org.deeplearning4j.nn.conf.dropout

Interface DropoutHelper

public interface DropoutHelper

A helper interface for native dropout implementations

Author:

Alex Black

Method Summary

Modifier and Type	Method and Description
void	applyDropout(INDArray inputActivations, INDArray resultArray, double dropoutInputRetainProb) Apply the dropout during forward pass
void	backprop(INDArray gradAtOutput, INDArray gradAtInput) Perform backpropagation.
boolean	checkSupported()

Method Detail

checkSupported

boolean checkSupported()

Returns:

Check if this dropout helper is supported in the current environment

applyDropout

void applyDropout(INDArray inputActivations,
                  INDArray resultArray,
                  double dropoutInputRetainProb)

Apply the dropout during forward pass

Parameters:

inputActivations - Input activations (pre dropout)

resultArray - Output activations (post dropout). May be same as (or different to) input array

dropoutInputRetainProb - Probability of retaining an activation

backprop

void backprop(INDArray gradAtOutput,
              INDArray gradAtInput)

Perform backpropagation. Note that the same dropout mask should be used for backprop as was used during the last call to applyDropout(INDArray, INDArray, double)

Parameters:

gradAtOutput - Gradient at output (from perspective of forward pass)

gradAtInput - Result array - gradient at input. May be same as (or different to) gradient at input