects ${e.length} ${i} Tensors, but only received one Tensor. Found: array with shape ${JSON.stringify(t.shape)}.`);r=[t]}if(null!=n)for(let t=0;t<e.length;++t){if(null==n[t])continue;const a=r[t];if(a.shape.length!==n[t].length)throw new o(`Error when checking ${i}: expected ${e[t]} to have ${n[t].length} dimension(s), but got array with shape ${JSON.stringify(a.shape)}`);for(let r=0;r<n[t].length;++r){if(0===r&&!s)continue;const l=a.shape[r],u=n[t][r];if(null!=u&&u!==l)throw new o(`Error when checking ${i}: expected ${e[t]} to have shape ${JSON.stringify(n[t])} but got array with shape ${JSON.stringify(a.shape)}.`)}}}class Ql extends _l{constructor(t){super(t),this.isTraining=!1}summary(t,e,n=console.log){if(!this.built)throw new o("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");zl(this,t,e,n)}compile(t){if(null==t.loss&&(t.loss=[]),this.loss=t.loss,"string"==typeof t.optimizer)this.optimizer_=function(t){const n={Adagrad:()=>e.train.adagrad(.01),Adadelta:()=>e.train.adadelta(1,.95,tt()),Adam:()=>e.train.adam(.001,.9,.999,tt()),Adamax:()=>e.train.adamax(.002,.9,.999,tt(),0),RMSProp:()=>e.train.rmsprop(.001,.9,0,tt()),SGD:()=>e.train.sgd(.01)};if(n.adagrad=n.Adagrad,n.adadelta=n.Adadelta,n.adam=n.Adam,n.adamax=n.Adamax,n.rmsprop=n.RMSProp,n.sgd=n.SGD,t in n)return n[t]();throw new o(`Unknown Optimizer ${t}`)}(t.optimizer),this.isOptimizerOwned=!0;else{if(!(t.optimizer instanceof e.Optimizer))throw new o("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=t.optimizer,this.isOptimizerOwned=!1}let n=[];if(Array.isArray(t.loss)||"string"==typeof t.loss||"function"==typeof t.loss)if(Array.isArray(t.loss)){if(t.loss.length!==this.outputs.length)throw new o(`When passing an Array as loss, it should have one entry per model output. The model has ${this.outputs.length} output(s), but you passed loss=${t.loss}.`);const e=t.loss;n=e.map((t=>cl(t)))}else{const e=cl(t.loss);this.outputs.forEach((t=>{n.push(e)}))}else{t.loss=t.loss;for(const e in t.loss)if(-1===this.outputNames.indexOf(e))throw new o(`Unknown entry in loss dictionary: "${e}". Only expected the following keys: ${this.outputNames}`);for(const e of this.outputNames)null==t.loss[e]&&console.warn(`Output "${e}" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data to be passed to ${e} during training`),n.push(cl(t.loss[e]))}this.lossFunctions=n,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(let t=0;t<this.outputs.length;++t){const e=this.internalOutputShapes[t],n=this.outputNames[t];this.feedOutputNames.push(n),this.feedOutputShapes.push(e),this.feedLossFns.push(this.lossFunctions[t])}const s=[];this.metrics=t.metrics,this.metricsNames=["loss"],this.metricsTensors=[],q("loss",(()=>{for(let t=0;t<this.outputs.length;++t){if(-1!==s.indexOf(t))continue;const e=this.lossFunctions[t];this.outputs.length>1&&(this.metricsTensors.push([e,t]),this.metricsNames.push(this.outputNames[t]+"_loss"))}}));const i=function(t,e){if(null==t||Array.isArray(t)&&0===t.length)return e.map((t=>[]));let n;if("string"==typeof t||"function"==typeof t)n=[t];else{if(!Array.isArray(t)&&"object"!=typeof t)throw new TypeError(`Type of metrics argument not understood. Expected an string,function, Array, or Object, found: ${t}`);n=t}if(Array.isArray(n))return e.map((t=>n));{const t=[];for(const s of e){let e=n.hasOwnProperty(s)?n[s]:[];Array.isArray(e)||(e=[e]),t.push(e)}return t}}(t.metrics,this.outputNames),r=(t,e,n)=>{this.outputNames.length>1&&(e=this.outputNames[t]+"_"+e),this.metricsNames.push(e),this.metricsTensors.push([n,t])};q("metric",(()=>{for(let t=0;t<this.outputs.length;++t){if(-1!==s.indexOf(t))continue;(e=>{let n,s,i;for(const a of e){if("string"==typeof a&&-1!==["accuracy","acc","crossentropy","ce"].indexOf(a)){const e=this.internalOutputShapes[t];let r;1===e[e.length-1]||this.lossFunctions[t]===ll?-1!==["accuracy","acc"].indexOf(a)?s=pl:-1!==["crossentropy","ce"].indexOf(a)&&(s=yl):this.lossFunctions[t]===ol?-1!==["accuracy","acc"].indexOf(a)?s=bl:-1!==["crossentropy","ce"].indexOf(a)&&(s=kl):-1!==["accuracy","acc"].indexOf(a)?s=dl:-1!==["crossentropy","ce"].indexOf(a)&&(s=wl),-1!==["accuracy","acc"].indexOf(a)?r="acc":-1!==["crossentropy","ce"].indexOf(a)&&(r="ce"),i=s,n=""+r}else{const t=Sl(a);i=t,n=""+xl(a)}let e;q(n,(()=>{e=i})),r(t,n,e)}})(i[t])}})),this.collectedTrainableWeights=this.trainableWeights}checkTrainableWeightsConsistency(){null!=this.collectedTrainableWeights&&this.trainableWeights.length!==this.collectedTrainableWeights.length&&console.warn("Discrepancy between trainableweights and collected trainable weights. Did you set `model.trainable` without calling `model.compile()` afterwards?")}evaluate(t,e,n={}){const s=null==n.batchSize?32:n.batchSize;jl(s);const i=this.standardizeUserDataXY(t,e,!0,s);try{const r=i[0].concat(i[1]);this.makeTestFunction();const a=this.testFunction;return f(this.testLoop(a,r,s,n.verbose,n.steps))}finally{Hl(i[0],t),Hl(i[1],e)}}async evaluateDataset(t,n){return this.makeTestFunction(),async function(t,n,s){const r=null!=(s=s||{}).batches,a=t.testFunction;let o=[];if(s.verbose>0)throw new l("Verbose mode is not implemented yet.");i.util.assert(!r||s.batches>0&&Number.isInteger(s.batches),(()=>`Test loop expects \`batches\` to be a positive integer, but received ${JSON.stringify(s.batches)}`));const u="function"==typeof n.next?n:await n.iterator();let h=0,c=0;for(;!r||c<s.batches;){const n=await u.next();if(o=i.tidy((()=>{if(n.value){const{xs:s,ys:r}=Bl(t,n.value),l=s.concat(r),u=i.tidy((()=>a(l)));if(i.dispose(l),0===c)for(let t=0;t<u.length;++t)o.push(e.scalar(0));const p=l[0].shape[0];for(let t=0;t<u.length;++t){const e=u[t],n=o[t];o[t]=i.tidy((()=>i.add(o[t],i.mul(p,e)))),c>0&&i.dispose(n)}i.dispose(u),h+=p,++c}return o})),n.done){r&&console.warn(`Your dataset iterator ran out of data during evaluateDataset(). Interrupting evalution. Make sure that your dataset can generate at least \`batches\` batches (in this case, ${s.batches} batches). You may need to use the repeat() function when building your dataset.`);break}}for(let t=0;t<o.length;++t){const e=o[t];o[t]=i.div(o[t],h),i.dispose(e)}return f(o)}(this,t,n)}checkNumSamples(t,e,n,s="steps"){let i;if(null!=n){if(i=null,null!=e)throw new o(`If ${s} is set, batchSize must be null or undefined.Got batchSize = ${e}`)}else{if(null==t)throw new o(`Either the input data should have a defined shape, or ${s} shoud be specified.`);i=Array.isArray(t)?t[0].shape[0]:t.shape[0]}return i}execute(t,n){if(Array.isArray(n)&&0===n.length)throw new o("`outputs` is an empty Array, which is not allowed.");const s=Array.isArray(n),i=s?n:[n],r=this.retrieveSymbolicTensors(i),a=new se;if(t instanceof e.Tensor&&(t=[t]),Array.isArray(t)){if(t.length!==this.inputs.length)throw new o(`The number of inputs provided (${t.length}) does not match the number of inputs of this model (${this.inputs.length}).`);for(let e=0;e<this.inputs.length;++e)a.add(this.inputs[e],t[e])}else for(const e of this.inputs){const n=t[e.name];if(null==n)throw new o(`No value is provided for the model's input ${e.name}`);a.add(e,n)}const l=ae(r,a);return s?l:l[0]}retrieveSymbolicTensors(t){const e=c(null,t.length);let n=t.length;for(const s of this.layers){const i=Array.isArray(s.output)?s.output:[s.output],r=i.map((t=>t.name));for(let s=0;s<t.length;++s){const a=r.indexOf(t[s]);if(-1!==a&&(e[s]=i[a],n--),0===n)break}if(0===n)break}if(n>0){const n=[];throw e.forEach(((e,s)=>{null==e&&n.push(t[s])})),new o(`Cannot find SymbolicTensors for output name(s): ${JSON.stringify(n)}`)}return e}predictLoop(t,e=32,n=!1){return i.tidy((()=>{const s=this.checkNumSamples(t);if(n)throw new l("Verbose predictLoop() is not implemented yet.");const r=Kl(s,e),a=this.outputs.map((t=>[]));for(let e=0;e<r.length;++e){i.tidy((()=>{const n=r[e][0],s=r[e][1],i=ql(t,n,s),a=[];if(Array.isArray(i))for(let t=0;t<i.length;++t)a.push({key:this.inputs[t],value:i[t]});else a.push({key:this.inputs[0],value:i});const o=new se(a);return ae(this.outputs,o)})).forEach(((t,e)=>a[e].push(t)))}return f(a.map((t=>i.concat(t,0))))}))}predict(t,e={}){const n=Gl(t);Xl(n,this.inputNames,this.feedInputShapes,!1);try{const s=null==e.batchSize?32:e.batchSize;return jl(s),this.predictLoop(n,s)}finally{Hl(n,t)}}predictOnBatch(t){Xl(t,this.inputNames,this.feedInputShapes,!0);const e=(Array.isArray(t)?t[0]:t).shape[0];return this.predictLoop(t,e)}standardizeUserDataXY(t,n,s=!0,i){if(null==this.optimizer_)throw new a("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");const r=[];for(let t=0;t<this.feedOutputShapes.length;++t){const e=this.feedOutputShapes[t];this.feedLossFns[t]===ol?r.push(e.slice(0,e.length-1).concat([1])):r.push(e)}if(function(t,n,s){const i=x(t.map((t=>t.shape[0])));i.sort();const r=x(n.map((t=>t.shape[0])));if(r.sort(),i.length>1)throw new o(`All input Tensors (x) should have the same number of samples. Got array shapes: ${JSON.stringify(t.map((t=>t.shape)))}`);if(r.length>1)throw new o(`All target Tensors (y) should have the same number of samples. Got array shapes: ${JSON.stringify(n.map((t=>t.shape)))}`);if(i.length>0&&r.length>0&&!e.util.arraysEqual(i,r))throw new o(`Input Tensors should have the same number of samples as target Tensors. Found ${i[0]} input sample(s) and ${r[0]} target sample(s).`)}(t=Yl(t,this.feedInputNames,this.feedInputShapes,!1,"input"),n=Yl(n,this.feedOutputNames,r,!1,"target")),function(t,e,n){const s=[sl,ll,al];for(let i=0;i<t.length;++i){const r=t[i],a=e[i],l=n[i];if(null!=a){if(a===al&&1===r.shape[r.shape.length-1])throw new o(`You are passing a target array of shape ${r.shape} while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].`);if(-1!==s.indexOf(a)){const t=r.shape.slice(1),e=l.slice(1);for(let n=0;n<t.length;++n){const s=t[n],i=e[n];if(null!=i&&s!==i)throw new o(`A target Tensor with shape ${r.shape} was passed for an output of shape ${l}, while using a loss function that expects targets to have the same shape as the output.`)}}}}}(n,this.feedLossFns,this.feedOutputShapes),this.stateful&&null!=i&&i>0&&t[0].shape[0]%i!=0)throw new o(`In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size ${i}. Found: ${t[0].shape[0]} sample(s).`);return[t,n]}async standardizeUserData(t,e,n,s,i=!0,r){const[a,o]=this.standardizeUserDataXY(t,e,i,r);if(null!=n)throw new Error("sample weight is not supported yet.");let l=null;if(null!=s){const t=Rl(s,this.outputNames);l=[];for(let e=0;e<t.length;++e)l.push(await Ol(o[e],null,t[e]))}return[a,o,l]}testLoop(t,n,s,r=0,a){return i.tidy((()=>{const o=this.checkNumSamples(n,s,a,"steps"),u=[];if(r>0)throw new l("Verbose mode is not implemented yet.");if(null!=a)throw new l("steps mode in testLoop() is not implemented yet");{const r=Kl(o,s),a=e.tensor1d(X(0,o));for(let s=0;s<r.length;++s){const o=r[s][0],l=r[s][1],h=st(a,o,l-o),c=Vl(n,h),p=t(c);if(0===s)for(let t=0;t<p.length;++t)u.push(e.scalar(0));for(let t=0;t<p.length;++t){const e=p[t];u[t]=i.add(u[t],i.mul(l-o,e))}}for(let t=0;t<u.length;++t)u[t]=i.div(u[t],o)}return u}))}getDedupedMetricsNames(){const t=this.metricsNames,e=[];for(let n=0;n<t.length;++n){const s=t[n];let i=s;if(d(t,s)>1){i+=`_${d(t.slice(0,n),s)}`}e.push(i)}return e}makeTrainFunction(){return t=>{const e=[],n=t.slice(0,this.inputs.length),s=t.slice(this.inputs.length,this.inputs.length+this.outputs.length),r=t.slice(this.inputs.length+this.outputs.length,this.inputs.length+2*this.outputs.length),a=[],o=this.collectedTrainableWeights.map((t=>t.read()));return[this.optimizer_.minimize((()=>{const t=[];for(let e=0;e<this.inputs.length;++e)t.push({key:this.inputs[e],value:n[e]});const o=new se(t),l=ae(this.outputs,o,{training:!0});let u;for(let t=0;t<this.lossFunctions.length;++t){let n=(0,this.lossFunctions[t])(s[t],l[t]);null!=r[t]&&(n=Ml(n,r[t]));const a=i.mean(n);e.push(a),u=0===t?n:i.add(u,n)}for(let t=0;t<this.metricsTensors.length;++t){let n;if(this.outputs.length>1&&t<this.outputs.length)n=e[t];else{const e=this.metricsTensors[t][0],r=this.metricsTensors[t][1];n=i.mean(e(s[r],l[r]))}i.keep(n),a.push(n)}return u=i.mean(u),this.calculateLosses().forEach((t=>{u=i.add(u,t)})),u}),!0,o)].concat(a)}}makeTestFunction(){this.testFunction=t=>i.tidy((()=>{const e=[];let n;const s=t.slice(0,this.inputs.length),r=t.slice(this.inputs.length,this.inputs.length+this.outputs.length),a=[];for(let t=0;t<this.inputs.length;++t)a.push({key:this.inputs[t],value:s[t]});const o=new se(a),l=ae(this.outputs,o);for(let t=0;t<this.lossFunctions.length;++t){const s=this.lossFunctions[t],a=i.mean(s(r[t],l[t]));n=0===t?a:i.add(n,a),e.push(n)}for(let t=0;t<this.metricsTensors.length;++t){const n=this.metricsTensors[t][0],s=this.metricsTensors[t][1],a=i.mean(n(r[s],l[s]));e.push(a)}return e}))}async fit(t,e,n={}){if(this.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");let s,r,a,u,h,c,p,d,f;this.isTraining=!0;try{const g=null==n.batchSize?32:n.batchSize;jl(g);const m=!1,y=await this.standardizeUserData(t,e,n.sampleWeight,n.classWeight,m,g);s=y[0],r=y[1],f=y[2];let b,w=!1;if(null!=n.validationData&&n.validationData.length>0){if(w=!0,2!==n.validationData.length)throw 3===n.validationData.length?new l("validationData including sample weights is not supported yet."):new o(`When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; ${n.validationData} is invalid.`);h=n.validationData[0],c=n.validationData[1];const t=!0,e=await this.standardizeUserData(h,c,null,null,t,g);p=e[0],d=e[1],b=p.concat(d)}else if(null!=n.validationSplit&&n.validationSplit>0&&n.validationSplit<1){w=!0;const t=Math.floor(s[0].shape[0]*(1-n.validationSplit)),e=s[0].shape[0];p=ql(s,t,e),a=s,s=ql(s,0,t),d=ql(r,t,e),u=r,r=ql(r,0,t),b=p.concat(d)}else null!=n.validationSteps&&(w=!0);const k=s.concat(r).concat(f);this.checkTrainableWeightsConsistency();const v=this.makeTrainFunction(),S=this.getDedupedMetricsNames();let x,N;w?(this.makeTestFunction(),x=this.testFunction,N=S.slice().concat(S.map((t=>"val_"+t)))):(x=null,b=[],N=S.slice());const I=Xo(n.callbacks,n.yieldEvery);return await this.fitLoop(v,k,S,g,n.epochs,n.verbose,I,x,b,n.shuffle,N,n.initialEpoch,null,null)}finally{this.isTraining=!1,Hl(s,t),Hl(r,e),Hl(a,t),Hl(u,e),Hl(p,h),Hl(d,c),null!=f&&i.dispose(f)}}async fitLoop(t,n,s,r,a,u,h,c,p,d,f,g,m,y){null==r&&(r=32),null==a&&(a=1),null==d&&(d=!0),null==g&&(g=0);let b=!1;if(null!=c&&null!=p&&(b=!0),null!=y&&(b=!0,null==m))throw new o("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");const w=this.checkNumSamples(n,r,m,"steps_per_epoch");let k;null!=w&&(k=X(0,w)),null==u&&(u=1);const{callbackList:v,history:S}=tl(h,u,a,g,w,m,r,b,f);v.setModel(this),this.history=S,await v.onTrainBegin(),this.stopTraining_=!1;for(let o=g;o<a;++o){await v.onEpochBegin(o);const a={};if(null!=m)throw new l("stepsPerEpoch mode is not implemented yet.");{if("batch"===d)throw new l("batch shuffling is not implemneted yet");d&&e.util.shuffle(k);const o=e.tensor1d(k),u=Kl(w,r);for(let e=0;e<u.length;++e){const l={};if(await v.onBatchBegin(e,l),i.tidy((()=>{const h=u[e][0],d=u[e][1],f=st(o,h,d-h);l.batch=e,l.size=d-h;const g=Vl(n,f),m=t(g);for(let t=0;t<s.length;++t){const e=s[t],n=m[t];l[e]=n,i.keep(n)}if(e===u.length-1&&b){const t=this.testLoop(c,p,r);for(let e=0;e<s.length;++e){const n=s[e],r=t[e];i.keep(r),a["val_"+n]=r}}})),await v.onBatchEnd(e,l),Ko(l),this.stopTraining_)break}o.dispose()}if(await v.onEpochEnd(o,a),this.stopTraining_)break}return await v.onTrainEnd(),await this.history.syncData(),this.history}async fitDataset(t,e){return Ul(this,t,e)}async trainOnBatch(t,e){const n=await this.standardizeUserData(t,e),s=n[0],r=n[1],a=this.makeTrainFunction()(s.concat(r)),o=[];for(const t of a){const e=await t.data();o.push(e[0])}return i.dispose(a),Hl(n[0],t),Hl(n[1],e),f(o)}getNamedWeights(t){const e=[],n=null!=t&&t.trainableOnly,s=n?this.trainableWeights:this.weights,i=this.getWeights(n);for(let t=0;t<s.length;++t)n&&!s[t].trainable||e.push({name:s[t].originalName,tensor:i[t]});return e}set stopTraining(t){this.stopTraining_=t}get stopTraining(){return this.stopTraining_}get optimizer(){return this.optimizer_}set optimizer(t){this.optimizer_!==t&&(this.optimizer_=t,this.isOptimizerOwned=!1)}dispose(){const t=super.dispose();if(0===t.refCountAfterDispose&&null!=this.optimizer&&this.isOptimizerOwned){const e=i.memory().numTensors;this.optimizer_.dispose(),t.numDisposedVariables+=e-i.memory().numTensors}return t}getLossIdentifiers(){let t;if("string"==typeof this.loss)t=m(this.loss);else if(Array.isArray(this.loss)){for(const t of this.loss)if("string"!=typeof t)throw new Error("Serialization of non-string loss is not supported.");t=this.loss.map((t=>m(t)))}else{const e=Object.keys(this.loss);t={};const n=this.loss;for(const s of e){if("string"!=typeof n[s])throw new Error("Serialization of non-string loss is not supported.");t[s]=m(n[s])}}return t}getMetricIdentifiers(){if("string"==typeof this.metrics||"function"==typeof this.metrics)return[m(xl(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map((t=>m(xl(t))));{const t={};for(const e in this.metrics)t[e]=m(xl(this.metrics[e]));return t}}getTrainingConfig(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}}loadTrainingConfig(t){if(null!=t.weighted_metrics)throw new Error("Loading weight_metrics is not supported yet.");if(null!=t.loss_weights)throw new Error("Loading loss_weights is not supported yet.");if(null!=t.sample_weight_mode)throw new Error("Loading sample_weight_mode is not supported yet.");const e=el(Fl(t.optimizer_config));let n,s;if("string"==typeof t.loss)n=y(t.loss);else if(Array.isArray(t.loss))n=t.loss.map((t=>y(t)));else if(null!=t.loss){n={};for(const e in t.loss)n[e]=y(t.loss[e])}if(Array.isArray(t.metrics))s=t.metrics.map((t=>y(t)));else if(null!=t.metrics){s={};for(const e in t.metrics)s[e]=y(t.metrics[e])}this.compile({loss:n,metrics:s,optimizer:e})}async save(t,n){if("string"==typeof t){const n=e.io.getSaveHandlers(t);if(0===n.length)throw new o(`Cannot find any save handlers for URL '${t}'`);if(n.length>1)throw new o(`Found more than one (${n.length}) save handlers for URL '${t}'`);t=n[0]}if(null==t.save)throw new o("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");const s=await e.io.encodeWeights(this.getNamedWeights(n)),i={modelTopology:this.toJSON(null,!1),format:"layers-model",generatedBy:"TensorFlow.js tfjs-layers v4.15.0",convertedBy:null};if(null!=n&&n.includeOptimizer&&null!=this.optimizer){i.trainingConfig=this.getTrainingConfig();const t="optimizer",{data:n,specs:r}=await e.io.encodeWeights(await this.optimizer.getWeights(),t);s.specs.push(...r),s.data=e.io.concatenateArrayBuffers([s.data,n])}if(null!=this.userDefinedMetadata){const t=!0;Il(this.userDefinedMetadata,this.name,t),i.userDefinedMetadata=this.userDefinedMetadata}return i.weightData=s.data,i.weightSpecs=s.specs,t.save(i)}setUserDefinedMetadata(t){Il(t,this.name),this.userDefinedMetadata=t}getUserDefinedMetadata(){return this.userDefinedMetadata}}Ql.className="Model",e.serialization.registerClass(Ql);class tu extends Ql{}tu.className="Functional",e.serialization.registerClass(tu);class eu extends Ql{constructor(t){if(super({inputs:[],outputs:[]}),t=t||{},this.trainable=!0,this.built=!1,this.name=null!=t.name?t.name:D("sequential_"),null!=t.layers)for(const e of t.layers)this.add(e)}checkShape(t){if(t.inboundNodes[0].outputTensors[0].shape.some((t=>t<0)))throw new o(`Negative dimension size caused by adding layer ${t.name} with input shape [${t.inboundNodes[0].inputTensors[0].shape}]`)}add(t){const e=t instanceof eu||t instanceof Ql;let n;if(e){if(n=t,1!==n.outputs.length)throw new o("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");if(1!==n.inputs.length)throw new o("All layers in a Sequential model should have a single input tensor. For multi-input layers, use the functional API.")}if(0===this.outputs.length){if(0===t.inboundNodes.length){if(null==t.batchInputShape)throw new o("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");const e=ne({batchShape:t.batchInputShape,dtype:t.dtype,name:t.name+"_input"});t.apply(e)}if(e)this.outputs=n.outputs,this.inputs=n.inputs;else{if(1!==t.inboundNodes.length)throw new o(`A layer added to a Sequential model must not already be connected somewhere else. LayersModel received layer ${t.name} which has ${t.inboundNodes.length} pre-existing inbound connections.`);if(1!==t.inboundNodes[0].outputTensors.length)throw new o("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(t),this.outputs=[t.inboundNodes[0].outputTensors[0]],this.inputs=te(this.outputs[0])}this.inboundNodes=[],new Yt({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:c(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map((t=>t.shape)),outputShapes:this.outputs[0].shape})}else{const e=t.apply(this.outputs[0]);if(Array.isArray(e))throw new TypeError("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(t),this.outputs=[e],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(t),this.built=!1}pop(){if(0===this.layers.length)throw new TypeError("There are no layers in the model.");if(this.layers.pop(),0===this.layers.length)this.outputs=[],this.inboundNodes=[],this.outboundNodes=[];else{const t=this.layers.length-1;this.layers[t].outboundNodes=[],this.outputs=[this.layers[t].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}}call(t,e){return null==this.model&&this.build(),this.model.call(t,e)}build(t){if(Wt(t),0===this.inputs.length||0===this.outputs.length)throw new TypeError("Sequential model cannot be built: model is empty. Add some layers first.");this.model=new Ql({inputs:this.inputs,outputs:this.outputs[0],name:this.name+"_model"}),this.model.trainable=this.trainable,this.supportsMasking=this.model.supportsMasking,this.inputLayers=this.model.inputLayers,this.inputLayersNodeIndices=this.model.inputLayersNodeIndices,this.inputLayersTensorIndices=this.model.inputLayersTensorIndices,this.outputLayers=this.model.outputLayers,this.outputLayersNodeIndices=this.model.outputLayersNodeIndices,this.outputLayersTensorIndices=this.model.outputLayersTensorIndices,this.nodesByDepth=this.model.nodesByDepth,this.containerNodes=this.model.containerNodes,this.outputNames=this.model.outputNames,this.inputNames=this.model.inputNames,this.built=!0}countParams(){return this.built||this.build(),super.countParams()}summary(t,e,n=console.log){this.built||this.build(),super.summary(t,e,n)}setWeights(t){null==this.model&&this.build(),this.model.setWeights(t)}evaluate(t,e,n={}){if(!this.built)throw new a("The model needs to be compiled before being used.");return this.model.evaluate(t,e,n)}async evaluateDataset(t,e){if(!this.built)throw new a("The model needs to be compiled before being used.");return this.model.evaluateDataset(t,e)}predict(t,e={}){return null==this.model&&this.build(),this.model.predict(t,e)}predictOnBatch(t){return null==this.model&&this.build(),this.model.predictOnBatch(t)}compile(t){this.build(),this.model.compile(t),this.optimizer_=this.model.optimizer,this.isOptimizerOwned=this.model.isOptimizerOwned,this.loss=this.model.loss,this.metrics=this.model.metrics,this.metricsTensors=this.model.metricsTensors,this.metricsNames=this.model.metricsNames}get optimizer(){return null==this.model?void 0:this.model.optimizer}set optimizer(t){this.model.optimizer=t}async fit(t,e,n={}){if(!this.built)throw new a("The model needs to be compiled before being used.");return this.model.fit(t,e,n)}async fitDataset(t,e){if(!this.built)throw new a("The model needs to be compiled before being used.");return this.model.fitDataset(t,e)}async trainOnBatch(t,e){return this.model.trainOnBatch(t,e)}static fromConfig(t,n,s={},i=!1){let r,a={};if(n instanceof Array){if(null==n[0].className||"Merge"===n[0].className)throw new o("Legacy serialization format not supported yet.");r=n}else e.util.assert(null!=n.layers,(()=>"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field.")),r=n.layers,delete n.layers,a=n;const u=new t(a);if(!(u instanceof eu))throw new l(`Sequential.fromConfig called on non-Sequential input: ${u}`);for(const t of r){const e=el(t,void 0,i);i&&e.setFastWeightInitDuringBuild(!0),u.add(e)}return u}set stopTraining(t){if(null==this.model)throw new o("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=t}get stopTraining(){if(null==this.model)throw new o("Cannot get the stopTraining property of a sequential model before it is compiled.");return this.model.stopTraining}getConfig(){const t=[];for(const e of this.layers){const n={};n.className=e.getClassName(),n.config=e.getConfig(),t.push(n)}return{name:this.name,layers:t}}}function nu(t){return ne(t)}eu.className="Sequential",e.serialization.registerClass(eu);let su=class extends e.serialization.Serializable{getConfig(){return{}}};class iu extends su{apply(t,e=1){return function(t,e=1){if(1!==e)throw new l(`Support for alpha values other than 1 (${e}) is not implemented yet.`);return i.elu(t)}(t,e)}}iu.className="elu",e.serialization.registerClass(iu);class ru extends su{apply(t){return i.selu(t)}}ru.className="selu",e.serialization.registerClass(ru);class au extends su{apply(t){return i.relu(t)}}au.className="relu",e.serialization.registerClass(au);class ou extends su{apply(t){return e.tidy((()=>i.minimum(6,i.relu(t))))}}ou.className="relu6",e.serialization.registerClass(ou);class lu extends su{apply(t){return t}}lu.className="linear",e.serialization.registerClass(lu);class uu extends su{apply(t){return i.sigmoid(t)}}uu.className="sigmoid",e.serialization.registerClass(uu);class hu extends su{apply(t){return function(t){return e.tidy((()=>{const e=i.add(.5,i.mul(.2,t));return i.clipByValue(e,0,1)}))}(t)}}hu.className="hardSigmoid",e.serialization.registerClass(hu);class cu extends su{apply(t){return i.softplus(t)}}cu.className="softplus",e.serialization.registerClass(cu);class pu extends su{apply(t){return function(t){return e.tidy((()=>i.div(t,i.add(i.abs(t),1))))}(t)}}pu.className="softsign",e.serialization.registerClass(pu);class du extends su{apply(t){return i.tanh(t)}}du.className="tanh",e.serialization.registerClass(du);let fu=class extends su{apply(t,e=-1){return i.softmax(t,e)}};fu.className="softmax",e.serialization.registerClass(fu);class gu extends su{apply(t,e=-1){return i.logSoftmax(t,e)}}gu.className="logSoftmax",e.serialization.registerClass(gu);class mu extends su{apply(t,n=1){return e.tidy((()=>i.mul(i.sigmoid(i.mul(t,n)),t)))}}mu.className="swish",e.serialization.registerClass(mu);class yu extends su{apply(t){return e.tidy((()=>i.mul(t,i.tanh(i.softplus(t)))))}}function bu(t){return t.getClassName()}function wu(t,n={}){return v(t,e.serialization.SerializationMap.getMap().classNameMap,n,"activation")}function ku(t){if(null==t){const t={className:"linear",config:{}};return wu(t)}if("string"==typeof t){const e={};return e.className=t,e.config={},wu(e)}return t instanceof su?t:wu(t)}function vu(t){if(null!=t&&"object"!=typeof t)throw new Error(`Argument to L1L2 regularizer's constructor is expected to be an object, but received: ${t}`)}yu.className="mish",e.serialization.registerClass(yu);class Su extends e.serialization.Serializable{}class xu extends Su{constructor(t){super(),vu(t),this.l1=null==t||null==t.l1?.01:t.l1,this.l2=null==t||null==t.l2?.01:t.l2,this.hasL1=0!==this.l1,this.hasL2=0!==this.l2}apply(t){return e.tidy((()=>{let n=e.zeros([1]);return this.hasL1&&(n=e.add(n,e.sum(i.mul(this.l1,e.abs(t))))),this.hasL2&&(n=e.add(n,e.sum(i.mul(this.l2,pt(t))))),i.reshape(n,[])}))}getConfig(){return{l1:this.l1,l2:this.l2}}static fromConfig(t,e){return new t({l1:e.l1,l2:e.l2})}}xu.className="L1L2",e.serialization.registerClass(xu);const Nu={l1l2:"L1L2"};function Iu(t){return w(t)}function Au(t,n={}){return v(t,e.serialization.SerializationMap.getMap().classNameMap,n,"regularizer")}function zu(t){if(null==t)return null;if("string"==typeof t){return Au({className:t in Nu?Nu[t]:t,config:{}})}return t instanceof Su?t:Au(t)}class Eu extends Qt{constructor(t){super(null==t?{}:t),this.supportsMasking=!0,null!=t&&(this.maxValue=t.maxValue)}call(t,n){t=Ut(t);let s=e.relu(t);return null!=this.maxValue&&(s=e.clipByValue(s,0,this.maxValue)),s}computeOutputShape(t){return t}getConfig(){const t={maxValue:this.maxValue},e=super.getConfig();return Object.assign(t,e),t}}Eu.className="ReLU",e.serialization.registerClass(Eu);class Tu extends Qt{constructor(t){super(null==t?{}:t),this.DEFAULT_ALPHA=.3,null==t&&(t={}),this.alpha=null==t.alpha?this.DEFAULT_ALPHA:t.alpha}call(t,n){const s=Ut(t);return e.leakyRelu(s,this.alpha)}computeOutputShape(t){return t}getConfig(){const t={alpha:this.alpha},e=super.getConfig();return Object.assign(t,e),t}}Tu.className="LeakyReLU",e.serialization.registerClass(Tu);class Cu extends Qt{constructor(t){if(super(null==t?{}:t),this.DEFAULT_ALPHA_INITIALIZER="zeros",null==t&&(t={}),this.supportsMasking=!0,this.alphaInitializer=Mt(t.alphaInitializer||this.DEFAULT_ALPHA_INITIALIZER),this.alphaRegularizer=zu(t.alphaRegularizer),this.alphaConstraint=Uo(t.alphaConstraint),null==t.sharedAxes)this.sharedAxes=null;else if(Array.isArray(t.sharedAxes))this.sharedAxes=t.sharedAxes;else{if("number"!=typeof t.sharedAxes)throw new o(`Expected sharedAxes to be a number or an array of numbers, but got ${t.sharedAxes}`);this.sharedAxes=[t.sharedAxes]}}build(t){const e=(t=Wt(t)).slice(1);if(null!=this.sharedAxes)for(const t of this.sharedAxes)e[t-1]=1;this.alpha=this.addWeight("alpha",e,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);const n={};if(null!=this.sharedAxes)for(let e=1;e<t.length;++e)n[e]=t[e];this.inputSpec=[new Ht({ndim:t.length,axes:n})],this.built=!0}call(t,n){return t=Ut(t),e.prelu(t,this.alpha.read())}getConfig(){const t={alphaInitializer:Ot(this.alphaInitializer),alphaRegularizer:Iu(this.alphaRegularizer),alphaConstraint:Bo(this.alphaConstraint),sharedAxes:this.sharedAxes},e=super.getConfig();return Object.assign(t,e),t}}Cu.className="PReLU",e.serialization.registerClass(Cu);class $u extends Qt{constructor(t){if(super(null==t?{}:t),this.DEFAULT_ALPHA=1,null==t&&(t={}),null!=t.alpha&&t.alpha!==this.DEFAULT_ALPHA)throw new l(`Non-default alpha value (${t.alpha}) is not supported by the ELU layer yet.`);this.alpha=null==t.alpha?this.DEFAULT_ALPHA:t.alpha}call(t,n){const s=Ut(t);return e.elu(s)}computeOutputShape(t){return t}getConfig(){const t={alpha:this.alpha},e=super.getConfig();return Object.assign(t,e),t}}$u.className="ELU",e.serialization.registerClass($u);class Fu extends Qt{constructor(t){super(null==t?{}:t),this.DEFAULT_THETA=1,null==t&&(t={}),this.theta=null==t.theta?this.DEFAULT_THETA:t.theta}call(t,n){const s=Ut(t);return e.mul(s,e.cast(e.greater(s,this.theta),"float32"))}computeOutputShape(t){return t}getConfig(){const t={theta:this.theta},e=super.getConfig();return Object.assign(t,e),t}}Fu.className="ThresholdedReLU",e.serialization.registerClass(Fu);class Du extends Qt{constructor(t){super(null==t?{}:t),this.DEFAULT_AXIS=1,null==t&&(t={}),this.softmax=(new fu).apply,this.axis=null==t.axis?this.DEFAULT_AXIS:t.axis}call(t,n){return e.tidy((()=>{let s=Ut(t);const i=n.mask;if(null!=i){const t=e.mul(e.sub(e.ones(s.shape),e.cast(i,s.dtype)),e.scalar(-1e9));s=e.add(s,t)}return this.axis instanceof Array?this.axis.length>1?e.exp(e.sub(s,e.logSumExp(s,this.axis,!0))):this.softmax(s,this.axis[0]):this.softmax(s,this.axis)}))}computeOutputShape(t){return t}getConfig(){const t={axis:this.axis},e=super.getConfig();return Object.assign(t,e),t}}function Lu(t,e,n){if("number"==typeof t)return c(t,e);if(t.length!==e)throw new o(`The ${n} argument must be an integer or tuple of ${e} integers. Received: ${t.length} elements.`);for(let i=0;i<e;++i){const r=t[i];if((s=r)!==parseInt(s.toString(),10))throw new o(`The ${n} argument must be an integer or tuple of ${e} integers. Received: ${JSON.stringify(t)} including a non-integer number ${r}`)}return t;var s}function _u(t,e,n,s,i=1){if(null==t)return t;let r;return r="same"===n?t:t-(e+(e-1)*(i-1))+1,Math.floor((r+s-1)/s)}function Ru(t,e,n,s){if(null==t)return null;if("valid"===s)t=t*e+Y([n-e,0]);else{if("same"!==s)throw new o(`Unsupport padding mode: ${s}.`);t*=e}return t}function Ou(t,n){return e.tidy((()=>(P(n),"channelsFirst"===n?i.transpose(t,[0,2,3,1]):t)))}function Mu(t,n){return e.tidy((()=>(P(n),"channelsFirst"===n?i.transpose(t,[0,2,3,4,1]):t)))}function Bu(t,n,s,r=[1,1],a="valid",u,h,c=null){return e.tidy((()=>{if(null==u&&(u="channelsLast"),P(u),3!==t.rank&&4!==t.rank)throw new o(`conv2dWithBiasActivation expects input to be of rank 3 or 4, but received ${t.rank}.`);if(3!==n.rank&&4!==n.rank)throw new o(`conv2dWithBiasActivation expects kernel to be of rank 3 or 4, but received ${t.rank}.`);let e=Ou(t,u);if("causal"===a)throw new l("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return e=i.fused.conv2d({x:e,filter:n,strides:r,pad:"same"===a?"same":"valid",dilations:h,dataFormat:"NHWC",bias:s,activation:c}),"channelsFirst"===u&&(e=i.transpose(e,[0,3,1,2])),e}))}Du.className="Softmax",e.serialization.registerClass(Du);class Pu extends Qt{constructor(t,e){if(super(e),this.bias=null,this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_BIAS_INITIALIZER="zeros",Pu.verifyArgs(e),this.rank=t,z(this.rank,"rank"),1!==this.rank&&2!==this.rank&&3!==this.rank)throw new l(`Convolution layer for rank other than 1, 2, or 3 (${this.rank}) is not implemented yet.`);if(this.kernelSize=Lu(e.kernelSize,t,"kernelSize"),this.strides=Lu(null==e.strides?1:e.strides,t,"strides"),this.padding=null==e.padding?"valid":e.padding,U(this.padding),this.dataFormat=null==e.dataFormat?"channelsLast":e.dataFormat,P(this.dataFormat),this.activation=ku(e.activation),this.useBias=null==e.useBias||e.useBias,this.biasInitializer=Mt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.biasConstraint=Uo(e.biasConstraint),this.biasRegularizer=zu(e.biasRegularizer),this.activityRegularizer=zu(e.activityRegularizer),this.dilationRate=Lu(null==e.dilationRate?1:e.dilationRate,t,"dilationRate"),1===this.rank&&Array.isArray(this.dilationRate)&&1!==this.dilationRate.length)throw new o(`dilationRate must be a number or an array of a single number for 1D convolution, but received ${JSON.stringify(this.dilationRate)}`);if(2===this.rank){if("number"==typeof this.dilationRate)this.dilationRate=[this.dilationRate,this.dilationRate];else if(2!==this.dilationRate.length)throw new o(`dilationRate must be a number or array of two numbers for 2D convolution, but received ${JSON.stringify(this.dilationRate)}`)}else if(3===this.rank)if("number"==typeof this.dilationRate)this.dilationRate=[this.dilationRate,this.dilationRate,this.dilationRate];else if(3!==this.dilationRate.length)throw new o(`dilationRate must be a number or array of three numbers for 3D convolution, but received ${JSON.stringify(this.dilationRate)}`)}static verifyArgs(t){if(p("kernelSize"in t,"required key 'kernelSize' not in config"),"number"!=typeof t.kernelSize&&!A(t.kernelSize,"number",1,3))throw new o(`BaseConv expects config.kernelSize to be number or number[] with length 1, 2, or 3, but received ${JSON.stringify(t.kernelSize)}.`)}getConfig(){const t={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:bu(this.activation),useBias:this.useBias,biasInitializer:Ot(this.biasInitializer),biasRegularizer:Iu(this.biasRegularizer),activityRegularizer:Iu(this.activityRegularizer),biasConstraint:Bo(this.biasConstraint)},e=super.getConfig();return Object.assign(t,e),t}}class Uu extends Pu{constructor(t,e){super(t,e),this.kernel=null,Uu.verifyArgs(e),this.filters=e.filters,z(this.filters,"filters"),this.kernelInitializer=Mt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.kernelConstraint=Uo(e.kernelConstraint),this.kernelRegularizer=zu(e.kernelRegularizer)}build(t){t=Wt(t);const e="channelsFirst"===this.dataFormat?1:t.length-1;if(null==t[e])throw new o(`The channel dimension of the input should be defined. Found ${t[e]}`);const n=t[e],s=this.kernelSize.concat([n,this.filters]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:{[e]:n}}],this.built=!0}call(t,n){return e.tidy((()=>{let n;t=Ut(t);const s=null==this.bias?null:this.bias.read(),r=T(this.activation.getClassName());if(null!=r&&2===this.rank)n=Bu(t,this.kernel.read(),s,this.strides,this.padding,this.dataFormat,this.dilationRate,r);else{if(1===this.rank)n=function(t,n,s,r=1,a="valid",u,h=1){return e.tidy((()=>{if(null==u&&(u="channelsLast"),P(u),3!==t.shape.length)throw new o(`The input of a conv1dWithBias operation should be 3, but is ${t.shape.length} instead.`);if(3!==n.shape.length)throw new o(`The kernel for a conv1dWithBias operation should be 3, but is ${n.shape.length} instead`);if(null!=s&&1!==s.shape.length)throw new o(`The bias for a conv1dWithBias operation should be 1, but is ${n.shape.length} instead`);if("channelsFirst"===u&&(t=i.transpose(t,[0,2,1])),"causal"===a)throw new l("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");let e=i.conv1d(t,n,r,"same"===a?"same":"valid","NWC",h);return null!=s&&(e=ft(e,s)),e}))}(t,this.kernel.read(),s,this.strides[0],this.padding,this.dataFormat,this.dilationRate[0]);else if(2===this.rank)n=Bu(t,this.kernel.read(),s,this.strides,this.padding,this.dataFormat,this.dilationRate);else{if(3!==this.rank)throw new l("convolutions greater than 3D are not implemented yet.");n=function(t,n,s,r=[1,1,1],a="valid",u,h){return e.tidy((()=>{if(null==u&&(u="channelsLast"),P(u),4!==t.rank&&5!==t.rank)throw new o(`conv3dWithBias expects input to be of rank 4 or 5, but received ${t.rank}.`);if(4!==n.rank&&5!==n.rank)throw new o(`conv3dWithBias expects kernel to be of rank 4 or 5, but received ${t.rank}.`);let e=Mu(t,u);if("causal"===a)throw new l("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return e=i.conv3d(e,n,r,"same"===a?"same":"valid","NDHWC",h),null!=s&&(e=ft(e,s)),"channelsFirst"===u&&(e=i.transpose(e,[0,4,1,2,3])),e}))}(t,this.kernel.read(),s,this.strides,this.padding,this.dataFormat,this.dilationRate)}null!=this.activation&&(n=this.activation.apply(n))}return n}))}computeOutputShape(t){t=Wt(t);const e=[],n="channelsLast"===this.dataFormat?t.slice(1,t.length-1):t.slice(2);for(let t=0;t<n.length;++t){const s=_u(n[t],this.kernelSize[t],this.padding,this.strides[t],"number"==typeof this.dilationRate?this.dilationRate:this.dilationRate[t]);e.push(s)}let s=[t[0]];return"channelsLast"===this.dataFormat?(s=s.concat(e),s.push(this.filters)):(s.push(this.filters),s=s.concat(e)),s}getConfig(){const t={filters:this.filters,kernelInitializer:Ot(this.kernelInitializer),kernelRegularizer:Iu(this.kernelRegularizer),kernelConstraint:Bo(this.kernelConstraint)},e=super.getConfig();return Object.assign(t,e),t}static verifyArgs(t){if(!("filters"in t)||"number"!=typeof t.filters||t.filters<1)throw new o(`Convolution layer expected config.filters to be a 'number' > 0 but got ${JSON.stringify(t.filters)}`)}}class Wu extends Uu{constructor(t){super(2,t),Wu.verifyArgs(t)}getConfig(){const t=super.getConfig();return delete t.rank,t}static verifyArgs(t){if("number"!=typeof t.kernelSize&&!A(t.kernelSize,"number",1,2))throw new o(`Conv2D expects config.kernelSize to be number or number[] with length 1 or 2, but received ${JSON.stringify(t.kernelSize)}.`)}}Wu.className="Conv2D",e.serialization.registerClass(Wu);class ju extends Uu{constructor(t){super(3,t),ju.verifyArgs(t)}getConfig(){const t=super.getConfig();return delete t.rank,t}static verifyArgs(t){if("number"!=typeof t.kernelSize&&(!Array.isArray(t.kernelSize)||1!==t.kernelSize.length&&3!==t.kernelSize.length))throw new o(`Conv3D expects config.kernelSize to be number or [number, number, number], but received ${JSON.stringify(t.kernelSize)}.`)}}ju.className="Conv3D",e.serialization.registerClass(ju);class qu extends Wu{constructor(t){if(super(t),this.inputSpec=[new Ht({ndim:4})],"same"!==this.padding&&"valid"!==this.padding)throw new o(`Conv2DTranspose currently supports only padding modes 'same' and 'valid', but received padding mode ${this.padding}`)}build(t){if(4!==(t=Wt(t)).length)throw new o("Input should have rank 4; Received input shape: "+JSON.stringify(t));const e="channelsFirst"===this.dataFormat?1:t.length-1;if(null==t[e])throw new o("The channel dimension of the inputs should be defined. Found `None`.");const n=t[e],s=this.kernelSize.concat([this.filters,n]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Ht({ndim:4,axes:{[e]:n}})],this.built=!0}call(t,e){return i.tidy((()=>{let e=Ut(t);if(4!==e.shape.length)throw new o(`Conv2DTranspose.call() expects input tensor to be rank-4, but received a tensor of rank-${e.shape.length}`);const n=e.shape,s=n[0];let r,a;"channelsFirst"===this.dataFormat?(r=2,a=3):(r=1,a=2);const l=n[r],u=n[a],h=this.kernelSize[0],c=this.kernelSize[1],p=this.strides[0],d=this.strides[1],f=[s,Ru(l,p,h,this.padding),Ru(u,d,c,this.padding),this.filters];"channelsLast"!==this.dataFormat&&(e=i.transpose(e,[0,2,3,1]));let g=i.conv2dTranspose(e,this.kernel.read(),f,this.strides,this.padding);return"channelsLast"!==this.dataFormat&&(g=i.transpose(g,[0,3,1,2])),null!=this.bias&&(g=ft(g,this.bias.read(),this.dataFormat)),null!=this.activation&&(g=this.activation.apply(g)),g}))}computeOutputShape(t){const e=(t=Wt(t)).slice();let n,s,i;"channelsFirst"===this.dataFormat?(n=1,s=2,i=3):(n=3,s=1,i=2);const r=this.kernelSize[0],a=this.kernelSize[1],o=this.strides[0],l=this.strides[1];return e[n]=this.filters,e[s]=Ru(e[s],o,r,this.padding),e[i]=Ru(e[i],l,a,this.padding),e}getConfig(){const t=super.getConfig();return delete t.dilationRate,t}}qu.className="Conv2DTranspose",e.serialization.registerClass(qu);class Vu extends ju{constructor(t){if(super(t),this.inputSpec=[new Ht({ndim:5})],"same"!==this.padding&&"valid"!==this.padding)throw new o(`Conv3DTranspose currently supports only padding modes 'same' and 'valid', but received padding mode ${this.padding}`)}build(t){if(5!==(t=Wt(t)).length)throw new o("Input should have rank 5; Received input shape: "+JSON.stringify(t));const e="channelsFirst"===this.dataFormat?1:t.length-1;if(null==t[e])throw new o("The channel dimension of the inputs should be defined. Found `None`.");const n=t[e],s=this.kernelSize.concat([this.filters,n]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Ht({ndim:5,axes:{[e]:n}})],this.built=!0}call(t,e){return i.tidy((()=>{let e=Ut(t);if(5!==e.shape.length)throw new o(`Conv3DTranspose.call() expects input tensor to be rank-4, but received a tensor of rank-${e.shape.length}`);const n=e.shape,s=n[0];let r,a,l;"channelsFirst"===this.dataFormat?(l=2,r=3,a=4):(l=1,r=2,a=3);const u=n[l],h=n[r],c=n[a],p=this.kernelSize[0],d=this.kernelSize[1],f=this.kernelSize[2],g=this.strides[0],m=this.strides[1],y=this.strides[2],b=[s,Ru(u,g,p,this.padding),Ru(h,m,d,this.padding),Ru(c,y,f,this.padding),this.filters];"channelsLast"!==this.dataFormat&&(e=i.transpose(e,[0,2,3,4,1]));let w=i.conv3dTranspose(e,this.kernel.read(),b,this.strides,this.padding);return"channelsLast"!==this.dataFormat&&(w=i.transpose(w,[0,4,1,2,3])),null!==this.bias&&(w=ft(w,this.bias.read(),this.dataFormat)),null!==this.activation&&(w=this.activation.apply(w)),w}))}computeOutputShape(t){const e=(t=Wt(t)).slice();let n,s,i,r;"channelsFirst"===this.dataFormat?(n=1,s=2,i=3,r=4):(n=4,s=1,i=2,r=3);const a=this.kernelSize[0],o=this.kernelSize[1],l=this.kernelSize[2],u=this.strides[0],h=this.strides[1],c=this.strides[2];return e[n]=this.filters,e[s]=Ru(e[s],u,a,this.padding),e[i]=Ru(e[i],h,o,this.padding),e[r]=Ru(e[r],c,l,this.padding),e}getConfig(){const t=super.getConfig();return delete t.dilationRate,t}}Vu.className="Conv3DTranspose",e.serialization.registerClass(Vu);class Ku extends Uu{constructor(t,e){if(super(t,e),this.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",this.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",this.depthwiseKernel=null,this.pointwiseKernel=null,null==e.filters)throw new o("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(null!=e.kernelInitializer||null!=e.kernelRegularizer||null!=e.kernelConstraint)throw new o("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(null!=e.padding&&"same"!==e.padding&&"valid"!==e.padding)throw new o(`SeparableConv${this.rank}D supports only padding modes: 'same' and 'valid', but received ${JSON.stringify(e.padding)}`);this.depthMultiplier=null==e.depthMultiplier?1:e.depthMultiplier,this.depthwiseInitializer=Mt(e.depthwiseInitializer||this.DEFAULT_DEPTHWISE_INITIALIZER),this.depthwiseRegularizer=zu(e.depthwiseRegularizer),this.depthwiseConstraint=Uo(e.depthwiseConstraint),this.pointwiseInitializer=Mt(e.depthwiseInitializer||this.DEFAULT_POINTWISE_INITIALIZER),this.pointwiseRegularizer=zu(e.pointwiseRegularizer),this.pointwiseConstraint=Uo(e.pointwiseConstraint)}build(t){if((t=Wt(t)).length<this.rank+2)throw new o(`Inputs to SeparableConv${this.rank}D should have rank ${this.rank+2}, but received input shape: ${JSON.stringify(t)}`);const e="channelsFirst"===this.dataFormat?1:t.length-1;if(null==t[e]||t[e]<0)throw new o(`The channel dimension of the inputs should be defined, but found ${JSON.stringify(t[e])}`);const n=t[e],s=this.kernelSize.concat([n,this.depthMultiplier]),i=[];for(let t=0;t<this.rank;++t)i.push(1);i.push(n*this.depthMultiplier,this.filters);const r=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",s,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,r,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",i,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,r,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,r,this.biasConstraint):this.bias=null,this.inputSpec=[new Ht({ndim:this.rank+2,axes:{[e]:n}})],this.built=!0}call(t,n){return e.tidy((()=>{let e;if(t=Ut(t),1===this.rank)throw new l("1D separable convolution is not implemented yet.");return 2===this.rank&&("channelsFirst"===this.dataFormat&&(t=i.transpose(t,[0,2,3,1])),e=i.separableConv2d(t,this.depthwiseKernel.read(),this.pointwiseKernel.read(),this.strides,this.padding,this.dilationRate,"NHWC")),this.useBias&&(e=ft(e,this.bias.read(),this.dataFormat)),null!=this.activation&&(e=this.activation.apply(e)),"channelsFirst"===this.dataFormat&&(e=i.transpose(e,[0,3,1,2])),e}))}getConfig(){const t=super.getConfig();return delete t.rank,delete t.kernelInitializer,delete t.kernelRegularizer,delete t.kernelConstraint,t.depthwiseInitializer=Ot(this.depthwiseInitializer),t.pointwiseInitializer=Ot(this.pointwiseInitializer),t.depthwiseRegularizer=Iu(this.depthwiseRegularizer),t.pointwiseRegularizer=Iu(this.pointwiseRegularizer),t.depthwiseConstraint=Bo(this.depthwiseConstraint),t.pointwiseConstraint=Bo(this.pointwiseConstraint),t}}Ku.className="SeparableConv";class Gu extends Ku{constructor(t){super(2,t)}}Gu.className="SeparableConv2D",e.serialization.registerClass(Gu);class Hu extends Uu{constructor(t){super(1,t),Hu.verifyArgs(t),this.inputSpec=[{ndim:3}]}getConfig(){const t=super.getConfig();return delete t.rank,delete t.dataFormat,t}static verifyArgs(t){if("number"!=typeof t.kernelSize&&!A(t.kernelSize,"number",1,1))throw new o(`Conv1D expects config.kernelSize to be number or number[] with length 1, but received ${JSON.stringify(t.kernelSize)}.`)}}Hu.className="Conv1D",e.serialization.registerClass(Hu);class Ju extends Qt{constructor(t){super(t),"number"==typeof t.cropping?this.cropping=[[t.cropping,t.cropping],[t.cropping,t.cropping]]:"number"==typeof t.cropping[0]?this.cropping=[[t.cropping[0],t.cropping[0]],[t.cropping[1],t.cropping[1]]]:this.cropping=t.cropping,this.dataFormat=void 0===t.dataFormat?"channelsLast":t.dataFormat,this.inputSpec=[{ndim:4}]}computeOutputShape(t){return"channelsFirst"===this.dataFormat?[t[0],t[1],t[2]-this.cropping[0][0]-this.cropping[0][1],t[3]-this.cropping[1][0]-this.cropping[1][1]]:[t[0],t[1]-this.cropping[0][0]-this.cropping[0][1],t[2]-this.cropping[1][0]-this.cropping[1][1],t[3]]}call(t,n){return e.tidy((()=>{if(t=Ut(t),"channelsLast"===this.dataFormat){const e=rt(t,this.cropping[0][0],t.shape[1]-this.cropping[0][0]-this.cropping[0][1],2);return rt(e,this.cropping[1][0],t.shape[2]-this.cropping[1][1]-this.cropping[1][0],3)}{const e=rt(t,this.cropping[0][0],t.shape[2]-this.cropping[0][0]-this.cropping[0][1],3);return rt(e,this.cropping[1][0],t.shape[3]-this.cropping[1][1]-this.cropping[1][0],4)}}))}getConfig(){const t={cropping:this.cropping,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}}Ju.className="Cropping2D",e.serialization.registerClass(Ju);class Zu extends Qt{constructor(t){var e;super(t),this.DEFAULT_SIZE=[2,2],this.inputSpec=[{ndim:4}],this.size=null==t.size?this.DEFAULT_SIZE:t.size,this.dataFormat=null==t.dataFormat?"channelsLast":t.dataFormat,P(this.dataFormat),this.interpolation=null==t.interpolation?"nearest":t.interpolation,e=this.interpolation,I(_,"InterpolationFormat",e)}computeOutputShape(t){if("channelsFirst"===this.dataFormat){const e=null==t[2]?null:this.size[0]*t[2],n=null==t[3]?null:this.size[1]*t[3];return[t[0],t[1],e,n]}{const e=null==t[1]?null:this.size[0]*t[1],n=null==t[2]?null:this.size[1]*t[2];return[t[0],e,n,t[3]]}}call(t,e){return i.tidy((()=>{let e=Ut(t);const n=e.shape;if("channelsFirst"===this.dataFormat){e=i.transpose(e,[0,2,3,1]);const t=this.size[0]*n[2],s=this.size[1]*n[3],r="nearest"===this.interpolation?i.image.resizeNearestNeighbor(e,[t,s]):i.image.resizeBilinear(e,[t,s]);return i.transpose(r,[0,3,1,2])}{const t=this.size[0]*n[1],s=this.size[1]*n[2];return"nearest"===this.interpolation?i.image.resizeNearestNeighbor(e,[t,s]):i.image.resizeBilinear(e,[t,s])}}))}getConfig(){const t={size:this.size,dataFormat:this.dataFormat,interpolation:this.interpolation},e=super.getConfig();return Object.assign(t,e),t}}Zu.className="UpSampling2D",e.serialization.registerClass(Zu);class Yu extends Pu{constructor(t){super(2,t),this.depthwiseKernel=null,this.depthMultiplier=null==t.depthMultiplier?1:t.depthMultiplier,this.depthwiseInitializer=Mt(t.depthwiseInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.depthwiseConstraint=Uo(t.depthwiseConstraint),this.depthwiseRegularizer=zu(t.depthwiseRegularizer)}build(t){if((t=Wt(t)).length<4)throw new o(`Inputs to DepthwiseConv2D should have rank 4. Received input shape: ${JSON.stringify(t)}.`);const e="channelsFirst"===this.dataFormat?1:3;if(null==t[e]||t[e]<0)throw new o(`The channel dimension of the inputs to DepthwiseConv2D should be defined, but is not (${t[e]}).`);const n=t[e],s=[this.kernelSize[0],this.kernelSize[1],n,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",s,null,this.depthwiseInitializer,this.depthwiseRegularizer,!0,this.depthwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[n*this.depthMultiplier],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0}call(t,n){return e.tidy((()=>{let n=function(t,n,s=[1,1],r="valid",a,l){return e.tidy((()=>{null==a&&(a="channelsLast"),P(a);let e=Ou(t,a);if(4!==t.rank)throw new o(`Input for depthwiseConv2d is required to be 4-D, but is instead ${t.rank}-D`);if(4!==n.rank)throw new o(`depthwiseKernel is required to be 4-D, but is instead ${n.rank}-D`);return e=i.depthwiseConv2d(e,n,s,"same"===r?"same":"valid","NHWC",l),"channelsFirst"===a&&(e=i.transpose(e,[0,3,1,2])),e}))}(t=Ut(t),this.depthwiseKernel.read(),this.strides,this.padding,this.dataFormat,null);return this.useBias&&(n=ft(n,this.bias.read(),this.dataFormat)),null!=this.activation&&(n=this.activation.apply(n)),n}))}computeOutputShape(t){t=Wt(t);const e="channelsFirst"===this.dataFormat?t[2]:t[1],n="channelsFirst"===this.dataFormat?t[3]:t[2],s="channelsFirst"===this.dataFormat?t[1]*this.depthMultiplier:t[3]*this.depthMultiplier,i=_u(e,this.kernelSize[0],this.padding,this.strides[0]),r=_u(n,this.kernelSize[1],this.padding,this.strides[1]);return"channelsFirst"===this.dataFormat?[t[0],s,i,r]:[t[0],i,r,s]}getConfig(){const t=super.getConfig();return t.depthMultiplier=this.depthMultiplier,t.depthwiseInitializer=Ot(this.depthwiseInitializer),t.depthwiseRegularizer=Iu(this.depthwiseRegularizer),t.depthwiseConstraint=Bo(this.depthwiseRegularizer),t}}function Xu(t,e,n,s){if(Array.isArray(t)){if(null!=e||null!=n)throw new o("When inputs is an array, neither initialState or constants should be provided");null!=s&&(n=t.slice(t.length-s,t.length),t=t.slice(0,t.length-s)),t.length>1&&(e=t.slice(1,t.length)),t=t[0]}function i(t){return null==t||Array.isArray(t)?t:[t]}return{inputs:t,initialState:e=i(e),constants:n=i(n)}}function Qu(t,e,n,s=!1,r,a,u=!1,h=!1){return i.tidy((()=>{const c=e.shape.length;if(c<3)throw new o(`Input should be at least 3D, but is ${c}D.`);const p=[1,0].concat(X(2,c));if(e=i.transpose(e,p),null!=a)throw new l("The rnn() functoin of the deeplearn.js backend does not support constants yet.");u&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),null!=r&&((r=i.cast(i.cast(r,"bool"),"float32")).rank===c-1&&(r=i.expandDims(r,-1)),r=i.transpose(r,p)),s&&(e=i.reverse(e,0),null!=r&&(r=i.reverse(r,0)));const d=[];let f,g=n;const m=e.shape[0],y=i.unstack(e);let b,w;null!=r&&(b=i.unstack(r));for(let e=0;e<m;++e){const n=y[e],s=i.tidy((()=>t(n,g)));if(null==r)f=s[0],g=s[1];else{const t=i.tidy((()=>{const t=b[e],n=i.sub(i.onesLike(t),t);return{output:i.add(i.mul(s[0],t),i.mul(g[0],n)),newStates:g.map(((e,r)=>i.add(i.mul(s[1][r],t),i.mul(e,n))))}}));f=t.output,g=t.newStates}h&&d.push(f)}if(h){const t=1;w=i.stack(d,t)}return[f,w,g]}))}Yu.className="DepthwiseConv2D",e.serialization.registerClass(Yu);class th extends Qt{constructor(t){let e;if(super(t),null==t.cell)throw new o("cell property is missing for the constructor of RNN.");if(e=Array.isArray(t.cell)?new lh({cells:t.cell}):t.cell,null==e.stateSize)throw new o("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");this.cell=e,this.returnSequences=null!=t.returnSequences&&t.returnSequences,this.returnState=null!=t.returnState&&t.returnState,this.goBackwards=null!=t.goBackwards&&t.goBackwards,this._stateful=null!=t.stateful&&t.stateful,this.unroll=null!=t.unroll&&t.unroll,this.supportsMasking=!0,this.inputSpec=[new Ht({ndim:3})],this.stateSpec=null,this.states_=null,this.numConstants=null,this.keptStates=[]}getStates(){if(null==this.states_){return X(0,Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1).map((t=>null))}return this.states_}setStates(t){this.states_=t}computeOutputShape(t){Bt(t)&&(t=t[0]);let e=this.cell.stateSize;Array.isArray(e)||(e=[e]);const n=e[0];let s;if(s=this.returnSequences?[t[0],t[1],n]:[t[0],n],this.returnState){const n=[];for(const s of e)n.push([t[0],s]);return[s].concat(n)}return s}computeMask(t,e){return i.tidy((()=>{Array.isArray(e)&&(e=e[0]);const t=this.returnSequences?e:null;if(this.returnState){const e=this.states.map((t=>null));return[t].concat(e)}return t}))}get states(){if(null==this.states_){const t=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,e=[];for(let n=0;n<t;++n)e.push(null);return e}return this.states_}set states(t){this.states_=t}build(t){if(null!=this.numConstants)throw new l("Constants support is not implemented in RNN yet.");Bt(t)&&(t=t[0]);const n=this.stateful?t[0]:null,s=t.slice(2);this.inputSpec[0]=new Ht({shape:[n,null,...s]});const i=[t[0]].concat(t.slice(2));let r;if(this.cell.build(i),r=Array.isArray(this.cell.stateSize)?this.cell.stateSize:[this.cell.stateSize],null!=this.stateSpec){if(!e.util.arraysEqual(this.stateSpec.map((t=>t.shape[t.shape.length-1])),r))throw new o(`An initialState was passed that is not compatible with cell.stateSize. Received stateSpec=${this.stateSpec}; However cell.stateSize is ${this.cell.stateSize}`)}else this.stateSpec=r.map((t=>new Ht({shape:[null,t]})));this.stateful&&this.resetStates()}resetStates(t,n=!1){e.tidy((()=>{if(!this.stateful)throw new r("Cannot call resetStates() on an RNN Layer that is not stateful.");const s=this.inputSpec[0].shape[0];if(null==s)throw new o("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(null==this.states_)Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map((t=>i.zeros([s,t]))):this.states_=[i.zeros([s,this.cell.stateSize])];else if(null==t)i.dispose(this.states_),null!=this.keptStates&&(i.dispose(this.keptStates),this.keptStates=[]),Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map((t=>i.zeros([s,t]))):this.states_[0]=i.zeros([s,this.cell.stateSize]);else{if(Array.isArray(t)||(t=[t]),t.length!==this.states_.length)throw new o(`Layer ${this.name} expects ${this.states_.length} state(s), but it received ${t.length} state value(s). Input received: ${t}`);!0===n?this.keptStates.push(this.states_.slice()):i.dispose(this.states_);for(let n=0;n<this.states_.length;++n){const i=t[n],r=Array.isArray(this.cell.stateSize)?this.cell.stateSize[n]:this.cell.stateSize,a=[s,r];if(!e.util.arraysEqual(i.shape,a))throw new o(`State ${n} is incompatible with layer ${this.name}: expected shape=${a}, received shape=${i.shape}`);this.states_[n]=i}}this.states_=this.states_.map((t=>i.keep(t.clone())))}))}apply(t,e){let n=null==e?null:e.initialState,s=null==e?null:e.constants;null==e&&(e={});const i=Xu(t,n,s,this.numConstants);t=i.inputs,n=i.initialState,s=i.constants;let r=[],a=[];if(null!=n){e.initialState=n,r=r.concat(n),this.stateSpec=[];for(const t of n)this.stateSpec.push(new Ht({shape:t.shape}));a=a.concat(this.stateSpec)}null!=s&&(e.constants=s,r=r.concat(s),this.numConstants=s.length);if(r[0]instanceof Jt){const n=[t].concat(r),s=this.inputSpec.concat(a),i=this.inputSpec;this.inputSpec=s;const o=super.apply(n,e);return this.inputSpec=i,o}return super.apply(t,e)}call(t,n){return e.tidy((()=>{const e=null==n?null:n.mask,s=null==n?null:n.training;let i=null==n?null:n.initialState;t=Ut(t),null==i&&(i=this.stateful?this.states_:this.getInitialState(t));const r=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;if(i.length!==r)throw new o(`RNN Layer has ${r} state(s) but was passed ${i.length} initial state(s).`);this.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");const a={training:s},l=Qu(((t,e)=>{const n=this.cell.call([t].concat(e),a);return[n[0],n.slice(1)]}),t,i,this.goBackwards,e,null,this.unroll,this.returnSequences),u=l[0],h=l[1],c=l[2];this.stateful&&this.resetStates(c,s);const p=this.returnSequences?h:u;return this.returnState?[p].concat(c):p}))}getInitialState(t){return e.tidy((()=>{let e=i.zeros(t.shape);return e=i.sum(e,[1,2]),e=nt(e),Array.isArray(this.cell.stateSize)?this.cell.stateSize.map((t=>t>1?lt(e,[1,t]):e)):this.cell.stateSize>1?[lt(e,[1,this.cell.stateSize])]:[e]}))}get trainableWeights(){return this.trainable?this.cell.trainableWeights:[]}get nonTrainableWeights(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights}setFastWeightInitDuringBuild(t){super.setFastWeightInitDuringBuild(t),null!=this.cell&&this.cell.setFastWeightInitDuringBuild(t)}getConfig(){const t=super.getConfig(),e={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};null!=this.numConstants&&(e.numConstants=this.numConstants);const n=this.cell.getConfig();return this.getClassName()===th.className&&(e.cell={className:this.cell.getClassName(),config:n}),Object.assign(Object.assign(Object.assign({},n),t),e)}static fromConfig(t,e,n={}){const s=el(e.cell,n);return new t(Object.assign(e,{cell:s}))}}th.className="RNN",e.serialization.registerClass(th);class eh extends Qt{}class nh extends eh{constructor(t){super(t),this.DEFAULT_ACTIVATION="tanh",this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_RECURRENT_INITIALIZER="orthogonal",this.DEFAULT_BIAS_INITIALIZER="zeros",this.units=t.units,z(this.units,"units"),this.activation=ku(null==t.activation?this.DEFAULT_ACTIVATION:t.activation),this.useBias=null==t.useBias||t.useBias,this.kernelInitializer=Mt(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=Mt(t.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=Mt(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelRegularizer=zu(t.kernelRegularizer),this.recurrentRegularizer=zu(t.recurrentRegularizer),this.biasRegularizer=zu(t.biasRegularizer),this.kernelConstraint=Uo(t.kernelConstraint),this.recurrentConstraint=Uo(t.recurrentConstraint),this.biasConstraint=Uo(t.biasConstraint),this.dropout=Z([1,Y([0,null==t.dropout?0:t.dropout])]),this.recurrentDropout=Z([1,Y([0,null==t.recurrentDropout?0:t.recurrentDropout])]),this.dropoutFunc=t.dropoutFunc,this.stateSize=this.units,this.dropoutMask=null,this.recurrentDropoutMask=null}build(t){t=Wt(t),this.kernel=this.addWeight("kernel",[t[t.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}call(t,n){return e.tidy((()=>{if(2!==t.length)throw new o(`SimpleRNNCell expects 2 input Tensors, got ${t.length}.`);let e=t[1];t=t[0];const s=null!=n.training&&n.training;let r;0<this.dropout&&this.dropout<1&&null==this.dropoutMask&&(this.dropoutMask=uh({ones:()=>i.onesLike(t),rate:this.dropout,training:s,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&null==this.recurrentDropoutMask&&(this.recurrentDropoutMask=uh({ones:()=>i.onesLike(e),rate:this.recurrentDropout,training:s,dropoutFunc:this.dropoutFunc}));const a=this.dropoutMask,l=this.recurrentDropoutMask;r=ht(null!=a?i.mul(t,a):t,this.kernel.read()),null!=this.bias&&(r=ft(r,this.bias.read())),null!=l&&(e=i.mul(e,l));let u=i.add(r,ht(e,this.recurrentKernel.read()));return null!=this.activation&&(u=this.activation.apply(u)),[u,u]}))}getConfig(){const t=super.getConfig(),e={units:this.units,activation:bu(this.activation),useBias:this.useBias,kernelInitializer:Ot(this.kernelInitializer),recurrentInitializer:Ot(this.recurrentInitializer),biasInitializer:Ot(this.biasInitializer),kernelRegularizer:Iu(this.kernelRegularizer),recurrentRegularizer:Iu(this.recurrentRegularizer),biasRegularizer:Iu(this.biasRegularizer),activityRegularizer:Iu(this.activityRegularizer),kernelConstraint:Bo(this.kernelConstraint),recurrentConstraint:Bo(this.recurrentConstraint),biasConstraint:Bo(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return Object.assign(Object.assign({},t),e)}}nh.className="SimpleRNNCell",e.serialization.registerClass(nh);class sh extends th{constructor(t){t.cell=new nh(t),super(t)}call(t,n){return e.tidy((()=>{null!=this.cell.dropoutMask&&(i.dispose(this.cell.dropoutMask),this.cell.dropoutMask=null),null!=this.cell.recurrentDropoutMask&&(i.dispose(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);const e=null==n?null:n.mask,s=null==n?null:n.training,r=null==n?null:n.initialState;return super.call(t,{mask:e,training:s,initialState:r})}))}static fromConfig(t,e){return new t(e)}}sh.className="SimpleRNN",e.serialization.registerClass(sh);class ih extends eh{constructor(t){if(super(t),this.DEFAULT_ACTIVATION="tanh",this.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_RECURRENT_INITIALIZER="orthogonal",this.DEFAULT_BIAS_INITIALIZER="zeros",t.resetAfter)throw new o("GRUCell does not support reset_after parameter set to true.");this.units=t.units,z(this.units,"units"),this.activation=ku(void 0===t.activation?this.DEFAULT_ACTIVATION:t.activation),this.recurrentActivation=ku(void 0===t.recurrentActivation?this.DEFAULT_RECURRENT_ACTIVATION:t.recurrentActivation),this.useBias=null==t.useBias||t.useBias,this.kernelInitializer=Mt(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=Mt(t.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=Mt(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelRegularizer=zu(t.kernelRegularizer),this.recurrentRegularizer=zu(t.recurrentRegularizer),this.biasRegularizer=zu(t.biasRegularizer),this.kernelConstraint=Uo(t.kernelConstraint),this.recurrentConstraint=Uo(t.recurrentConstraint),this.biasConstraint=Uo(t.biasConstraint),this.dropout=Z([1,Y([0,null==t.dropout?0:t.dropout])]),this.recurrentDropout=Z([1,Y([0,null==t.recurrentDropout?0:t.recurrentDropout])]),this.dropoutFunc=t.dropoutFunc,this.implementation=t.implementation,this.stateSize=this.units,this.dropoutMask=null,this.recurrentDropoutMask=null}build(t){const e=(t=Wt(t))[t.length-1];this.kernel=this.addWeight("kernel",[e,3*this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,3*this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[3*this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0}call(t,n){return e.tidy((()=>{if(2!==t.length)throw new o(`GRUCell expects 2 input Tensors (inputs, h, c), got ${t.length}.`);const e=null!=n.training&&n.training;let s=t[1];t=t[0],0<this.dropout&&this.dropout<1&&null==this.dropoutMask&&(this.dropoutMask=uh({ones:()=>i.onesLike(t),rate:this.dropout,training:e,count:3,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&null==this.recurrentDropoutMask&&(this.recurrentDropoutMask=uh({ones:()=>i.onesLike(s),rate:this.recurrentDropout,training:e,count:3,dropoutFunc:this.dropoutFunc}));const r=this.dropoutMask,a=this.recurrentDropoutMask;let l,u,h;0<this.dropout&&this.dropout<1&&(t=i.mul(t,r[0]));let c=ht(t,this.kernel.read());this.useBias&&(c=ft(c,this.bias.read())),0<this.recurrentDropout&&this.recurrentDropout<1&&(s=i.mul(s,a[0]));const p=this.recurrentKernel.read(),[d,f]=i.split(p,[2*this.units,this.units],p.rank-1),g=ht(s,d),[m,y,b]=i.split(c,3,c.rank-1),[w,k]=i.split(g,2,g.rank-1);l=this.recurrentActivation.apply(i.add(m,w)),u=this.recurrentActivation.apply(i.add(y,k));const v=ht(i.mul(u,s),f);h=this.activation.apply(i.add(b,v));const S=i.add(i.mul(l,s),i.mul(i.add(1,i.neg(l)),h));return[S,S]}))}getConfig(){const t=super.getConfig(),e={units:this.units,activation:bu(this.activation),recurrentActivation:bu(this.recurrentActivation),useBias:this.useBias,kernelInitializer:Ot(this.kernelInitializer),recurrentInitializer:Ot(this.recurrentInitializer),biasInitializer:Ot(this.biasInitializer),kernelRegularizer:Iu(this.kernelRegularizer),recurrentRegularizer:Iu(this.recurrentRegularizer),biasRegularizer:Iu(this.biasRegularizer),activityRegularizer:Iu(this.activityRegularizer),kernelConstraint:Bo(this.kernelConstraint),recurrentConstraint:Bo(this.recurrentConstraint),biasConstraint:Bo(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return Object.assign(Object.assign({},t),e)}}ih.className="GRUCell",e.serialization.registerClass(ih);class rh extends th{constructor(t){0===t.implementation&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),t.cell=new ih(t),super(t)}call(t,n){return e.tidy((()=>{null!=this.cell.dropoutMask&&(i.dispose(this.cell.dropoutMask),this.cell.dropoutMask=null),null!=this.cell.recurrentDropoutMask&&(i.dispose(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);const e=null==n?null:n.mask,s=null==n?null:n.training,r=null==n?null:n.initialState;return super.call(t,{mask:e,training:s,initialState:r})}))}static fromConfig(t,e){return 0===e.implmentation&&(e.implementation=1),new t(e)}}rh.className="GRU",e.serialization.registerClass(rh);class ah extends eh{constructor(t){super(t),this.DEFAULT_ACTIVATION="tanh",this.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_RECURRENT_INITIALIZER="orthogonal",this.DEFAULT_BIAS_INITIALIZER="zeros",this.units=t.units,z(this.units,"units"),this.activation=ku(void 0===t.activation?this.DEFAULT_ACTIVATION:t.activation),this.recurrentActivation=ku(void 0===t.recurrentActivation?this.DEFAULT_RECURRENT_ACTIVATION:t.recurrentActivation),this.useBias=null==t.useBias||t.useBias,this.kernelInitializer=Mt(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=Mt(t.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=Mt(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.unitForgetBias=t.unitForgetBias,this.kernelRegularizer=zu(t.kernelRegularizer),this.recurrentRegularizer=zu(t.recurrentRegularizer),this.biasRegularizer=zu(t.biasRegularizer),this.kernelConstraint=Uo(t.kernelConstraint),this.recurrentConstraint=Uo(t.recurrentConstraint),this.biasConstraint=Uo(t.biasConstraint),this.dropout=Z([1,Y([0,null==t.dropout?0:t.dropout])]),this.recurrentDropout=Z([1,Y([0,null==t.recurrentDropout?0:t.recurrentDropout])]),this.dropoutFunc=t.dropoutFunc,this.implementation=t.implementation,this.stateSize=[this.units,this.units],this.dropoutMask=null,this.recurrentDropoutMask=null}build(t){var e;const n=(t=Wt(t))[t.length-1];let s;if(this.kernel=this.addWeight("kernel",[n,4*this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,4*this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){if(this.unitForgetBias){const t=this.biasInitializer,n=this.units;s=new((e=class extends wt{apply(e,s){const i=t.apply([n]),r=(new vt).apply([n]),a=t.apply([2*n]);return ot(ot(i,r),a)}}).className="CustomInit",e)}else s=this.biasInitializer;this.bias=this.addWeight("bias",[4*this.units],null,s,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0}call(t,n){return e.tidy((()=>{const e=null!=n.training&&n.training;if(3!==t.length)throw new o(`LSTMCell expects 3 input Tensors (inputs, h, c), got ${t.length}.`);let s=t[1];const r=t[2];t=t[0],0<this.dropout&&this.dropout<1&&null==this.dropoutMask&&(this.dropoutMask=uh({ones:()=>i.onesLike(t),rate:this.dropout,training:e,count:4,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&null==this.recurrentDropoutMask&&(this.recurrentDropoutMask=uh({ones:()=>i.onesLike(s),rate:this.recurrentDropout,training:e,count:4,dropoutFunc:this.dropoutFunc}));const a=this.dropoutMask,l=this.recurrentDropoutMask;let u,h,c,p;0<this.dropout&&this.dropout<1&&(t=i.mul(t,a[0]));let d=ht(t,this.kernel.read());0<this.recurrentDropout&&this.recurrentDropout<1&&(s=i.mul(s,l[0])),d=i.add(d,ht(s,this.recurrentKernel.read())),this.useBias&&(d=ft(d,this.bias.read()));const[f,g,m,y]=i.split(d,4,d.rank-1);u=this.recurrentActivation.apply(f),h=this.recurrentActivation.apply(g),c=i.add(i.mul(h,r),i.mul(u,this.activation.apply(m))),p=this.recurrentActivation.apply(y);const b=i.mul(p,this.activation.apply(c));return[b,b,c]}))}getConfig(){const t=super.getConfig(),e={units:this.units,activation:bu(this.activation),recurrentActivation:bu(this.recurrentActivation),useBias:this.useBias,kernelInitializer:Ot(this.kernelInitializer),recurrentInitializer:Ot(this.recurrentInitializer),biasInitializer:Ot(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:Iu(this.kernelRegularizer),recurrentRegularizer:Iu(this.recurrentRegularizer),biasRegularizer:Iu(this.biasRegularizer),activityRegularizer:Iu(this.activityRegularizer),kernelConstraint:Bo(this.kernelConstraint),recurrentConstraint:Bo(this.recurrentConstraint),biasConstraint:Bo(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return Object.assign(Object.assign({},t),e)}}ah.className="LSTMCell",e.serialization.registerClass(ah);class oh extends th{constructor(t){0===t.implementation&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),t.cell=new ah(t),super(t)}call(t,n){return e.tidy((()=>{null!=this.cell.dropoutMask&&(i.dispose(this.cell.dropoutMask),this.cell.dropoutMask=null),null!=this.cell.recurrentDropoutMask&&(i.dispose(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);const e=null==n?null:n.mask,s=null==n?null:n.training,r=null==n?null:n.initialState;return super.call(t,{mask:e,training:s,initialState:r})}))}static fromConfig(t,e){return 0===e.implmentation&&(e.implementation=1),new t(e)}}oh.className="LSTM",e.serialization.registerClass(oh);class lh extends eh{constructor(t){super(t),this.cells=t.cells}get stateSize(){const t=[];for(const e of this.cells.slice().reverse())Array.isArray(e.stateSize)?t.push(...e.stateSize):t.push(e.stateSize);return t}call(t,n){return e.tidy((()=>{let e=t.slice(1);const s=[];for(const t of this.cells.slice().reverse())Array.isArray(t.stateSize)?s.push(e.splice(0,t.stateSize.length)):s.push(e.splice(0,1));s.reverse();const i=[];let r;for(let a=0;a<this.cells.length;++a){const o=this.cells[a];e=s[a],r=0===a?[t[0]].concat(e):[r[0]].concat(e),r=o.call(r,n),i.push(r.slice(1))}e=[];for(const t of i.slice().reverse())e.push(...t);return[r[0]].concat(e)}))}build(t){let e;Bt(t)&&(t=t[0]),this.cells.forEach(((n,s)=>{q(`RNNCell_${s}`,(()=>{n.build(t),e=Array.isArray(n.stateSize)?n.stateSize[0]:n.stateSize,t=[t[0],e]}))})),this.built=!0}getConfig(){const t=super.getConfig(),e={cells:this.cells.map((t=>({className:t.getClassName(),config:t.getConfig()})))};return Object.assign(Object.assign({},t),e)}static fromConfig(t,e,n={}){const s=[];for(const t of e.cells)s.push(el(t,n));return new t({cells:s})}get trainableWeights(){if(!this.trainable)return[];const t=[];for(const e of this.cells)t.push(...e.trainableWeights);return t}get nonTrainableWeights(){const t=[];for(const e of this.cells)t.push(...e.nonTrainableWeights);if(!this.trainable){const e=[];for(const t of this.cells)e.push(...t.trainableWeights);return e.concat(t)}return t}getWeights(){const t=[];for(const e of this.cells)t.push(...e.weights);return Kt(t)}setWeights(t){const e=[];for(const n of this.cells){const s=n.weights.length,i=t.splice(s);for(let t=0;t<n.weights.length;++t)e.push([n.weights[t],i[t]])}Gt(e)}}function uh(t){const{ones:e,rate:n,training:s=!1,count:r=1,dropoutFunc:a}=t,o=()=>null!=a?a(e(),n):gt(e(),n),l=()=>mt(o,e,s);if(!r||r<=1)return i.keep(l().clone());return Array(r).fill(void 0).map(l).map((t=>i.keep(t.clone())))}lh.className="StackedRNNCells",e.serialization.registerClass(lh);var hh=function(t,e){var n={};for(var s in t)Object.prototype.hasOwnProperty.call(t,s)&&e.indexOf(s)<0&&(n[s]=t[s]);if(null!=t&&"function"==typeof Object.getOwnPropertySymbols){var i=0;for(s=Object.getOwnPropertySymbols(t);i<s.length;i++)e.indexOf(s[i])<0&&Object.prototype.propertyIsEnumerable.call(t,s[i])&&(n[s[i]]=t[s[i]])}return n};class ch extends th{constructor(t){if(t.unroll)throw new l("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(t.cell))throw new l("It is not possible at the moment to stack convolutional cells.");super(t),this.inputSpec=[new Ht({ndim:5})]}call(t,e){return i.tidy((()=>{if(null!=this.cell.dropoutMask&&(i.dispose(this.cell.dropoutMask),this.cell.dropoutMask=null),null!=this.cell.recurrentDropoutMask&&(i.dispose(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null),e&&e.constants)throw new o("ConvRNN2D cell does not support constants");const n=null==e?null:e.mask,s=null==e?null:e.training,r=null==e?null:e.initialState;return super.call(t,{mask:n,training:s,initialState:r})}))}computeOutputShape(t){let e=this.computeSingleOutputShape(t);return this.returnSequences||(e=[e[0],...e.slice(2)]),this.returnState&&(e=[e,...Array(2).fill([t[0],...e.slice(-3)])]),e}getInitialState(t){return i.tidy((()=>{const{stateSize:e}=this.cell,n=t.shape,s=this.computeSingleOutputShape(n),r=[s[0],...s.slice(2)],a=i.zeros(r);return Array.isArray(e)?Array(e.length).fill(a):[a]}))}resetStates(t,n=!1){i.tidy((()=>{if(!this.stateful)throw new r("Cannot call resetStates() on an RNN Layer that is not stateful.");const s=this.inputSpec[0].shape,a=this.computeSingleOutputShape(s),l=[a[0],...a.slice(2)];if(null==s[0])throw new o("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(null==this.getStates())Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map((()=>i.zeros(l))):this.states_=[i.zeros(l)];else if(null==t)i.dispose(this.states_),null!=this.keptStates&&(i.dispose(this.keptStates),this.keptStates=[]),Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map((()=>i.zeros(l))):this.states_[0]=i.zeros(l);else{if(Array.isArray(t)||(t=[t]),t.length!==this.states_.length)throw new o(`Layer ${this.name} expects ${this.states_.length} state(s), but it received ${t.length} state value(s). Input received: ${t}`);n?this.keptStates.push(this.states_.slice()):i.dispose(this.states_);for(let n=0;n<this.states_.length;++n){const s=t[n],i=l;if(!e.util.arraysEqual(s.shape,i))throw new o(`State ${n} is incompatible with layer ${this.name}: expected shape=${i}, received shape=${s.shape}`);this.states_[n]=s}}this.states_=this.states_.map((t=>i.keep(t.clone())))}))}computeSingleOutputShape(t){const{dataFormat:e,filters:n,kernelSize:s,padding:i,strides:r,dilationRate:a}=this.cell,o="channelsFirst"===e,l=t[o?3:2],u=t[o?4:3],h=_u(l,s[0],i,r[0],a[0]),c=_u(u,s[1],i,r[1],a[1]);return[...t.slice(0,2),...o?[n,h,c]:[h,c,n]]}}ch.className="ConvRNN2D";class ph extends ah{constructor(t){const{filters:e,kernelSize:n,strides:s,padding:i,dataFormat:r,dilationRate:a}=t;super(Object.assign(Object.assign({},t),{units:e})),this.filters=e,z(this.filters,"filters"),this.kernelSize=Lu(n,2,"kernelSize"),this.kernelSize.forEach((t=>z(t,"kernelSize"))),this.strides=Lu(s||1,2,"strides"),this.strides.forEach((t=>z(t,"strides"))),this.padding=i||"valid",U(this.padding),this.dataFormat=r||"channelsLast",P(this.dataFormat),this.dilationRate=Lu(a||1,2,"dilationRate"),this.dilationRate.forEach((t=>z(t,"dilationRate")))}build(t){var e;t=Wt(t);const n="channelsFirst"===this.dataFormat?1:t.length-1;if(null==t[n])throw new o(`The channel dimension of the input should be defined. Found ${t[n]}`);const s=t[n],r=this.kernelSize.concat([s,4*this.filters]);this.kernel=this.addWeight("kernel",r,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);const a=this.kernelSize.concat([this.filters,4*this.filters]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",a,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){let t;if(this.unitForgetBias){const n=this.biasInitializer,s=this.filters;t=new((e=class extends wt{apply(t,e){return at([n.apply([s]),i.ones([s]),n.apply([2*s])])}}).className="CustomInit",e)}else t=this.biasInitializer;this.bias=this.addWeight("bias",[4*this.filters],null,t,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0}call(t,e){return i.tidy((()=>{if(3!==t.length)throw new o(`ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ${t.length}.`);const n=e.training||!1,s=t[0],r=t[1],a=t[2];0<this.dropout&&this.dropout<1&&null==this.dropoutMask&&(this.dropoutMask=uh({ones:()=>i.onesLike(s),rate:this.dropout,training:n,count:4,dropoutFunc:this.dropoutFunc}));const l=this.dropoutMask,u=(t,e,n)=>e&&e[n]?i.mul(e[n],t):t;let h=u(s,l,0),c=u(s,l,1),p=u(s,l,2),d=u(s,l,3);0<this.recurrentDropout&&this.recurrentDropout<1&&null==this.recurrentDropoutMask&&(this.recurrentDropoutMask=uh({ones:()=>i.onesLike(r),rate:this.recurrentDropout,training:n,count:4,dropoutFunc:this.dropoutFunc}));const f=this.recurrentDropoutMask;let g=u(r,f,0),m=u(r,f,1),y=u(r,f,2),b=u(r,f,3);const[w,k,v,S]=i.split(this.kernel.read(),4,3),[x,N,I,A]=this.useBias?i.split(this.bias.read(),4):[null,null,null,null];h=this.inputConv(h,w,x,this.padding),c=this.inputConv(c,k,N,this.padding),p=this.inputConv(p,v,I,this.padding),d=this.inputConv(d,S,A,this.padding);const[z,E,T,C]=i.split(this.recurrentKernel.read(),4,3);g=this.recurrentConv(g,z),m=this.recurrentConv(m,E),y=this.recurrentConv(y,T),b=this.recurrentConv(b,C);const $=this.recurrentActivation.apply(i.add(h,g)),F=this.recurrentActivation.apply(i.add(c,m)),D=i.add(i.mul(F,a),i.mul($,this.activation.apply(i.add(p,y)))),L=i.mul(this.recurrentActivation.apply(i.add(d,b)),this.activation.apply(D));return[L,L,D]}))}getConfig(){const t=super.getConfig(),e=hh(t,["units"]),n={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return Object.assign(Object.assign({},e),n)}inputConv(t,e,n,s){const r=i.conv2d(t,e,this.strides,s||"valid","channelsFirst"===this.dataFormat?"NCHW":"NHWC",this.dilationRate);return n?ft(r,n,this.dataFormat):r}recurrentConv(t,e){return i.conv2d(t,e,1,"same","channelsFirst"===this.dataFormat?"NCHW":"NHWC")}}ph.className="ConvLSTM2DCell",i.serialization.registerClass(ph);class dh extends ch{constructor(t){const e=new ph(t);super(Object.assign(Object.assign({},t),{cell:e}))}static fromConfig(t,e){return new t(e)}}dh.className="ConvLSTM2D",i.serialization.registerClass(dh);class fh extends Qt{constructor(t){super(t),this.rate=Math.max(Math.min(t.rate,1),0),this.noiseShape=t.noiseShape,this.seed=t.seed,this.supportsMasking=!0}getNoiseShape(t){if(null==this.noiseShape)return this.noiseShape;const e=t.shape,n=[];for(let t=0;t<this.noiseShape.length;++t)n.push(null==this.noiseShape[t]?e[t]:this.noiseShape[t]);return n}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);const e=Ut(t);if(0<this.rate&&this.rate<1){const t=null!=n.training&&n.training,s=this.getNoiseShape(e);return mt((()=>gt(e,this.rate,s,this.seed)),(()=>e),t)}return t}))}getConfig(){const t={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},e=super.getConfig();return Object.assign(t,e),t}dispose(){return super.dispose()}}fh.className="Dropout",e.serialization.registerClass(fh);class gh extends fh{constructor(t){super(t),this.inputSpec=[{ndim:3}]}getNoiseShape(t){const e=t.shape;return[e[0],1,e[2]]}}gh.className="SpatialDropout1D",e.serialization.registerClass(gh);class mh extends Qt{constructor(t){if(super(t),this.activation=null,this.useBias=!0,this.kernel=null,this.bias=null,this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_BIAS_INITIALIZER="zeros",null==t.batchInputShape&&null==t.inputShape&&null!=t.inputDim){let e=null;null!=t.batchSize&&(e=t.batchSize),this.batchInputShape=[e,t.inputDim]}this.units=t.units,z(this.units,"units"),this.activation=ku(t.activation),null!=t.useBias&&(this.useBias=t.useBias),this.kernelInitializer=Mt(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.biasInitializer=Mt(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelConstraint=Uo(t.kernelConstraint),this.biasConstraint=Uo(t.biasConstraint),this.kernelRegularizer=zu(t.kernelRegularizer),this.biasRegularizer=zu(t.biasRegularizer),this.activityRegularizer=zu(t.activityRegularizer),this.supportsMasking=!0,this.inputSpec=[{minNDim:2}]}build(t){const e=(t=Wt(t))[t.length-1];null==this.kernel&&(this.kernel=this.addWeight("kernel",[e,this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint))),this.inputSpec=[{minNDim:2,axes:{[-1]:e}}],this.built=!0}computeOutputShape(t){const e=(t=Wt(t)).slice();return e[e.length-1]=this.units,e}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);const e=Ut(t),s=T(this.activation.getClassName());let i;return null!=s?i=ht(e,this.kernel.read(),s,this.bias?this.bias.read():null):(i=ht(e,this.kernel.read()),null!=this.bias&&(i=ft(i,this.bias.read())),null!=this.activation&&(i=this.activation.apply(i))),i}))}getConfig(){const t={units:this.units,activation:bu(this.activation),useBias:this.useBias,kernelInitializer:Ot(this.kernelInitializer),biasInitializer:Ot(this.biasInitializer),kernelRegularizer:Iu(this.kernelRegularizer),biasRegularizer:Iu(this.biasRegularizer),activityRegularizer:Iu(this.activityRegularizer),kernelConstraint:Bo(this.kernelConstraint),biasConstraint:Bo(this.biasConstraint)},e=super.getConfig();return Object.assign(t,e),t}}mh.className="Dense",e.serialization.registerClass(mh);class yh extends Qt{constructor(t){super(t=t||{}),this.inputSpec=[{minNDim:3}],this.dataFormat=t.dataFormat}computeOutputShape(t){t=Wt(t);for(const e of t.slice(1))if(null==e)throw new o(`The shape of the input to "Flatten" is not fully defined (got ${t.slice(1)}). Make sure to pass a complete "input_shape" or "batch_input_shape" argument to the first layer in your model.`);return[t[0],J(t,1)]}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);let s=Ut(t);if("channelsFirst"===this.dataFormat&&s.rank>1){const t=[0];for(let e=2;e<s.rank;++e)t.push(e);t.push(1),s=e.transpose(s,t)}return function(t){if(t.rank<=1)throw new o(`batchFlatten requires a minimum rank of 2. Got rank: ${t.rank}.`);const e=[t.shape[0],J(t.shape,1)];return i.reshape(t,e)}(s)}))}getConfig(){const t={};null!=this.dataFormat&&(t.dataFormat=this.dataFormat);const e=super.getConfig();return Object.assign(t,e),t}}yh.className="Flatten",e.serialization.registerClass(yh);class bh extends Qt{constructor(t){super(t),this.supportsMasking=!0,this.activation=ku(t.activation)}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);const e=Ut(t);return this.activation.apply(e)}))}getConfig(){const t={activation:bu(this.activation)},e=super.getConfig();return Object.assign(t,e),t}}bh.className="Activation",e.serialization.registerClass(bh);class wh extends Qt{constructor(t){super(t),this.n=t.n,this.inputSpec=[{ndim:2}]}computeOutputShape(t){return[t[0],this.n,t[1]]}call(t,n){return e.tidy((()=>{return t=Ut(t),n=t,s=this.n,e.tidy((()=>{if(2!==n.shape.length)throw new o(`repeat() expects a rank-2 tensor, but received a rank-${n.shape.length} tensor.`);return lt(nt(n,1),[1,s,1])}));var n,s}))}getConfig(){const t={n:this.n},e=super.getConfig();return Object.assign(t,e),t}}wh.className="RepeatVector",e.serialization.registerClass(wh);class kh extends Qt{constructor(t){super(t),this.targetShape=t.targetShape;for(let t=0;t<this.targetShape.length;++t)this.isUnknown(this.targetShape[t])&&(this.targetShape[t]=null)}isUnknown(t){return t<0||null==t}fixUnknownDimension(t,e){const n="Total size of new array must be unchanged.",s=e.slice();let i=1,r=null;for(let t=0;t<s.length;++t){const e=s[t];if(this.isUnknown(e)){if(null!==r)throw new o("Can only specifiy one unknown dimension.");r=t}else i*=e}const a=J(t);if(null!==r){if(0===i||a%i!=0)throw new o(n);s[r]=a/i}else if(a!==i)throw new o(n);return s}computeOutputShape(t){let e=!1;for(let n=0;n<t.length;++n)if(this.isUnknown(t[n])){e=!0;break}return e?t.slice(0,1).concat(this.targetShape):t.slice(0,1).concat(this.fixUnknownDimension(t.slice(1),this.targetShape))}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);const s=Ut(t),i=s.shape,r=i.slice(0,1).concat(this.fixUnknownDimension(i.slice(1),this.targetShape));return e.reshape(s,r)}))}getConfig(){const t={targetShape:this.targetShape},e=super.getConfig();return Object.assign(t,e),t}}kh.className="Reshape",e.serialization.registerClass(kh);class vh extends Qt{constructor(t){if(super(t),null==t.dims)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(t.dims))throw new Error(`Permute constructor requires \`dims\` to be an Array, but received ${t.dims} instead.`);const n=X(1,t.dims.length+1);if(!e.util.arraysEqual(t.dims.slice().sort(),n))throw new Error("Invalid permutation `dims`: "+JSON.stringify(t.dims)+" `dims` must contain consecutive integers starting from 1.");this.dims=t.dims,this.dimsIncludingBatch=[0].concat(this.dims),this.inputSpec=[new Ht({ndim:this.dims.length+1})]}computeOutputShape(t){const e=(t=Wt(t)).slice();return this.dims.forEach(((n,s)=>{e[s+1]=t[n]})),e}call(t,n){return e.transpose(Ut(t),this.dimsIncludingBatch)}getConfig(){const t={dims:this.dims},e=super.getConfig();return Object.assign(t,e),t}}vh.className="Permute",e.serialization.registerClass(vh);class Sh extends Qt{constructor(t){super(null==t?{}:t),this.supportsMasking=!0,this.maskValue=null!=t?null==t.maskValue?0:t.maskValue:0}computeOutputShape(t){return t}getConfig(){const t=super.getConfig(),e={maskValue:this.maskValue};return Object.assign(e,t),e}computeMask(t,n){const s=Ut(t);return e.any(e.notEqual(s,this.maskValue),-1)}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);const s=Ut(t),i=e.any(e.notEqual(s,this.maskValue),-1,!0);return e.mul(s,e.cast(i,s.dtype))}))}}Sh.className="Masking",e.serialization.registerClass(Sh);class xh extends Qt{constructor(t){if(super(t),this.embeddings=null,this.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",null==t.batchInputShape&&null==t.inputShape){let e=null;null!=t.batchSize&&(e=t.batchSize),null==t.inputLength?this.batchInputShape=[e,null]:this.batchInputShape=[e].concat(g(t.inputLength))}this.inputDim=t.inputDim,z(this.inputDim,"inputDim"),this.outputDim=t.outputDim,z(this.outputDim,"outputDim"),this.embeddingsInitializer=Mt(t.embeddingsInitializer||this.DEFAULT_EMBEDDINGS_INITIALIZER),this.embeddingsRegularizer=zu(t.embeddingsRegularizer),this.activityRegularizer=zu(t.activityRegularizer),this.embeddingsConstraint=Uo(t.embeddingsConstraint),this.maskZero=t.maskZero,this.supportsMasking=t.maskZero,this.inputLength=t.inputLength}build(t){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0}warnOnIncompatibleInputShape(t){}computeMask(t,n){return e.tidy((()=>this.maskZero?(t=Ut(t),e.notEqual(t,e.zerosLike(t))):null))}computeOutputShape(t){if(t=Wt(t),null==this.inputLength)return[...t,this.outputDim];const e=g(this.inputLength);if(e.length!==t.length-1)throw new o(`"inputLength" is ${this.inputLength}, but received input shape has shape ${t}`);{let n=0;for(let s=0;s<e.length;++s){const i=e[s],r=t[s+1];if(null!=i&&null!=r&&i!==r)throw new o(`"inputLength" is ${this.inputLength}, but received input shape has shape ${t}`);null==i&&(e[n]=r),n++}}return[t[0],...e,this.outputDim]}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);let s=Ut(t);"int32"!==s.dtype&&(s=et(s,"int32"));const i=ct(this.embeddings.read(),e.reshape(s,[s.size]));return e.reshape(i,Wt(this.computeOutputShape(s.shape)))}))}getConfig(){const t={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:Ot(this.embeddingsInitializer),embeddingsRegularizer:Iu(this.embeddingsRegularizer),activityRegularizer:Iu(this.activityRegularizer),embeddingsConstraint:Bo(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},e=super.getConfig();return Object.assign(t,e),t}}xh.className="Embedding",e.serialization.registerClass(xh);class Nh extends Qt{constructor(t){super(t||{}),this.supportsMasking=!0}mergeFunction(t){throw new l}computeElementwiseOpOutputShape(t,e){if(null==t||null==e)return null;if(t.length<e.length)return this.computeElementwiseOpOutputShape(e,t);if(0===e.length)return t;const n=t.slice(0,t.length-e.length);for(let s=0;s<e.length;++s){const i=t[t.length-e.length+s],r=e[s];if(null==i||null==r||i<0||r<0)n.push(null);else if(1===i)n.push(r);else if(1===r)n.push(i);else{if(i!==r)throw new o("Operands could not be broadcast together with shapes "+JSON.stringify(t)+" "+JSON.stringify(e));n.push(i)}}return n}build(t){if(Array.isArray(t)&&!Array.isArray(t[0])&&(t=[Wt(t)]),t.length<2)throw new o(`A merge layer should be called on an Array of at least 2 inputs. Got ${t.length} input(s).`);let e=[];for(const n of t)null!=n&&null!==n[0]&&e.push(n[0]);if(e=x(e),e.length>1)throw new o(`Can not merge tensors with different batch sizes. Got tensors with shapes: ${JSON.stringify(t)}.`);let n=null==t[0]?null:t[0].slice(1);for(let e=1;e<t.length;++e){const s=null==t[e]?null:t[e].slice(1);n=this.computeElementwiseOpOutputShape(n,s)}const s=t.map((t=>t.length));-1===t.indexOf(null)&&1===x(s).length?this.reshapeRequired=!1:this.reshapeRequired=!0}call(t,n){return e.tidy((()=>{if(this.reshapeRequired){const e=[],n=t.map((t=>t.rank));if(-1===n.indexOf(null)){const s=Y(n);for(let n of t){const t=n.rank;for(let e=0;e<s-t;++e)n=nt(n,1);e.push(n)}return this.mergeFunction(e)}{let n=!1;for(const s of t){const t=s.rank;if(null==t){const t=s.shape,r=t[0],a=t.slice(1).concat([r]);let o=i.reshape(s,[r].concat(J(t.slice(1))));o=i.transpose(o,[1,0]),o=i.reshape(o,a),e.push(o),n=!0}else if(t>1){const r=X(1,t).concat([0]);e.push(i.transpose(s,r)),n=!0}else e.push(s)}let s=this.mergeFunction(e);const r=s.rank;if(n)if(null==r){const t=s.shape,e=t[t.length-1],n=[e].concat(t.slice(0,t.length-1));s=i.reshape(i.transpose(i.reshape(s,[-1,e]),[1,0]),n)}else if(r>1){const t=[r-1].concat(X(0,r-1));s=i.transpose(s,t)}return s}}return this.mergeFunction(t)}))}computeOutputShape(t){let e;e=null==t[0]?null:t[0].slice(1);for(let n=1;n<t.length;++n){const s=null==t[n]?null:t[n].slice(1);e=this.computeElementwiseOpOutputShape(e,s)}let n=[];for(const e of t)null!=e&&null!==e[0]&&n.push(e[0]);return n=x(n),e=1===n.length?n.concat(e):[null].concat(e),e}computeMask(t,e){return i.tidy((()=>{if(null==e)return null;if(!Array.isArray(e))throw new o("`mask` should be an Array");if(!Array.isArray(t))throw new o("`inputs` should be an Array");if(e.length!==t.length)throw new o(`The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths (${t.length} vs ${e.length})`);if(e.every((t=>null==t)))return null;let n=(e=e.map((t=>null==t?t:i.expandDims(t,0))))[0];for(let t=1;t<e.length-1;++t)n=i.logicalAnd(n,e[t]);return n}))}}class Ih extends Nh{constructor(t){super(t)}mergeFunction(t){return e.tidy((()=>{let e=t[0].clone();for(let n=1;n<t.length;++n)e=i.add(e,t[n]);return e}))}}Ih.className="Add",e.serialization.registerClass(Ih);class Ah extends Nh{constructor(t){super(t)}mergeFunction(t){return e.tidy((()=>{let e=t[0].clone();for(let n=1;n<t.length;++n)e=i.mul(e,t[n]);return e}))}}Ah.className="Multiply",e.serialization.registerClass(Ah);class zh extends Nh{constructor(t){super(t)}mergeFunction(t){return e.tidy((()=>{let e=t[0].clone();for(let n=1;n<t.length;++n)e=i.add(e,t[n]);return i.mul(1/t.length,e)}))}}zh.className="Average",e.serialization.registerClass(zh);class Eh extends Nh{constructor(t){super(t)}mergeFunction(t){return e.tidy((()=>{let e=t[0];for(let n=1;n<t.length;++n)e=i.maximum(e,t[n]);return e}))}}Eh.className="Maximum",e.serialization.registerClass(Eh);class Th extends Nh{constructor(t){super(t)}mergeFunction(t){return e.tidy((()=>{let e=t[0];for(let n=1;n<t.length;++n)e=i.minimum(e,t[n]);return e}))}}Th.className="Minimum",e.serialization.registerClass(Th);class Ch extends Nh{constructor(t){super(t),this.DEFAULT_AXIS=-1,null==t&&(t={}),this.axis=null==t.axis?this.DEFAULT_AXIS:t.axis,this.supportsMasking=!0,this.reshapeRequired=!1}build(t){if(!Array.isArray(t)||!Array.isArray(t[0])||1===t.length)throw new o("A `Concatenate` layer should be called on a list of at least 2 inputs");let n=!0;for(const e of t)if(null!=e){n=!1;break}if(n)return;const s=[];for(let n=0;n<t.length;++n){const i=t[n].slice();i.splice(this.axis,1);let r=!1;for(const t of s)if(e.util.arraysEqual(t,i)){r=!0;break}r||s.push(i)}if(s.length>1)throw new o("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(t))}mergeFunction(t){return e.tidy((()=>at(t,this.axis)))}computeOutputShape(t){if(!Array.isArray(t)||!Array.isArray(t[0]))throw new o("A `Concatenate` layer should be called on a list of inputs.");const e=t,n=e[0].slice(),s=this.axis<0?n.length+this.axis:this.axis;for(const t of e.slice(1)){if(null==n[s]||null==t[s]){n[s]=null;break}n[s]+=t[s]}return n}computeMask(t,e){if(null==e)return null;if(!Array.isArray(e))throw new o("`mask` should be an array for Concatenate");if(!Array.isArray(t))throw new o("`inputs` should be an array for Concatenate");if(e.length!==t.length)throw new o(`Mismatch in the length of mask (${e.length}) and the legnth of inputs (${t.length})`);return i.tidy((()=>{let n=!0;if(e.forEach((t=>{null==t||(n=!1)})),n)return null;const s=[];for(let n=0;n<t.length;++n)null==e[n]?s.push(i.cast(i.onesLike(t[n]),"bool")):e[n].rank<t[n].rank?s.push(i.expandDims(e[n],-1)):s.push(e[n]);const r=i.concat(s,this.axis);return i.all(r,-1,!1)}))}getConfig(){const t={axis:this.axis},e=super.getConfig();return Object.assign(t,e),t}}function $h(t,e){for(;t<0;)t+=e;return t}Ch.className="Concatenate",e.serialization.registerClass(Ch);class Fh extends Nh{constructor(t){super(t),this.axes=t.axes,this.normalize=null!=t.normalize&&t.normalize,this.supportsMasking=!0,this.reshapeRequired=!1}build(t){i.util.assert(Array.isArray(t)&&2===t.length&&Array.isArray(t[0])&&Array.isArray(t[1]),(()=>"A `Dot` layer should be called on a list of exactly 2 inputs."));const e=t[0],n=t[1];if(e.length>3||n.length>3)throw new l("Dot layer does not support tensors of 4D or higher rank yet.");const s=this.interpretAxes(e,n);if(e[s[0]]!==n[s[1]])throw new o(`Dimension incompatibility: ${e[s[0]]} !== ${n[s[1]]}`)}mergeFunction(t){if(2!==t.length)throw new o(`A \`Dot\` layer must be called on exactly 2 inputs, but received ${t.length} input(s).`);let e,n=t[0],s=t[1];return e=Array.isArray(this.axes)?this.axes.map(((e,n)=>$h(e,t[n].shape.length))):[$h(this.axes,n.shape.length),$h(this.axes,s.shape.length)],this.normalize&&(n=nl(n,e[0]),s=nl(s,e[1])),function(t,e,n){if(t.shape.length>3||e.shape.length>3)throw new l("batchDot is not implemented for tensors of 4D or higher rank yet");if(i.util.assert(t.shape.length>=2,(()=>`batchDot requires the rank of x to be >= 2, but got ${t.shape.length}`)),i.util.assert(t.shape.length>=2,(()=>`batchDot requires the rank of y to be >= 2, but got ${e.shape.length}`)),"number"==typeof n&&(n=[n,n]),"complex64"===t.dtype||"complex64"===e.dtype)throw new l("batchDot is not implemented for complex64-type Tensors yet.");const s=t.shape.length,r=e.shape.length;null==n&&(n=[s-1,r-2]);const a=n;return i.tidy((()=>{let n,o;if(s>r){n=s-r;const t=[];for(let e=0;e<n;++e)t.push(1);e=i.reshape(e,e.shape.concat(t))}else if(r>s){n=r-s;const e=[];for(let t=0;t<n;++t)e.push(1);t=i.reshape(t,t.shape.concat(e))}else n=0;if(2===t.shape.length&&2===e.shape.length)o=a[0]===a[1]?i.sum(i.mul(t,e),a[0]):i.sum(i.mul(i.transpose(t,[1,0]),e),a[1]);else{const n=a[0]!==t.shape.length-1,s=a[1]===e.shape.length-1;o=i.matMul(t,e,n,s)}if(n>0){let t;t=s>r?s+r-3:s-1;const e=[];for(let s=t;s<t+n;++s)e.push(s);o=i.squeeze(o,e)}return 1===o.shape.length&&(o=i.expandDims(o,1)),o}))}(n,s,e)}interpretAxes(t,e){let n;return n=Array.isArray(this.axes)?this.axes:[$h(this.axes,t.length),$h(this.axes,e.length)],n}computeOutputShape(t){i.util.assert(Array.isArray(t)&&2===t.length&&Array.isArray(t[0])&&Array.isArray(t[1]),(()=>"A `Dot` layer should be called on a list of exactly 2 inputs."));const e=t[0].slice(),n=t[1].slice();if(e.length>3||n.length>3)throw new l("Dot layer does not support tensors of 4D or higher rank yet.");const s=this.interpretAxes(e,n);e.splice(s[0],1),n.splice(s[1],1),n.splice(0,1);const r=e.concat(n);return 1===r.length&&r.push(1),r}computeMask(t,e){return null}getConfig(){const t={axes:this.axes,normalize:this.normalize},e=super.getConfig();return Object.assign(t,e),t}}Fh.className="Dot",e.serialization.registerClass(Fh);class Dh extends Qt{constructor(t){super(t),this.supportsMasking=!0,this.stddev=t.stddev}computeOutputShape(t){return t}getConfig(){const t=super.getConfig(),e={stddev:this.stddev};return Object.assign(e,t),e}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);const s=Ut(t);return mt((()=>e.add(ut(s.shape,0,this.stddev),s)),(()=>s),n.training||!1)}))}}Dh.className="GaussianNoise",e.serialization.registerClass(Dh);class Lh extends Qt{constructor(t){super(t),this.supportsMasking=!0,this.rate=t.rate}computeOutputShape(t){return t}getConfig(){const t=super.getConfig(),e={rate:this.rate};return Object.assign(e,t),e}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n);const s=Ut(t);if(this.rate>0&&this.rate<1){return mt((()=>{const t=Math.sqrt(this.rate/(1-this.rate));return e.mul(s,ut(s.shape,1,t))}),(()=>s),n.training||!1)}return s}))}}Lh.className="GaussianDropout",e.serialization.registerClass(Lh);class _h extends Qt{constructor(t){super(t),this.supportsMasking=!0,this.rate=t.rate,this.noiseShape=t.noiseShape}_getNoiseShape(t){return this.noiseShape||Ut(t).shape}computeOutputShape(t){return t}getConfig(){const t=super.getConfig(),e={rate:this.rate};return Object.assign(e,t),e}call(t,n){return e.tidy((()=>{if(this.rate<1&&this.rate>0){const s=this._getNoiseShape(t),i=()=>{const n=Ut(t),i=-1.7580993408473766;let r=e.greaterEqual(e.randomUniform(s),this.rate);r=et(r,"float32");const a=((1-this.rate)*(1+this.rate*i**2))**-.5,o=-a*i*this.rate,l=e.add(e.mul(n,r),e.mul(e.add(r,-1),i));return e.add(e.mul(l,a),o)};return mt(i,(()=>Ut(t)),n.training||!1)}return t}))}}function Rh(t,e,n,s,r,a=.001){let o;if(2===t.rank)o=i.batchNorm2d(t,e,n,s,r,a);else if(3===t.rank)o=i.batchNorm3d(t,e,n,s,r,a);else{if(4!==t.rank)throw new l(`batchNormalization is not implemented for array of rank ${t.rank} yet`);o=i.batchNorm4d(t,e,n,s,r,a)}return o}function Oh(t,n,s,r,a=.001){return e.util.arraysEqual(r.slice().sort(),X(0,t.rank-1))?function(t,n,s,r,a=.001){return e.tidy((()=>{const e=i.moments(t,r),o=e.mean,l=e.variance;return[Rh(t,o,l,s,n,a),o,l]}))}(t,n,s,r,a):function(t,n,s,r,a=.001){return e.tidy((()=>{const o=i.moments(t,r),l=o.mean,u=o.variance,h=[];for(const e of X(0,t.rank))-1!==r.indexOf(e)?h.push(1):h.push(t.shape[e]);const c=e.reshape(l,h),p=e.reshape(u,h),d=null==n?null:e.reshape(n,h),f=null==s?null:e.reshape(s,h);return[Rh(t,c,p,f,d,a),l,u]}))}(t,n,s,r,a)}_h.className="AlphaDropout",e.serialization.registerClass(_h);class Mh extends Qt{constructor(t){null==t&&(t={}),super(t),this.supportsMasking=!0,this.axis=null==t.axis?-1:t.axis,this.momentum=null==t.momentum?.99:t.momentum,this.epsilon=null==t.epsilon?.001:t.epsilon,this.center=null==t.center||t.center,this.scale=null==t.scale||t.scale,this.betaInitializer=Mt(t.betaInitializer||"zeros"),this.gammaInitializer=Mt(t.gammaInitializer||"ones"),this.movingMeanInitializer=Mt(t.movingMeanInitializer||"zeros"),this.movingVarianceInitializer=Mt(t.movingVarianceInitializer||"ones"),this.betaConstraint=Uo(t.betaConstraint),this.gammaConstraint=Uo(t.gammaConstraint),this.betaRegularizer=zu(t.betaRegularizer),this.gammaRegularizer=zu(t.gammaRegularizer)}build(t){t=Wt(t);const e=this.axis>=0?this.axis:this.axis+t.length,n=t[e];if(null==n)throw new o(`Axis ${e} of input tensor should have a defined dimension but the layer received an input with shape ${JSON.stringify(t)}.`);this.inputSpec=[new Ht({ndim:t.length,axes:{[e]:n}})];const s=[n];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}call(t,n){return e.tidy((()=>{const s=null!=n.training&&n.training,r=Ut(t),a=r.shape,o=a.length,l=X(0,o),u=this.axis>=0?this.axis:this.axis+o;l.splice(u,1);const h=c(1,o);h[u]=a[u];const p=l.slice();p.sort();const d=!e.util.arraysEqual(p,X(0,o).slice(0,o-1));if(!s)return(()=>{if(d){const t=e.reshape(this.movingMean.read(),h),n=e.reshape(this.movingVariance.read(),h),s=this.center?e.reshape(this.beta.read(),h):null,i=this.scale?e.reshape(this.gamma.read(),h):null;return Rh(r,t,n,s,i,this.epsilon)}return Rh(r,this.movingMean.read(),this.movingVariance.read(),null==this.beta?null:this.beta.read(),null==this.gamma?null:this.gamma.read(),this.epsilon)})();const[f,g,m]=Oh(r,this.gamma.read(),this.beta.read(),l,this.epsilon),y=(t,e,n)=>{i.tidy((()=>{const s=1-n,r=t.read(),a=i.mul(i.sub(r,e),s);t.write(i.sub(r,a))}))};return(()=>{y(this.movingMean,g,this.momentum),y(this.movingVariance,m,this.momentum)})(),f}))}getConfig(){const t={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:Ot(this.betaInitializer),gammaInitializer:Ot(this.gammaInitializer),movingMeanInitializer:Ot(this.movingMeanInitializer),movingVarianceInitializer:Ot(this.movingVarianceInitializer),betaRegularizer:Iu(this.betaRegularizer),gammaRegularizer:Iu(this.gammaRegularizer),betaConstraint:Bo(this.betaConstraint),gammaConstraint:Bo(this.gammaConstraint)},e=super.getConfig();return Object.assign(t,e),t}}Mh.className="BatchNormalization",e.serialization.registerClass(Mh);class Bh extends Qt{constructor(t){if(null==t&&(t={}),super(t),this.axis=null==t.axis?-1:t.axis,"number"==typeof this.axis){if(!Number.isInteger(this.axis))throw new Error(`Expected axis to be an integer, but received ${this.axis}`)}else{if(!Array.isArray(this.axis))throw new Error(`Expected axis to be an integer or an array of integers, but received ${JSON.stringify(this.axis)}`);for(const t of this.axis)if(!Number.isInteger(t))throw new Error(`Expected axis to be an array of integers, but received ${JSON.stringify(this.axis)}`)}this.epsilon=null==t.epsilon?.001:t.epsilon,this.center=null==t.center||t.center,this.scale=null==t.scale||t.scale,this.betaInitializer=Mt(t.betaInitializer||"zeros"),this.gammaInitializer=Mt(t.gammaInitializer||"ones"),this.betaRegularizer=zu(t.betaRegularizer),this.gammaRegularizer=zu(t.gammaRegularizer),this.supportsMasking=!0}build(t){const e=(t=Wt(t)).length;"number"==typeof this.axis&&(this.axis=[this.axis]);for(let t=0;t<this.axis.length;++t)this.axis[t]<0&&(this.axis[t]+=e);for(const t of this.axis)if(t<0||t>=e)throw new Error(`Invalid axis: ${t}`);if(this.axis.length!==x(this.axis).length)throw new Error(`Found duplicate axes in: ${this.axis}`);const n=this.axis.map((e=>t[e]));this.scale?this.gamma=this.addWeight("gamma",n,"float32",this.gammaInitializer,this.gammaRegularizer,true):this.gamma=null,this.center?this.beta=this.addWeight("beta",n,"float32",this.betaInitializer,this.betaRegularizer,true):this.beta=null,this.built=!0}call(t,n){const s=Ut(t),r=s.shape,a=r.length;return e.tidy((()=>{let{mean:t,variance:n}=e.moments(s,this.axis,!0);const o=c(1,a);for(const t of this.axis)o[t]=r[t];const l=t=>null!=t&&t.shape.length!==a?i.reshape(t,o):t;let u=this.scale?l(this.gamma.read()):null,h=this.center?l(this.beta.read()):null;const p=[],d=[];for(let t=0;t<a;++t)-1!==this.axis.indexOf(t)?(p.push(r[t]),d.push(1)):(p.push(1),d.push(r[t]));return t=i.tile(t,p),n=i.tile(n,p),null!=u&&(u=i.tile(u,d)),null!=h&&(h=i.tile(h,d)),Rh(s,t,n,h,u,this.epsilon)}))}getConfig(){const t={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:Ot(this.betaInitializer),gammaInitializer:Ot(this.gammaInitializer),betaRegularizer:Iu(this.betaRegularizer),gammaRegularizer:Iu(this.gammaRegularizer)},e=super.getConfig();return Object.assign(t,e),t}}Bh.className="LayerNormalization",e.serialization.registerClass(Bh);class Ph extends Qt{constructor(t){if(null==t&&(t={}),super(t),this.dataFormat=null==t.dataFormat?"channelsLast":t.dataFormat,null==t.padding)this.padding=[[1,1],[1,1]];else if("number"==typeof t.padding)this.padding=[[t.padding,t.padding],[t.padding,t.padding]];else{if(t.padding=t.padding,2!==t.padding.length)throw new o(`ZeroPadding2D expects padding to be a length-2 array, but received a length-${t.padding.length} array.`);let e,n;if("number"==typeof t.padding[0])e=[t.padding[0],t.padding[0]],n=[t.padding[1],t.padding[1]];else{if(t.padding=t.padding,2!==t.padding[0].length)throw new o(`ZeroPadding2D expects height padding to be a length-2 array, but received a length-${t.padding[0].length} array.`);if(e=t.padding[0],2!==t.padding[1].length)throw new o(`ZeroPadding2D expects width padding to be a length-2 array, but received a length-${t.padding[1].length} array.`);n=t.padding[1]}this.padding=[e,n]}this.inputSpec=[new Ht({ndim:4})]}computeOutputShape(t){let e,n;return t=Wt(t),"channelsFirst"===this.dataFormat?(e=null!=t[2]&&t[2]>=0?t[2]+this.padding[0][0]+this.padding[0][1]:null,n=null!=t[3]&&t[3]>=0?t[3]+this.padding[1][0]+this.padding[1][1]:null,[t[0],t[1],e,n]):(e=null!=t[1]&&t[1]>=0?t[1]+this.padding[0][0]+this.padding[0][1]:null,n=null!=t[2]&&t[2]>=0?t[2]+this.padding[1][0]+this.padding[1][1]:null,[t[0],e,n,t[3]])}call(t,n){return e.tidy((()=>{return n=Ut(t),s=this.padding,r=this.dataFormat,e.tidy((()=>{if(4!==n.rank)throw new o(`temporalPadding expects input tensor to be 4-D, but received a ${n.rank}-D tensor.`);if(null==s&&(s=[[1,1],[1,1]]),2!==s.length||2!==s[0].length||2!==s[1].length)throw new o("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(null==r&&(r="channelsLast"),"channelsLast"!==r&&"channelsFirst"!==r)throw new o(`Unknown data format: ${r}. Supported data formats are 'channelsLast' and 'channelsFirst.`);let t;return t="channelsFirst"===r?[[0,0],[0,0],s[0],s[1]]:[[0,0],s[0],s[1],[0,0]],i.pad(n,t)}));var n,s,r}))}getConfig(){const t={padding:this.padding,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}}function Uh(t,n,s,r,a,o){return e.tidy((()=>{let e;P(a),W(o),U(r),null==s&&(s=[1,1]),null==r&&(r="valid"),null==a&&(a="channelsLast"),null==o&&(o="max"),t=Ou(t,a);const l="same"===r?"same":"valid";return e="max"===o?i.maxPool(t,n,s,l):i.avgPool(t,n,s,l),"channelsFirst"===a&&(e=i.transpose(e,[0,3,1,2])),e}))}function Wh(t,n,s,r,a,o){return e.tidy((()=>{let e;P(a),W(o),U(r),null==s&&(s=[1,1,1]),null==r&&(r="valid"),null==a&&(a="channelsLast"),null==o&&(o="max"),t=Mu(t,a);const l="same"===r?"same":"valid";return e="max"===o?i.maxPool3d(t,n,s,l):i.avgPool3d(t,n,s,l),"channelsFirst"===a&&(e=i.transpose(e,[0,4,1,2,3])),e}))}Ph.className="ZeroPadding2D",e.serialization.registerClass(Ph);class jh extends Qt{constructor(t){if(null==t.poolSize&&(t.poolSize=2),super(t),"number"==typeof t.poolSize)this.poolSize=[t.poolSize];else{if(!Array.isArray(t.poolSize)||1!==t.poolSize.length||"number"!=typeof t.poolSize[0])throw new o(`poolSize for 1D convolutional layer must be a number or an Array of a single number, but received ${JSON.stringify(t.poolSize)}`);this.poolSize=t.poolSize}if(z(this.poolSize,"poolSize"),null==t.strides)this.strides=this.poolSize;else if("number"==typeof t.strides)this.strides=[t.strides];else{if(!Array.isArray(t.strides)||1!==t.strides.length||"number"!=typeof t.strides[0])throw new o(`strides for 1D convolutional layer must be a number or an Array of a single number, but received ${JSON.stringify(t.strides)}`);this.strides=t.strides}z(this.strides,"strides"),this.padding=null==t.padding?"valid":t.padding,U(this.padding),this.inputSpec=[new Ht({ndim:3})]}computeOutputShape(t){const e=_u((t=Wt(t))[1],this.poolSize[0],this.padding,this.strides[0]);return[t[0],e,t[2]]}call(t,n){return e.tidy((()=>{this.invokeCallHook(t,n),t=nt(Ut(t),2);const e=this.poolingFunction(Ut(t),[this.poolSize[0],1],[this.strides[0],1],this.padding,"channelsLast");return i.squeeze(e,[2])}))}getConfig(){const t={poolSize:this.poolSize,padding:this.padding,strides:this.strides},e=super.getConfig();return Object.assign(t,e),t}}class qh extends jh{constructor(t){super(t)}poolingFunction(t,e,n,s,i){return P(i),U(s),Uh(t,e,n,s,i,"max")}}qh.className="MaxPooling1D",e.serialization.registerClass(qh);class Vh extends jh{constructor(t){super(t)}poolingFunction(t,e,n,s,i){return P(i),U(s),Uh(t,e,n,s,i,"avg")}}Vh.className="AveragePooling1D",e.serialization.registerClass(Vh);class Kh extends Qt{constructor(t){if(null==t.poolSize&&(t.poolSize=[2,2]),super(t),this.poolSize=Array.isArray(t.poolSize)?t.poolSize:[t.poolSize,t.poolSize],null==t.strides)this.strides=this.poolSize;else if(Array.isArray(t.strides)){if(2!==t.strides.length)throw new o(`If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length ${t.strides.length}.`);this.strides=t.strides}else this.strides=[t.strides,t.strides];z(this.poolSize,"poolSize"),z(this.strides,"strides"),this.padding=null==t.padding?"valid":t.padding,this.dataFormat=null==t.dataFormat?"channelsLast":t.dataFormat,P(this.dataFormat),U(this.padding),this.inputSpec=[new Ht({ndim:4})]}computeOutputShape(t){t=Wt(t);let e="channelsFirst"===this.dataFormat?t[2]:t[1],n="channelsFirst"===this.dataFormat?t[3]:t[2];return e=_u(e,this.poolSize[0],this.padding,this.strides[0]),n=_u(n,this.poolSize[1],this.padding,this.strides[1]),"channelsFirst"===this.dataFormat?[t[0],t[1],e,n]:[t[0],e,n,t[3]]}call(t,n){return e.tidy((()=>(this.invokeCallHook(t,n),this.poolingFunction(Ut(t),this.poolSize,this.strides,this.padding,this.dataFormat))))}getConfig(){const t={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}}class Gh extends Kh{constructor(t){super(t)}poolingFunction(t,e,n,s,i){return P(i),U(s),Uh(t,e,n,s,i,"max")}}Gh.className="MaxPooling2D",e.serialization.registerClass(Gh);class Hh extends Kh{constructor(t){super(t)}poolingFunction(t,e,n,s,i){return P(i),U(s),Uh(t,e,n,s,i,"avg")}}Hh.className="AveragePooling2D",e.serialization.registerClass(Hh);class Jh extends Qt{constructor(t){if(null==t.poolSize&&(t.poolSize=[2,2,2]),super(t),this.poolSize=Array.isArray(t.poolSize)?t.poolSize:[t.poolSize,t.poolSize,t.poolSize],null==t.strides)this.strides=this.poolSize;else if(Array.isArray(t.strides)){if(3!==t.strides.length)throw new o(`If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length ${t.strides.length}.`);this.strides=t.strides}else this.strides=[t.strides,t.strides,t.strides];z(this.poolSize,"poolSize"),z(this.strides,"strides"),this.padding=null==t.padding?"valid":t.padding,this.dataFormat=null==t.dataFormat?"channelsLast":t.dataFormat,P(this.dataFormat),U(this.padding),this.inputSpec=[new Ht({ndim:5})]}computeOutputShape(t){t=Wt(t);let e="channelsFirst"===this.dataFormat?t[2]:t[1],n="channelsFirst"===this.dataFormat?t[3]:t[2],s="channelsFirst"===this.dataFormat?t[4]:t[3];return e=_u(e,this.poolSize[0],this.padding,this.strides[0]),n=_u(n,this.poolSize[1],this.padding,this.strides[1]),s=_u(s,this.poolSize[2],this.padding,this.strides[2]),"channelsFirst"===this.dataFormat?[t[0],t[1],e,n,s]:[t[0],e,n,s,t[4]]}call(t,n){return e.tidy((()=>(this.invokeCallHook(t,n),this.poolingFunction(Ut(t),this.poolSize,this.strides,this.padding,this.dataFormat))))}getConfig(){const t={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}}class Zh extends Jh{constructor(t){super(t)}poolingFunction(t,e,n,s,i){return P(i),U(s),Wh(t,e,n,s,i,"max")}}Zh.className="MaxPooling3D",e.serialization.registerClass(Zh);class Yh extends Jh{constructor(t){super(t)}poolingFunction(t,e,n,s,i){return P(i),U(s),Wh(t,e,n,s,i,"avg")}}Yh.className="AveragePooling3D",e.serialization.registerClass(Yh);class Xh extends Qt{constructor(t){super(t),this.inputSpec=[new Ht({ndim:3})]}computeOutputShape(t){return[t[0],t[2]]}call(t,e){throw new l}}class Qh extends Xh{constructor(t){super(t||{})}call(t,n){return e.tidy((()=>{const e=Ut(t);return i.mean(e,1)}))}}Qh.className="GlobalAveragePooling1D",e.serialization.registerClass(Qh);class tc extends Xh{constructor(t){super(t||{})}call(t,n){return e.tidy((()=>{const e=Ut(t);return i.max(e,1)}))}}tc.className="GlobalMaxPooling1D",e.serialization.registerClass(tc);class ec extends Qt{constructor(t){super(t),this.dataFormat=null==t.dataFormat?"channelsLast":t.dataFormat,P(this.dataFormat),this.inputSpec=[new Ht({ndim:4})]}computeOutputShape(t){return"channelsLast"===this.dataFormat?[t[0],t[3]]:[t[0],t[1]]}call(t,e){throw new l}getConfig(){const t={dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}}class nc extends ec{call(t,n){return e.tidy((()=>{const e=Ut(t);return"channelsLast"===this.dataFormat?i.mean(e,[1,2]):i.mean(e,[2,3])}))}}nc.className="GlobalAveragePooling2D",e.serialization.registerClass(nc);class sc extends ec{call(t,n){return e.tidy((()=>{const e=Ut(t);return"channelsLast"===this.dataFormat?i.max(e,[1,2]):i.max(e,[2,3])}))}}sc.className="GlobalMaxPooling2D",e.serialization.registerClass(sc);class ic extends Qt{constructor(t){super(t),this.layer=t.layer}build(t){this.built=!0}get trainable(){return null!=this.layer&&this.layer.trainable}set trainable(t){null!=this.layer&&(this.layer.trainable=t)}get trainableWeights(){return this.layer.trainableWeights}get nonTrainableWeights(){return this.layer.nonTrainableWeights}get updates(){return this.layer._updates}get losses(){return this.layer.losses}getWeights(){return this.layer.getWeights()}setWeights(t){this.layer.setWeights(t)}getConfig(){const t={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},e=super.getConfig();return Object.assign(t,e),t}setFastWeightInitDuringBuild(t){super.setFastWeightInitDuringBuild(t),null!=this.layer&&this.layer.setFastWeightInitDuringBuild(t)}static fromConfig(t,e,n={}){const s=el(e.layer,n);delete e.layer;const i={layer:s};return Object.assign(i,e),new t(i)}}class rc extends ic{constructor(t){super(t),this.supportsMasking=!0}build(t){if((t=Wt(t)).length<3)throw new o(`TimeDistributed layer expects an input shape >= 3D, but received input shape ${JSON.stringify(t)}`);this.inputSpec=[{shape:t}];const e=[t[0]].concat(t.slice(2));this.layer.built||(this.layer.build(e),this.layer.built=!0),super.build(t)}computeOutputShape(t){const e=[(t=Wt(t))[0]].concat(t.slice(2)),n=this.layer.computeOutputShape(e),s=t[1];return[n[0],s].concat(n.slice(1))}call(t,n){return e.tidy((()=>Qu(((t,e)=>[Ut(this.layer.call(t,n)),[]]),t=Ut(t),[],!1,null,null,!1,!0)[1]))}}rc.className="TimeDistributed",e.serialization.registerClass(rc);class ac extends ic{constructor(t){super(t);const e=t.layer.getConfig(),n={};n.className=t.layer.getClassName(),n.config=e,this.forwardLayer=el(n),e.goBackwards=!0!==e.goBackwards;const s={};var i;if(s.className=t.layer.getClassName(),s.config=e,this.backwardLayer=el(s),this.forwardLayer.name="forward_"+this.forwardLayer.name,this.backwardLayer.name="backward_"+this.backwardLayer.name,this.mergeMode=void 0===t.mergeMode?"concat":t.mergeMode,i=this.mergeMode,I(M,"BidirectionalMergeMode",i),t.weights)throw new l("weights support is not implemented for Bidirectional layer yet.");this._stateful=t.layer.stateful,this.returnSequences=t.layer.returnSequences,this.returnState=t.layer.returnState,this.supportsMasking=!0,this._trainable=!0,this.inputSpec=t.layer.inputSpec,this.numConstants=null}get trainable(){return this._trainable}set trainable(t){this._trainable=t,null!=this.forwardLayer&&(this.forwardLayer.trainable=t),null!=this.backwardLayer&&(this.backwardLayer.trainable=t)}getWeights(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())}setWeights(t){const e=t.length,n=Math.floor(e/2);this.forwardLayer.setWeights(t.slice(0,n)),this.backwardLayer.setWeights(t.slice(n))}computeOutputShape(t){let e,n,s,i=this.forwardLayer.computeOutputShape(t);return Array.isArray(i)&&Array.isArray(i[0])||(i=[i]),this.returnState?(s=i.slice(1),e=i[0]):e=i[0],"concat"===this.mergeMode?(e[e.length-1]*=2,n=[e]):n=null==this.mergeMode?[e,e.slice()]:[e],this.returnState?null==this.mergeMode?n.concat(s).concat(s.slice()):[e].concat(s).concat(s.slice()):f(n)}apply(t,e){let n=null==e?null:e.initialState,s=null==e?null:e.constants;null==e&&(e={});const i=Xu(t,n,s,this.numConstants);if(t=i.inputs,n=i.initialState,s=i.constants,Array.isArray(t)&&(n=t.slice(1),t=t[0]),(null==n||0===n.length)&&null==s)return super.apply(t,e);const r=[],a=[];if(null!=n){const t=n.length;if(t%2>0)throw new o("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");e.initialState=n,r.push(...n);const s=n.map((t=>new Ht({shape:t.shape})));this.forwardLayer.stateSpec=s.slice(0,t/2),this.backwardLayer.stateSpec=s.slice(t/2),a.push(...s)}if(null!=s)throw new l("Support for constants in Bidirectional layers is not implemented yet.");const u=r[0]instanceof Jt;for(const t of r)if(t instanceof Jt!==u)throw new o("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors");if(u){const n=[t].concat(r),s=this.inputSpec.concat(a),i=this.inputSpec;this.inputSpec=s;const o=super.apply(n,e);return this.inputSpec=i,o}return super.apply(t,e)}call(t,n){return e.tidy((()=>{const e=n.initialState;let s,r,a,o;if(null==e)s=this.forwardLayer.call(t,n),r=this.backwardLayer.call(t,n);else{const i=e.slice(0,e.length/2),a=e.slice(e.length/2);s=this.forwardLayer.call(t,Object.assign(n,{initialState:i})),r=this.backwardLayer.call(t,Object.assign(n,{initialState:a}))}return this.returnState&&(Array.isArray(s)&&(a=s.slice(1).concat(r.slice(1))),s=s[0],r=r[0]),this.returnSequences&&(r=i.reverse(r,1)),"concat"===this.mergeMode?o=at([s,r]):"sum"===this.mergeMode?o=i.add(s,r):"ave"===this.mergeMode?o=i.mul(.5,i.add(s,r)):"mul"===this.mergeMode?o=i.mul(s,r):null==this.mergeMode&&(o=[s,r]),this.returnState?null==this.mergeMode?o.concat(a):[o].concat(a):o}))}resetStates(t){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()}build(t){q(this.forwardLayer.name,(()=>{this.forwardLayer.build(t)})),q(this.backwardLayer.name,(()=>{this.backwardLayer.build(t)})),this.built=!0}computeMask(t,e){let n;if(Array.isArray(e)&&(e=e[0]),n=this.returnSequences?null==this.mergeMode?[e,e]:e:null==this.mergeMode?[null,null]:null,this.returnState){const t=this.forwardLayer.states.map((t=>null));return Array.isArray(n)?n.concat(t).concat(t):[n].concat(t).concat(t)}return n}get trainableWeights(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)}get nonTrainableWeights(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)}setFastWeightInitDuringBuild(t){super.setFastWeightInitDuringBuild(t),null!=this.forwardLayer&&this.forwardLayer.setFastWeightInitDuringBuild(t),null!=this.backwardLayer&&this.backwardLayer.setFastWeightInitDuringBuild(t)}getConfig(){const t={mergeMode:this.mergeMode},e=super.getConfig();return Object.assign(t,e),t}static fromConfig(t,e){const n=el(e.layer);if(delete e.layer,null!=e.numConstants)throw new l("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");const s=e;return s.layer=n,new t(s)}}ac.className="Bidirectional",e.serialization.registerClass(ac);class oc extends Qt{constructor(t){super(t),this.scale=t.scale,t.offset?this.offset=t.offset:this.offset=0}getConfig(){const t={scale:this.scale,offset:this.offset},e=super.getConfig();return Object.assign(t,e),t}call(t,n){return e.tidy((()=>("float32"!==(t=Ut(t)).dtype&&(t=et(t,"float32")),e.add(e.mul(t,this.scale),this.offset))))}}oc.className="Rescaling",e.serialization.registerClass(oc);const{resizeBilinear:lc,cropAndResize:uc}=e.image;class hc extends Qt{constructor(t){super(t),this.height=t.height,this.width=t.width}centerCrop(t,n,s,i,r,a,o,l){return e.tidy((()=>{let u,h=!1;const c=[n/a,s/o,(i+n)/a,(r+s)/o],p=[];3===t.rank?(h=!0,u=e.stack([t])):u=t;for(let t=0;t<u.shape[0];t++)p.push(c);const d=e.tensor(p,[p.length,4]),f=e.range(0,p.length,1,"int32"),g=uc(u,d,f,[i,r],"nearest");return et(h?Ut(e.unstack(g)):g,l)}))}upsize(t,n,s,i){return e.tidy((()=>et(lc(t,[n,s]),i)))}call(t,n){return e.tidy((()=>{const e=Ut(t),n=e.dtype,s=e.shape,i=s[s.length-3],r=s[s.length-2];let a=0;i!==this.height&&(a=Math.floor((i-this.height)/2));let o=0;return r!==this.width&&(o=Math.floor((r-this.width)/2),0===o&&(o=1)),a>=0&&o>=0?this.centerCrop(e,a,o,this.height,this.width,i,r,n):this.upsize(t,this.height,this.width,n)}))}getConfig(){const t={height:this.height,width:this.width},e=super.getConfig();return Object.assign(t,e),t}computeOutputShape(t){const e=(t=Wt(t)).length-3,n=t.length-2;return t[e]=this.height,t[n]=this.width,t}}hc.className="CenterCrop",e.serialization.registerClass(hc);class cc extends Qt{constructor(t){super(t),this.numTokens=t.numTokens,t.outputMode?this.outputMode=t.outputMode:this.outputMode="multiHot"}getConfig(){const t={numTokens:this.numTokens,outputMode:this.outputMode},e=super.getConfig();return Object.assign(t,e),t}computeOutputShape(t){return null==(t=Wt(t))?[this.numTokens]:"oneHot"===this.outputMode&&1!==t[t.length-1]?(t.push(this.numTokens),t):(t[t.length-1]=this.numTokens,t)}call(t,n){return e.tidy((()=>{let s;if("int32"!==(t=Ut(t)).dtype&&(t=et(t,"int32")),"undefined"!=typeof n.countWeights){if("count"!==this.outputMode)throw new o(`countWeights is not used when outputMode !== count.\n              Received countWeights=${n.countWeights}`);s=Ut(n.countWeights)}const i=e.max(t),r=e.min(t),a=e.greater(this.numTokens,i).bufferSync().get(0),l=e.greaterEqual(r,0).bufferSync().get(0);if(!a||!l)throw new o(`Input values must be between 0 < values <= numTokens with numTokens=${this.numTokens}`);return function(t,n,s,i){let r=Ut(t);if("int32"!==r.dtype&&(r=et(r,"int32")),"int"===n)return r;const a=r.shape;if(0===r.rank&&(r=e.expandDims(r,-1)),"oneHot"===n&&1!==r.shape[r.shape.length-1]&&(r=e.expandDims(r,-1)),r.rank>2)throw new o(`When outputMode is not int, maximum output rank is 2 Received outputMode ${n} and input shape ${a} which would result in output rank ${r.rank}.`);const l=["multiHot","oneHot"].includes(n),u=r;let h;if(h="undefined"!=typeof i&&"count"===n?e.denseBincount(u,i,s,l):e.denseBincount(u,[],s,l),"tfIdf"!==n)return h;if(i)return e.mul(h,i);throw new o("When outputMode is 'tfIdf', weights must be provided.")}(t,this.outputMode,this.numTokens,s)}))}}cc.className="CategoryEncoding",e.serialization.registerClass(cc);const pc=new Set(["bilinear","nearest"]);class dc extends Qt{constructor(t){if(super(t),this.height=t.height,this.width=t.width,t.interpolation){if(!pc.has(t.interpolation))throw new o(`Invalid interpolation parameter: ${t.interpolation} is not implemented`);this.interpolation=t.interpolation}else this.interpolation="bilinear";this.cropToAspectRatio=Boolean(t.cropToAspectRatio)}computeOutputShape(t){const e=(t=Wt(t))[2];return[this.height,this.width,e]}getConfig(){const t={height:this.height,width:this.width,interpolation:this.interpolation,cropToAspectRatio:this.cropToAspectRatio},e=super.getConfig();return Object.assign(t,e),t}call(t,n){return e.tidy((()=>{const n=[this.height,this.width];if("bilinear"===this.interpolation)return e.image.resizeBilinear(t,n,!this.cropToAspectRatio);if("nearest"===this.interpolation)return e.image.resizeNearestNeighbor(t,n,!this.cropToAspectRatio);throw new Error(`Interpolation is ${this.interpolation} but only ${[...pc]} are supported`)}))}}dc.className="Resizing",e.serialization.registerClass(dc);class fc{constructor(t){this.seed=t}next(){if(void 0!==this.seed)return this.seed++}}fc.className="RandomSeed";class gc extends Qt{constructor(t){super(t),this.randomGenerator=new fc(t.seed)}getConfig(){const t={seed:this.randomGenerator.seed},e=super.getConfig();return Object.assign(t,e),t}}gc.className="BaseRandomLayer";const mc=new Set(["bilinear","nearest"]);class yc extends gc{constructor(t){super(t);const{factor:e,interpolation:n="bilinear"}=t;if(this.factor=e,Array.isArray(this.factor)&&2===this.factor.length)this.widthLower=this.factor[0],this.widthUpper=this.factor[1];else{if(Array.isArray(this.factor)||!(this.factor>0))throw new o(`Invalid factor: ${this.factor}. Must be positive number or tuple of 2 numbers`);this.widthLower=-this.factor,this.widthUpper=this.factor}if(this.widthLower<-1||this.widthUpper<-1)throw new o(`factor must have values larger than -1. Got: ${this.factor}`);if(this.widthUpper<this.widthLower)throw new o(`factor cannot have upper bound less than lower bound.\n        Got upper bound: ${this.widthUpper}.\n        Got lower bound: ${this.widthLower}\n      `);if(n){if(!mc.has(n))throw new o(`Invalid interpolation parameter: ${n} is not implemented`);this.interpolation=n}}getConfig(){const t={factor:this.factor,interpolation:this.interpolation},e=super.getConfig();return Object.assign(t,e),t}computeOutputShape(t){const e=(t=Wt(t))[2];return[this.imgHeight,-1,e]}call(t,n){return e.tidy((()=>{const n=Ut(t);this.imgHeight=n.shape[n.shape.length-3];const s=n.shape[n.shape.length-2];this.widthFactor=e.randomUniform([1],1+this.widthLower,1+this.widthUpper,"float32",this.randomGenerator.next());let i=this.widthFactor.dataSync()[0]*s;i=Math.round(i);const r=[this.imgHeight,i];switch(this.interpolation){case"bilinear":return e.image.resizeBilinear(t,r);case"nearest":return e.image.resizeNearestNeighbor(t,r);default:throw new Error(`Interpolation is ${this.interpolation}\n          but only ${[...mc]} are supported`)}}))}}function bc(t){return new Vh(t)}function wc(t){return new Hh(t)}function kc(t){return new Yh(t)}function vc(t){return new tc(t)}function Sc(t){return new sc(t)}function xc(t){return new qh(t)}function Nc(t){return new Gh(t)}yc.className="RandomWidth",e.serialization.registerClass(yc);var Ic={__proto__:null,Layer:Qt,RNN:th,RNNCell:eh,activation:function(t){return new bh(t)},add:function(t){return new Ih(t)},alphaDropout:function(t){return new _h(t)},average:function(t){return new zh(t)},averagePooling1d:bc,averagePooling2d:wc,averagePooling3d:kc,avgPool1d:function(t){return bc(t)},avgPool2d:function(t){return wc(t)},avgPool3d:function(t){return kc(t)},avgPooling1d:function(t){return bc(t)},avgPooling2d:function(t){return wc(t)},avgPooling3d:function(t){return kc(t)},batchNormalization:function(t){return new Mh(t)},bidirectional:function(t){return new ac(t)},categoryEncoding:function(t){return new cc(t)},centerCrop:function(t){return new hc(t)},concatenate:function(t){return new Ch(t)},conv1d:function(t){return new Hu(t)},conv2d:function(t){return new Wu(t)},conv2dTranspose:function(t){return new qu(t)},conv3d:function(t){return new ju(t)},conv3dTranspose:function(t){return new Vu(t)},convLstm2d:function(t){return new dh(t)},convLstm2dCell:function(t){return new ph(t)},cropping2D:function(t){return new Ju(t)},dense:function(t){return new mh(t)},depthwiseConv2d:function(t){return new Yu(t)},dot:function(t){return new Fh(t)},dropout:function(t){return new fh(t)},elu:function(t){return new $u(t)},embedding:function(t){return new xh(t)},flatten:function(t){return new yh(t)},gaussianDropout:function(t){return new Lh(t)},gaussianNoise:function(t){return new Dh(t)},globalAveragePooling1d:function(t){return new Qh(t)},globalAveragePooling2d:function(t){return new nc(t)},globalMaxPool1d:vc,globalMaxPool2d:Sc,globalMaxPooling1d:vc,globalMaxPooling2d:Sc,gru:function(t){return new rh(t)},gruCell:function(t){return new ih(t)},input:nu,inputLayer:function(t){return new ee(t)},layerNormalization:function(t){return new Bh(t)},leakyReLU:function(t){return new Tu(t)},lstm:function(t){return new oh(t)},lstmCell:function(t){return new ah(t)},masking:function(t){return new Sh(t)},maxPool1d:xc,maxPool2d:Nc,maxPooling1d:xc,maxPooling2d:Nc,maxPooling3d:function(t){return new Zh(t)},maximum:function(t){return new Eh(t)},minimum:function(t){return new Th(t)},multiply:function(t){return new Ah(t)},permute:function(t){return new vh(t)},prelu:function(t){return new Cu(t)},randomWidth:function(t){return new yc(t)},reLU:function(t){return new Eu(t)},repeatVector:function(t){return new wh(t)},rescaling:function(t){return new oc(t)},reshape:function(t){return new kh(t)},resizing:function(t){return new dc(t)},rnn:function(t){return new th(t)},separableConv2d:function(t){return new Gu(t)},simpleRNN:function(t){return new sh(t)},simpleRNNCell:function(t){return new nh(t)},softmax:function(t){return new Du(t)},spatialDropout1d:function(t){return new gh(t)},stackedRNNCells:function(t){return new lh(t)},thresholdedReLU:function(t){return new Fu(t)},timeDistributed:function(t){return new rc(t)},upSampling2d:function(t){return new Zu(t)},zeroPadding2d:function(t){return new Ph(t)}};var Ac={__proto__:null,MAPE:function(t,e){return rl(t,e)},MSE:function(t,e){return sl(t,e)},binaryAccuracy:function(t,e){return pl(t,e)},binaryCrossentropy:function(t,e){return yl(t,e)},categoricalAccuracy:function(t,e){return dl(t,e)},categoricalCrossentropy:function(t,e){return wl(t,e)},cosineProximity:function(t,e){return ul(t,e)},mape:function(t,e){return rl(t,e)},meanAbsoluteError:function(t,e){return il(t,e)},meanAbsolutePercentageError:function(t,e){return rl(t,e)},meanSquaredError:function(t,e){return sl(t,e)},mse:function(t,e){return sl(t,e)},precision:function(t,e){return gl(t,e)},recall:function(t,e){return ml(t,e)},sparseCategoricalAccuracy:function(t,e){return bl(t,e)}},zc={__proto__:null,modelFromJSON:async function(t,n){"modelTopology"in t||(t={modelTopology:t});let s=t.modelTopology;null!=s.model_config&&(s=s.model_config);const i=el(Fl(s),n);if(null!=t.weightsManifest){const n=await e.io.loadWeights(t.weightsManifest,t.pathPrefix,i.weights.map((t=>t.originalName))),s={};for(const t of i.weights)s[t.originalName]=n[t.originalName];i.loadWeights(s),e.dispose(n)}return i}};var Ec={__proto__:null,l1:function(t){return vu(e=t),new xu({l1:null!=e?e.l1:null,l2:0});var e},l1l2:function(t){return new xu(t)},l2:function(t){return vu(e=t),new xu({l2:null!=e?e.l2:null,l1:0});var e}};class Tc extends Go{constructor(){super(...arguments),this.model=null}setModel(t){if(!(t instanceof Ql))throw new Error("model must be a LayersModel, not some other Container");this.model=t}}function Cc(t,e){return t<e}function $c(t,e){return t>e}class Fc extends Tc{constructor(t){if(super(),null==t&&(t={}),t.restoreBestWeights)throw new l("restoreBestWeights = True is not implemented in EarlyStopping yet.");this.monitor=t.monitor||"val_loss",this.minDelta=Math.abs(t.minDelta||0),this.patience=t.patience||0,this.verbose=t.verbose||0,this.mode=t.mode||"auto",this.baseline=t.baseline,-1===["auto","min","max"].indexOf(this.mode)&&(console.warn(`EarlyStopping mode '${this.mode}' is invalid. Falling back to mode 'auto'.`),this.mode="auto"),"min"===this.mode?this.monitorFunc=Cc:"max"===this.mode||-1!==this.monitor.indexOf("acc")?this.monitorFunc=$c:this.monitorFunc=Cc,this.monitorFunc===Cc&&(this.minDelta*=-1)}async onTrainBegin(t){this.wait=0,this.stoppedEpoch=0,null!=this.baseline?this.best=this.baseline:this.best=this.monitorFunc===Cc?1/0:-1/0}async onEpochEnd(t,e){await Vo(e);const n=this.getMonitorValue(e);null!=n&&(this.monitorFunc(n-this.minDelta,this.best)?(this.best=n,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=t,this.model.stopTraining=!0)))}async onTrainEnd(t){this.stoppedEpoch>0&&this.verbose&&console.log(`Epoch ${this.stoppedEpoch}: early stopping.`)}getMonitorValue(t){null==t&&(t={});const e=t[this.monitor];return null==e&&console.warn(`Metric for EarlyStopping ${this.monitor} is not available. Available metrics are: ${Object.keys(t)}`),e}}const Dc={earlyStopping:function(t){return new Fc(t)}};t.Callback=Tc,t.CallbackList=Ho,t.CustomCallback=Yo,t.EarlyStopping=Fc,t.History=Zo,t.InputSpec=Ht,t.LayerVariable=Vt,t.LayersModel=Ql,t.RNN=th,t.Sequential=eu,t.SymbolicTensor=Jt,t.callbacks=Dc,t.constraints=Wo,t.initializers=qo,t.input=nu,t.layers=Ic,t.loadLayersModel=async function(t,n){if(null==n&&(n={}),"string"==typeof t){const s=e.io.getLoadHandlers(t,n);if(0===s.length)s.push(e.io.browserHTTPRequest(t,n));else if(s.length>1)throw new o(`Found more than one (${s.length}) load handlers for URL '${t}'`);t=s[0]}return async function(t,n,s){null==s&&(s={});if(null==t.load)throw new o("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");const i=await t.load();let r=i.modelTopology;null!=r.model_config&&(r=r.model_config);const a=null==s.strict||s.strict,l=null!=i.weightData&&null!=i.weightSpecs&&a,u=el(Fl(r),n,l),h=i.trainingConfig;null!=h&&u.loadTrainingConfig(h);null!=i.userDefinedMetadata&&u.setUserDefinedMetadata(i.userDefinedMetadata);if(null!=i.weightData){if(null==i.weightSpecs)throw new o("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");const{modelWeights:t,optimizerWeights:n}=function(t,n){const s=e.io.decodeWeights(t,n),i={},r=[];return n.forEach((t=>{"optimizer"===t.group?r.push({name:t.name,tensor:s[t.name]}):i[t.name]=s[t.name]})),{modelWeights:i,optimizerWeights:r}}(i.weightData,i.weightSpecs);u.loadWeights(t,a),null!=u.optimizer&&n.length>0&&await u.optimizer.setWeights(n),e.dispose(t),e.dispose(n.map((t=>t.tensor)))}return u}(t,void 0,n)},t.metrics=Ac,t.model=function(t){return new Ql(t)},t.models=zc,t.registerCallbackConstructor=function(t,e){Qo.registerCallbackConstructor(t,e)},t.regularizers=Ec,t.sequential=function(t){return new eu(t)},t.version_layers=Ll}));
//# sourceMappingURL=tf-layers.es2017.min.js.map