@tensorflow/tfjs-core/src/optimizers/adamax_optimizer.ts

Hardware-accelerated JavaScript library for machine intelligence

/**
 * @license
 * Copyright 2018 Google LLC. All Rights Reserved.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 * =============================================================================
 */

import {ENGINE} from '../engine';
import {dispose, tidy} from '../globals';
import {abs} from '../ops/abs';
import {add} from '../ops/add';
import {div} from '../ops/div';
import {maximum} from '../ops/maximum';
import {mul} from '../ops/mul';
import {scalar} from '../ops/scalar';
import {sub} from '../ops/sub';
import {zerosLike} from '../ops/zeros_like';
import {ConfigDict, registerClass, Serializable, SerializableConstructor} from '../serialization';
import {Variable} from '../tensor';
import {NamedTensor, NamedVariableMap} from '../tensor_types';

import {Optimizer, OptimizerVariable} from './optimizer';

export class AdamaxOptimizer extends Optimizer {
  /** @nocollapse */
  static className = 'Adamax';  // Note: Name matters for Python compatbility.
  private accBeta1: Variable;
  private iteration: Variable;

  private accumulatedFirstMoment: OptimizerVariable[] = [];
  private accumulatedWeightedInfNorm: OptimizerVariable[] = [];

  constructor(
      protected learningRate: number, protected beta1: number,
      protected beta2: number, protected epsilon: number = null,
      protected decay = 0.0) {
    super();

    tidy(() => {
      this.iteration = scalar(0).variable();
      this.accBeta1 = scalar(beta1).variable();
    });

    if (epsilon == null) {
      this.epsilon = ENGINE.backend.epsilon();
    }
  }

  applyGradients(variableGradients: NamedVariableMap|NamedTensor[]) {
    const variableNames = Array.isArray(variableGradients) ?
        variableGradients.map(item => item.name) :
        Object.keys(variableGradients);

    tidy(() => {
      const oneMinusAccBeta1 = sub(1, this.accBeta1);
      const lr =
          div(-this.learningRate, add(mul(this.iteration, this.decay), 1));

      variableNames.forEach((name, i) => {
        const value = ENGINE.registeredVariables[name];
        const trainable = false;
        if (this.accumulatedFirstMoment[i] == null) {
          this.accumulatedFirstMoment[i] = {
            originalName: `${name}/m`,
            variable: zerosLike(value).variable(trainable)
          };
        }
        if (this.accumulatedWeightedInfNorm[i] == null) {
          this.accumulatedWeightedInfNorm[i] = {
            originalName: `${name}/v`,
            variable: zerosLike(value).variable(trainable)
          };
        }

        const gradient = Array.isArray(variableGradients) ?
            variableGradients[i].tensor :
            variableGradients[name];
        if (gradient == null) {
          return;
        }

        const firstMoment = this.accumulatedFirstMoment[i].variable;
        const weightedInfNorm = this.accumulatedWeightedInfNorm[i].variable;

        const newFirstMoment =
            add(mul(firstMoment, this.beta1), mul(gradient, 1 - this.beta1));

        const ut0 = mul(weightedInfNorm, this.beta2);
        const ut1 = abs(gradient);

        const newWeightedInfNorm = maximum(ut0, ut1);

        firstMoment.assign(newFirstMoment);
        weightedInfNorm.assign(newWeightedInfNorm);

        const newValue =
            add(mul(div(lr, oneMinusAccBeta1),
                    div(newFirstMoment, add(newWeightedInfNorm, this.epsilon))),
                value);

        value.assign(newValue);
      });

      this.iteration.assign(add(this.iteration, 1));
      this.accBeta1.assign(mul(this.accBeta1, this.beta1));
    });
    this.incrementIterations();
  }

  dispose(): void {
    this.accBeta1.dispose();
    this.iteration.dispose();

    if (this.accumulatedFirstMoment != null) {
      dispose(this.accumulatedFirstMoment.map(v => v.variable));
    }
    if (this.accumulatedWeightedInfNorm != null) {
      dispose(this.accumulatedWeightedInfNorm.map(v => v.variable));
    }
  }

  async getWeights(): Promise<NamedTensor[]> {
    throw new Error('getWeights() is not implemented for Adamax yet.');
  }

  async setWeights(weightValues: NamedTensor[]): Promise<void> {
    throw new Error('setWeights() is not implemented for Adamax yet.');
  }

  getConfig(): ConfigDict {
    return {
      'learningRate': this.learningRate,
      'beta1': this.beta1,
      'beta2': this.beta2,
      'epsilon': this.epsilon,
      'decay': this.decay
    };
  }

  /** @nocollapse */
  static fromConfig<T extends Serializable>(
      cls: SerializableConstructor<T>, config: ConfigDict): T {
    return new cls(
        config['learningRate'], config['beta1'], config['beta2'],
        config['epsilon'], config['decay']);
  }
}
registerClass(AdamaxOptimizer);