All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.nd4j.linalg.learning.config.RmsProp Maven / Gradle / Ivy

There is a newer version: 1.0.0-M2.1
Show newest version
package org.nd4j.linalg.learning.config;

import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Data;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.learning.GradientUpdater;
import org.nd4j.linalg.learning.RmsPropUpdater;
import org.nd4j.linalg.schedule.ISchedule;
import org.nd4j.shade.jackson.annotation.JsonProperty;

/**
 * RMS Prop updates:
 * 

* http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf * http://cs231n.github.io/neural-networks-3/#ada * * @author Adam Gibson */ @Data @Builder(builderClassName = "Builder") public class RmsProp implements IUpdater { public static final double DEFAULT_RMSPROP_LEARNING_RATE = 1e-1; public static final double DEFAULT_RMSPROP_EPSILON = 1e-8; public static final double DEFAULT_RMSPROP_RMSDECAY = 0.95; @lombok.Builder.Default private double learningRate = DEFAULT_RMSPROP_LEARNING_RATE; private ISchedule learningRateSchedule; @lombok.Builder.Default private double rmsDecay = DEFAULT_RMSPROP_RMSDECAY; @lombok.Builder.Default private double epsilon = DEFAULT_RMSPROP_EPSILON; public RmsProp(){ this(DEFAULT_RMSPROP_LEARNING_RATE, null, DEFAULT_RMSPROP_RMSDECAY, DEFAULT_RMSPROP_EPSILON); } public RmsProp(double learningRate){ this(learningRate, null, DEFAULT_RMSPROP_RMSDECAY, DEFAULT_RMSPROP_EPSILON); } public RmsProp(ISchedule learningRateSchedule){ this(Double.NaN, learningRateSchedule, DEFAULT_RMSPROP_RMSDECAY, DEFAULT_RMSPROP_EPSILON); } public RmsProp(double learningRate, double rmsDecay, double epsilon){ this(learningRate, null, rmsDecay, epsilon); } private RmsProp(@JsonProperty("learningRate") double learningRate, @JsonProperty("learningRateSchedule") ISchedule learningRateSchedule, @JsonProperty("rmsDecay") double rmsDecay, @JsonProperty("epsilon") double epsilon){ this.learningRate = learningRate; this.learningRateSchedule = learningRateSchedule; this.rmsDecay = rmsDecay; this.epsilon = epsilon; } @Override public long stateSize(long numParams) { return numParams; } @Override public GradientUpdater instantiate(INDArray viewArray, boolean initializeViewArray) { RmsPropUpdater u = new RmsPropUpdater(this); u.setStateViewArray(viewArray, viewArray.shape(), viewArray.ordering(), initializeViewArray); return u; } @Override public RmsProp clone() { return new RmsProp(learningRate, learningRateSchedule, rmsDecay, epsilon); } @Override public double getLearningRate(int iteration, int epoch){ if(learningRateSchedule != null){ return learningRateSchedule.valueAt(iteration, epoch); } return learningRate; } @Override public boolean hasLearningRate() { return true; } @Override public void setLrAndSchedule(double lr, ISchedule lrSchedule) { this.learningRate = lr; this.learningRateSchedule = lrSchedule; } //Partial builder implementation to give public no-arg constructor public static class Builder { public Builder(){ } } }





© 2015 - 2024 Weber Informatics LLC | Privacy Policy