Package org.deeplearning4j.optimize.solvers.accumulation.encoding.residual

Class ResidualClippingPostProcessor

java.lang.Object

org.deeplearning4j.optimize.solvers.accumulation.encoding.residual.ResidualClippingPostProcessor

All Implemented Interfaces:

Serializable, Cloneable, ResidualPostProcessor

public class ResidualClippingPostProcessor
extends Object
implements ResidualPostProcessor

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor	Description
ResidualClippingPostProcessor(double thresholdMultipleClipValue, int frequency)

Method Summary

Modifier and Type	Method	Description
ResidualClippingPostProcessor	clone()
void	processResidual(int iteration, int epoch, double lastThreshold, INDArray residualVector)

Methods inherited from class java.lang.Object

equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

ResidualClippingPostProcessor

public ResidualClippingPostProcessor(double thresholdMultipleClipValue,
                                     int frequency)

Parameters:

thresholdMultipleClipValue - The multiple of the current threshold to use for clipping. A value of C means that the residual vector will be clipped to the range [-C*T, C*T] for the current threshold T

frequency - Frequency with which to apply the clipping

Method Detail

processResidual

public void processResidual(int iteration,
                            int epoch,
                            double lastThreshold,
                            INDArray residualVector)

Specified by:

processResidual in interface ResidualPostProcessor

Parameters:

iteration - Current iteration

epoch - Current epoch

lastThreshold - Last threshold that was used

residualVector - The current residual vector. Should be modified in-place

clone

public ResidualClippingPostProcessor clone()

Specified by:

clone in interface ResidualPostProcessor

Overrides:

clone in class Object