org.nd4j.linalg.api.activation

Class SoftMax

java.lang.Object

org.nd4j.linalg.api.activation.BaseActivationFunction

org.nd4j.linalg.api.activation.SoftMax

All Implemented Interfaces:

com.google.common.base.Function<INDArray,INDArray>, Serializable, ActivationFunction

public class SoftMax
extends BaseActivationFunction

Softmax function

Author:

Adam Gibson

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
SoftMax() Initializes softmax with column wise features
SoftMax(boolean rows) Initialize softmax with whether to use row wise or column wise features

Method Summary

Methods

Modifier and Type	Method and Description
INDArray	apply(INDArray input) Returns the result of applying this function to input.
INDArray	applyDerivative(INDArray input) Applies the derivative of this function
static INDArray	softmax(INDArray input, boolean row) Soft max function row_maxes is a row vector (max for each row) row_maxes = rowmaxes(input) diff = exp(input - max) / diff.rowSums()
Class<? extends ElementWiseOp>	transformClazz() The class used for transformation

Methods inherited from class org.nd4j.linalg.api.activation.BaseActivationFunction

equals, toString, type

Methods inherited from class java.lang.Object

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

Constructor Detail

SoftMax

public SoftMax(boolean rows)

Initialize softmax with whether to use row wise or column wise features

Parameters:

rows - whether to use row wise or column wise features for calculation

SoftMax

public SoftMax()

Initializes softmax with column wise features

Method Detail

softmax

public static INDArray softmax(INDArray input,
               boolean row)

Soft max function row_maxes is a row vector (max for each row) row_maxes = rowmaxes(input) diff = exp(input - max) / diff.rowSums()

Parameters:

input - the input for the softmax

row - whether the row maxes should be taken or the column maxes, this is dependent on whether the features are column wise or row wise

Returns:

the softmax output (a probability matrix) scaling each row to between 0 and 1

apply

public INDArray apply(INDArray input)

Description copied from class: BaseActivationFunction

Returns the result of applying this function to input. This method is generally expected, but not absolutely required, to have the following properties:

• Its execution does not cause any observable side effects.
• The computation is consistent with equals; that is, Objects.equal(a, b) implies that Objects.equal(function.apply(a), function.apply(b)).

Specified by:

apply in interface com.google.common.base.Function<INDArray,INDArray>

Overrides:

apply in class BaseActivationFunction

transformClazz

public Class<? extends ElementWiseOp> transformClazz()

The class used for transformation

Returns:

the class used for transformation

applyDerivative

public INDArray applyDerivative(INDArray input)

Description copied from interface: ActivationFunction

Applies the derivative of this function

Parameters:

input - the input to apply it to

Returns:

the derivative of this function applied to the input