Interface ActivationFunction

All Superinterfaces:

Consumer<Tensor>

All Known Implementing Classes:

LeakyRelu, Linear, Relu, Sigmoid, SoftSign, Tanh

public interface ActivationFunction
extends Consumer<Tensor>

Common base interface for all activation functions used in layers. Classes implementing this interface should provide methods for calculating value and first derivative of the activation function. Activation function performs non-linear transformation of its input before its sent to layer output. First derivative of a function shows how fast and in what direction function is changing if its input changes, and it is used by training algorithm. For more see https://en.wikipedia.org/wiki/Activation_function

See Also:

• ActivationType

Method Summary

Modifier and Type	Method	Description
default void	accept(Tensor tensor)
void	apply(Tensor3D tensor, int channel)
void	apply(Tensor4D tensor, int batchIdx)
void	apply(Tensor tensor, int from, int to)
static ActivationFunction	create(ActivationType type)	Creates and returns specified type of activation function.
float	getPrime(float y)	Returns the first derivative of activation function for specified output y
float	getValue(float x)	Returns the value of activation function for specified input x

Methods inherited from interface Consumer

andThen

Method Details

getValue

float getValue(float x)

Returns the value of activation function for specified input x

Parameters:

x - input for activation

Returns:

value of activation function

apply

void apply(Tensor3D tensor,
 int channel)

apply

void apply(Tensor4D tensor,
 int batchIdx)

apply

void apply(Tensor tensor,
 int from,
 int to)

accept

default void accept(Tensor tensor)

Specified by:

accept in interface Consumer<Tensor>

getPrime

float getPrime(float y)

Returns the first derivative of activation function for specified output y

Parameters:

y - output of activation function

Returns:

first derivative of activation function

create

static ActivationFunction create(ActivationType type)

Creates and returns specified type of activation function. A factory method for creating activation functions;

Parameters:

type - type of the activation function

Returns:

returns instance of specified activation function type