All Downloads are FREE. Search and download functionalities are using the official Maven repository.
Please wait. This can take some minutes ...
Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance.
Project price only 1 $
You can buy this project and download/modify it how often you want.
org.nd4j.linalg.learning.config.AMSGrad Maven / Gradle / Ivy
package org.nd4j.linalg.learning.config;
import lombok.Builder;
import lombok.Data;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.learning.AMSGradUpdater;
import org.nd4j.linalg.learning.GradientUpdater;
import org.nd4j.linalg.schedule.ISchedule;
import org.nd4j.shade.jackson.annotation.JsonProperty;
import java.util.Arrays;
/**
* The AMSGrad updater
* Reference: On the Convergence of Adam and Beyond - https://openreview.net/forum?id=ryQu7f-RZ
*
* @author Alex Black
*/
@Data
@Builder(builderClassName = "Builder")
public class AMSGrad implements IUpdater {
public static final double DEFAULT_AMSGRAD_LEARNING_RATE = 1e-3;
public static final double DEFAULT_AMSGRAD_EPSILON = 1e-8;
public static final double DEFAULT_AMSGRAD_BETA1_MEAN_DECAY = 0.9;
public static final double DEFAULT_AMSGRAD_BETA2_VAR_DECAY = 0.999;
@lombok.Builder.Default private double learningRate = DEFAULT_AMSGRAD_LEARNING_RATE; // learning rate
private ISchedule learningRateSchedule;
@lombok.Builder.Default private double beta1 = DEFAULT_AMSGRAD_BETA1_MEAN_DECAY; // gradient moving avg decay rate
@lombok.Builder.Default private double beta2 = DEFAULT_AMSGRAD_BETA2_VAR_DECAY; // gradient sqrt decay rate
@lombok.Builder.Default private double epsilon = DEFAULT_AMSGRAD_EPSILON;
public AMSGrad() {
this(DEFAULT_AMSGRAD_LEARNING_RATE, DEFAULT_AMSGRAD_BETA1_MEAN_DECAY, DEFAULT_AMSGRAD_BETA2_VAR_DECAY,
DEFAULT_AMSGRAD_EPSILON);
}
public AMSGrad(double learningRate){
this(learningRate, null, DEFAULT_AMSGRAD_BETA1_MEAN_DECAY, DEFAULT_AMSGRAD_BETA2_VAR_DECAY, DEFAULT_AMSGRAD_EPSILON);
}
public AMSGrad(ISchedule learningRateSchedule){
this(Double.NaN, learningRateSchedule, DEFAULT_AMSGRAD_BETA1_MEAN_DECAY, DEFAULT_AMSGRAD_BETA2_VAR_DECAY, DEFAULT_AMSGRAD_EPSILON);
}
public AMSGrad(double learningRate, double beta1, double beta2, double epsilon) {
this(learningRate, null, beta1, beta2, epsilon);
}
private AMSGrad(@JsonProperty("learningRate") double learningRate,
@JsonProperty("learningRateSchedule") ISchedule learningRateSchedule,
@JsonProperty("beta1") double beta1,
@JsonProperty("beta2") double beta2,
@JsonProperty("epsilon") double epsilon){
this.learningRate = learningRate;
this.learningRateSchedule = learningRateSchedule;
this.beta1 = beta1;
this.beta2 = beta2;
this.epsilon = epsilon;
}
@Override
public long stateSize(long numParams) {
return 3 * numParams;
}
@Override
public GradientUpdater instantiate(INDArray viewArray, boolean initializeViewArray) {
AMSGradUpdater u = new AMSGradUpdater(this);
int[] gradientShape = viewArray.shape();
gradientShape = Arrays.copyOf(gradientShape, gradientShape.length);
gradientShape[1] /= 3;
u.setStateViewArray(viewArray, gradientShape, viewArray.ordering(), initializeViewArray);
return u;
}
@Override
public AMSGrad clone() {
return new AMSGrad(learningRate, learningRateSchedule, beta1, beta2, 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(){ }
}
}