The base class for activation functions. Typically, this are sigmoid-shaped functions used to "squash" the output of a network node. These are typically used in conjunction with the GNeuralNet class. More...

#include <GActivation.h>

Inheritance diagram for GClasses::GActivationFunction:

Public Member Functions
	GActivationFunction ()

virtual	~GActivationFunction ()

virtual double	center ()=0
	The center output value. This should return the value of squash(0.0). More...

virtual GActivationFunction *	clone ()=0
	Returns a clone of this object. More...

virtual double	derivative (double x)=0
	The derivative of the activation function. More...

virtual double	derivativeOfNet (double net, double activation)
	This computes the derivative of the net value. (Sometimes, such as with GActivationLogistic, it is more efficient to compute this from the activation value, so both are provided.) More...

virtual double	halfRange ()=0
	The absolute difference between the max (or min) output value and the center. More...

virtual double	identityBias ()=0
	Returns a value to use for the bias when initializing this unit to approximate the identity function. More...

virtual double	identityDiag ()=0
	Returns a value to use for the diagonal elements when initializing this unit to approximate the identity function. More...

virtual double	inverse (double y)=0
	The inverse of the activation function. (This function may throw an exception if the activation function cannot be inverted.) More...

virtual const char *	name () const =0
	Returns the name of this activation function. More...

GDomNode *	serialize (GDom *pDoc) const
	Serialize this object. More...

virtual double	squash (double x)=0
	The activation function. More...

Static Public Member Functions
static GActivationFunction *	deserialize (GDomNode *pNode)
	Deserialize this object. More...

Detailed Description

The base class for activation functions. Typically, this are sigmoid-shaped functions used to "squash" the output of a network node. These are typically used in conjunction with the GNeuralNet class.

Constructor & Destructor Documentation

GClasses::GActivationFunction::GActivationFunction ( )

inline

virtual GClasses::GActivationFunction::~GActivationFunction ( )

inlinevirtual

Member Function Documentation

virtual double GClasses::GActivationFunction::center ( )

pure virtual

The center output value. This should return the value of squash(0.0).

Implemented in GClasses::GActivationSoftPlus2, GClasses::GActivationSoftPlus, GClasses::GActivationRectifiedLinear, GClasses::GActivationLogisticDerivative, GClasses::GActivationSinc, GClasses::GActivationSin, GClasses::GActivationGaussian, GClasses::GActivationBiDir, GClasses::GActivationBend, GClasses::GActivationIdentity, GClasses::GActivationAlgebraic, GClasses::GActivationTanH, GClasses::GActivationArcTan, and GClasses::GActivationLogistic.

virtual GActivationFunction* GClasses::GActivationFunction::clone ( )

pure virtual

Returns a clone of this object.

Implemented in GClasses::GActivationSoftPlus2, GClasses::GActivationSoftPlus, GClasses::GActivationRectifiedLinear, GClasses::GActivationLogisticDerivative, GClasses::GActivationSinc, GClasses::GActivationSin, GClasses::GActivationGaussian, GClasses::GActivationBiDir, GClasses::GActivationBend, GClasses::GActivationIdentity, GClasses::GActivationAlgebraic, GClasses::GActivationTanH, GClasses::GActivationArcTan, and GClasses::GActivationLogistic.

virtual double GClasses::GActivationFunction::derivative ( double x )

pure virtual

The derivative of the activation function.

Implemented in GClasses::GActivationSoftPlus2, GClasses::GActivationSoftPlus, GClasses::GActivationRectifiedLinear, GClasses::GActivationLogisticDerivative, GClasses::GActivationSinc, GClasses::GActivationSin, GClasses::GActivationGaussian, GClasses::GActivationBiDir, GClasses::GActivationBend, GClasses::GActivationIdentity, GClasses::GActivationAlgebraic, GClasses::GActivationTanH, GClasses::GActivationArcTan, and GClasses::GActivationLogistic.

virtual double GClasses::GActivationFunction::derivativeOfNet	(	double	net,
		double	activation
	)

inlinevirtual

This computes the derivative of the net value. (Sometimes, such as with GActivationLogistic, it is more efficient to compute this from the activation value, so both are provided.)

Reimplemented in GClasses::GActivationLogisticDerivative, GClasses::GActivationIdentity, GClasses::GActivationTanH, and GClasses::GActivationLogistic.

static GActivationFunction* GClasses::GActivationFunction::deserialize ( GDomNode * pNode )

static

Deserialize this object.

virtual double GClasses::GActivationFunction::halfRange ( )

pure virtual

The absolute difference between the max (or min) output value and the center.

Implemented in GClasses::GActivationSoftPlus2, GClasses::GActivationSoftPlus, GClasses::GActivationRectifiedLinear, GClasses::GActivationLogisticDerivative, GClasses::GActivationSinc, GClasses::GActivationSin, GClasses::GActivationGaussian, GClasses::GActivationBiDir, GClasses::GActivationBend, GClasses::GActivationIdentity, GClasses::GActivationAlgebraic, GClasses::GActivationTanH, GClasses::GActivationArcTan, and GClasses::GActivationLogistic.