@tensorflow/tfjs-core
Version:
Hardware-accelerated JavaScript library for machine intelligence
127 lines • 4.78 kB
JavaScript
/**
* @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 { dispose } from '../globals';
import { variableGrads } from '../gradients';
import { scalar } from '../ops/ops';
import { Serializable } from '../serialization';
/** @doc {heading: 'Training', subheading: 'Classes', namespace: 'train'} */
export class Optimizer extends Serializable {
/**
* Executes `f()` and minimizes the scalar output of `f()` by computing
* gradients of y with respect to the list of trainable variables provided by
* `varList`. If no list is provided, it defaults to all trainable variables.
*
* @param f The function to execute and whose output to minimize.
* @param returnCost Whether to return the scalar cost value produced by
* executing `f()`.
* @param varList An optional list of variables to update. If specified, only
* the trainable variables in varList will be updated by minimize. Defaults to
* all trainable variables.
*
* @doc {heading: 'Training', subheading: 'Optimizers'}
*/
minimize(f, returnCost = false, varList) {
const { value, grads } = this.computeGradients(f, varList);
if (varList != null) {
const gradArray = varList.map(v => ({ name: v.name, tensor: grads[v.name] }));
this.applyGradients(gradArray);
}
else {
this.applyGradients(grads);
}
// Dispose gradients.
dispose(grads);
if (returnCost) {
return value;
}
else {
value.dispose();
return null;
}
}
/**
* The number of iterations that this optimizer instance has been invoked for.
*/
get iterations() {
if (this.iterations_ == null) {
this.iterations_ = 0;
}
return this.iterations_;
}
incrementIterations() {
this.iterations_ = this.iterations + 1;
}
/**
* Executes f() and computes the gradient of the scalar output of f() with
* respect to the list of trainable variables provided by `varList`. If no
* list is provided, it defaults to all trainable variables.
*
* @param f The function to execute and whose output to use for computing
* gradients with respect to variables.
* @param varList An optional list of variables to compute gradients with
* respect to. If specified, only the trainable variables in varList will have
* gradients computed with respect to. Defaults to all trainable variables.
*
* @doc {heading: 'Training', subheading: 'Optimizers'}
*/
computeGradients(f, varList) {
return variableGrads(f, varList);
}
/**
* Dispose the variables (if any) owned by this optimizer instance.
*/
dispose() {
if (this.iterations_ != null) {
dispose(this.iterations_);
}
}
async saveIterations() {
if (this.iterations_ == null) {
this.iterations_ = 0;
}
return {
name: 'iter',
// TODO(cais): Use 'int64' type when available.
tensor: scalar(this.iterations_, 'int32')
};
}
async getWeights() {
throw new Error('getWeights() is not implemented for this optimizer yet.');
}
async setWeights(weightValues) {
throw new Error(`setWeights() is not implemented for this optimizer class ` +
`${this.getClassName()}`);
}
/**
* Extract the first element of the weight values and set it
* as the iterations counter variable of this instance of optimizer.
*
* @param weightValues
* @returns Weight values with the first element consumed and excluded.
*/
async extractIterations(weightValues) {
this.iterations_ = (await weightValues[0].tensor.data())[0];
return weightValues.slice(1);
}
}
Object.defineProperty(Optimizer, Symbol.hasInstance, {
value: (instance) => {
return instance.minimize != null && instance.computeGradients != null &&
instance.applyGradients != null;
}
});
//# sourceMappingURL=optimizer.js.map