org.nd4j.linalg.lossfunctions

Class LossUtil

java.lang.Object

org.nd4j.linalg.lossfunctions.LossUtil

public class LossUtil
extends Object

Created by Alex on 14/09/2016.

Constructor Summary

Constructors

Constructor and Description
LossUtil()

Method Summary

Modifier and Type	Method and Description
static void	applyMask(INDArray to, INDArray mask)
static INDArray	dLdZsoftmax(INDArray dlda, INDArray a) Deprecated. No longer used
static INDArray	dLdZsoftmaxi(INDArray dlda, INDArray a) Deprecated. No longer used
static INDArray	dLdZsoftmaxPreOut(INDArray dlda, INDArray z) Deprecated. No longer used
static boolean	isPerOutputMasking(INDArray to, INDArray mask)

Methods inherited from class java.lang.Object

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

Constructor Detail

LossUtil

public LossUtil()

Method Detail

dLdZsoftmaxPreOut

@Deprecated
public static INDArray dLdZsoftmaxPreOut(INDArray dlda,
                                                     INDArray z)

Deprecated. No longer used

dLdZsoftmax

@Deprecated
public static INDArray dLdZsoftmax(INDArray dlda,
                                               INDArray a)

Deprecated. No longer used

Calculate dL/dz for softmax activation function, from dL/da and a, where
a: output activations
dL/da: derivative of loss function with respect to the output activations
Note: This version does NOT modify either input arrays. This is less efficient. For a more efficient implementation that modifies the input arrays, use dLdZsoftmaxi(INDArray, INDArray).

Parameters:

dlda - derivative of loss function with respect to the output activations (shape [minibatchSize, nOut])

a - output activations array (shape [minibatchSize, nOut])

dLdZsoftmaxi

@Deprecated
public static INDArray dLdZsoftmaxi(INDArray dlda,
                                                INDArray a)

Deprecated. No longer used

Calculate dL/dz for softmax activation function, from dL/da and a, where
a: output activations
dL/da: derivative of loss function with respect to the output activations
Note: This version WILL modify both input arrays (for efficiency). If this is not acceptable, use dLdZsoftmax(INDArray, INDArray).

Parameters:

dlda - derivative of loss function with respect to the output activations (shape [minibatchSize, nOut])

a - output activations array (shape [minibatchSize, nOut])

isPerOutputMasking

public static boolean isPerOutputMasking(INDArray to,
                                         INDArray mask)

applyMask

public static void applyMask(INDArray to,
                             INDArray mask)