MetaSVM Class Reference

Base class of all meta models for SVM training. More...

#include <Svm.h>

Inheritance diagram for MetaSVM:

List of all members.

Public Member Functions
	MetaSVM (SVM *pSVM, unsigned int numberOfHyperParameters)
	Constructor.
	MetaSVM (MultiClassSVM *pSVM, unsigned int numberOfHyperParameters)
	Constructor.
	~MetaSVM ()
	Descructor.
SVM *	getSVM ()
	return the underlying SVM model
MultiClassSVM *	getMultiClassSVM ()
	return the underlying MultiClassSVM model
void	model (const Array< double > &input, Array< double > &output)
	Just calls the underlying SVM.
void	setParameter (unsigned int index, double value)
	overloaded version of Model::setParameter
bool	isFeasible ()
	ensure the kernel is feasible
Protected Attributes
Model *	svm
	pointer to the underlying SVM model
KernelFunction *	kernel
	pointer to the kernel function object
unsigned int	hyperparameters
	number of hyperparameters of the SVM training scheme

---

Detailed Description

Base class of all meta models for SVM training.

The MetaSVM is the base class of all SVM training schemes. It stores the hyperparameters as its model parameters. These usually include the kernel parameters and at least one complexity control parameter.

Author:

T. Glasmachers

Date:

2007

Definition at line 497 of file Svm.h.

---

Constructor & Destructor Documentation

MetaSVM::MetaSVM	(	SVM *	pSVM,
		unsigned int	numberOfHyperParameters
	)

Constructor.

Parameters:

	pSVM	Pointer to the SVM to be optimized.
	numberOfHyperParameters	number of hyperparameters additional to the kernel parameters

Definition at line 932 of file Svm.cpp.

References SVM::getKernel(), Model::getParameter(), Model::getParameterDimension(), hyperparameters, kernel, Model::parameter, and svm.

MetaSVM::MetaSVM	(	MultiClassSVM *	pSVM,
		unsigned int	numberOfHyperParameters
	)

Constructor.

Parameters:

	pSVM	Pointer to the MultiClassSVM to be optimized.
	numberOfHyperParameters	number of hyperparameters additional to the kernel parameters

Definition at line 944 of file Svm.cpp.

References MultiClassSVM::getKernel(), Model::getParameter(), Model::getParameterDimension(), hyperparameters, kernel, Model::parameter, and svm.

MetaSVM::~MetaSVM

(

)

Descructor.

Definition at line 956 of file Svm.cpp.

---

Member Function Documentation

MultiClassSVM* MetaSVM::getMultiClassSVM

(

)

[inline]

return the underlying MultiClassSVM model

Definition at line 523 of file Svm.h.

References svm.

Referenced by SVM_Optimizer::optimize().

SVM* MetaSVM::getSVM

(

)

[inline]

return the underlying SVM model

Definition at line 517 of file Svm.h.

References svm.

Referenced by SpanBound1::bound(), SpanBound1::boundDerivative(), GaussianProcessEvidence::dEvidenceDTheta(), SpanBound1::error(), SpanBound::error(), RadiusMargin::error(), NegativePolarization::error(), LOO::error(), NegativeBKTA::error(), NegativeKTA::error(), GaussianProcessVariance::error(), SpanBound1::errorDerivative(), RadiusMargin::errorDerivative(), NegativePolarization::errorDerivative(), NegativeBKTA::errorDerivative(), NegativeKTA::errorDerivative(), GaussianProcessEvidence::evidence(), C_SVM::modelDerivative(), GaussianProcess::operator()(), SVM_Optimizer::optimize(), C_SVM::PrepareDerivative(), and GaussianProcess::train().

bool MetaSVM::isFeasible

(

)

[virtual]

ensure the kernel is feasible

Reimplemented from Model.

Reimplemented in GaussianProcess, C_SVM, Epsilon_SVM, OneClassSVM, RegularizationNetwork, AllInOneMcSVM, CrammerSingerMcSVM, OVAMcSVM, and OCCMcSVM.

Definition at line 976 of file Svm.cpp.

References Model::isFeasible(), and kernel.

Referenced by OCCMcSVM::isFeasible(), OVAMcSVM::isFeasible(), CrammerSingerMcSVM::isFeasible(), AllInOneMcSVM::isFeasible(), RegularizationNetwork::isFeasible(), Epsilon_SVM::isFeasible(), OneClassSVM::isFeasible(), and C_SVM::isFeasible().

void MetaSVM::model	(	const Array< double > &	input,
		Array< double > &	output
	)			[virtual]

Just calls the underlying SVM.

Implements Model.

Definition at line 961 of file Svm.cpp.

References Model::model(), and svm.

void MetaSVM::setParameter	(	unsigned int	index,
		double	value
	)			[virtual]

overloaded version of Model::setParameter

Reimplemented from Model.

Reimplemented in C_SVM, Epsilon_SVM, and OneClassSVM.

Definition at line 966 of file Svm.cpp.

References hyperparameters, kernel, and Model::setParameter().

Referenced by CrammerSingerMcSVM::set_beta(), OCCMcSVM::set_C(), OVAMcSVM::set_C(), AllInOneMcSVM::set_C(), RegularizationNetwork::set_gamma(), GaussianProcess::setBetaInv(), and GaussianProcess::setSigma().

---

Member Data Documentation

unsigned int MetaSVM::hyperparameters [protected]

number of hyperparameters of the SVM training scheme

Definition at line 545 of file Svm.h.

Referenced by MetaSVM(), Epsilon_SVM::setParameter(), and setParameter().

KernelFunction* MetaSVM::kernel [protected]

pointer to the kernel function object

Definition at line 542 of file Svm.h.

Referenced by isFeasible(), MetaSVM(), C_SVM::modelDerivative(), C_SVM::PrepareDerivative(), and setParameter().

Model* MetaSVM::svm [protected]

pointer to the underlying SVM model

Definition at line 539 of file Svm.h.

Referenced by getMultiClassSVM(), getSVM(), MetaSVM(), model(), C_SVM::modelDerivative(), C_SVM::PrepareDerivative(), and GaussianProcess::train().

---

The documentation for this class was generated from the following files:

• Svm.h
• Svm.cpp