The hinge loss function for the linear SVM objective function. More...

#include <linear_svm_function.hpp>

Public Member Functions
	LinearSVMFunction (const MatType &dataset, const arma::Row< size_t > &labels, const size_t numClasses, const double lambda=0.0001, const double delta=1.0, const bool fitIntercept=false)
	Construct the Linear SVM objective function with given parameters. More...

void	Shuffle ()
	Shuffle the dataset. More...

void	GetGroundTruthMatrix (const arma::Row< size_t > &labels, arma::sp_mat &groundTruth)
	Constructs the ground truth label matrix with the passed labels. More...

double	Evaluate (const arma::mat &parameters)
	Evaluate the hinge loss function for all the datapoints. More...

double	Evaluate (const arma::mat &parameters, const size_t firstId, const size_t batchSize=1)
	Evaluate the hinge loss function on the specified datapoints. More...

template<typename GradType >
void	Gradient (const arma::mat &parameters, GradType &gradient)
	Evaluate the gradient of the hinge loss function following the LinearFunctionType requirements on the Gradient function. More...

template<typename GradType >
void	Gradient (const arma::mat &parameters, const size_t firstId, GradType &gradient, const size_t batchSize=1)
	Evaluate the gradient of the hinge loss function, following the LinearFunctionType requirements on the Gradient function. More...

template<typename GradType >
double	EvaluateWithGradient (const arma::mat &parameters, GradType &gradient) const
	Evaluate the gradient of the hinge loss function, following the LinearFunctionType requirements on the Gradient function followed by evaluation of the hinge loss function on all the datapoints. More...

template<typename GradType >
double	EvaluateWithGradient (const arma::mat &parameters, const size_t firstId, GradType &gradient, const size_t batchSize=1) const
	Evaluate the gradient of the hinge loss function, following the LinearFunctionType requirements on the Gradient function followed by evaluation of the hinge loss function on the specified datapoints. More...

const arma::mat &	InitialPoint () const
	Return the initial point for the optimization.

arma::mat &	InitialPoint ()
	Modify the initial point for the optimization.

const arma::sp_mat &	Dataset () const
	Get the dataset.

arma::sp_mat &	Dataset ()
	Modify the dataset.

double &	Lambda ()
	Sets the regularization parameter.

double	Lambda () const
	Gets the regularization parameter.

bool	FitIntercept () const
	Gets the intercept flag.

size_t	NumFunctions () const
	Return the number of functions.

Static Public Member Functions
static void	InitializeWeights (arma::mat &weights, const size_t featureSize, const size_t numClasses, const bool fitIntercept=false)
	Initialize Linear SVM weights (trainable parameters) with the given parameters. More...

Detailed Description

template<typename MatType = arma::mat>
class mlpack::svm::LinearSVMFunction< MatType >

The hinge loss function for the linear SVM objective function.

This is used by various ensmallen optimizers to train the linear SVM model.

Constructor & Destructor Documentation

◆ LinearSVMFunction()

template<typename MatType >

mlpack::svm::LinearSVMFunction< MatType >::LinearSVMFunction	(	const MatType &	dataset,
		const arma::Row< size_t > &	labels,
		const size_t	numClasses,
		const double	lambda = `0.0001`,
		const double	delta = `1.0`,
		const bool	fitIntercept = `false`
	)

Construct the Linear SVM objective function with given parameters.

Parameters

dataset	Input training data, each column associate with one sample
labels	Labels associated with the feature data.
numClasses	Number of classes for classification.
lambda	L2-regularization constant.
delta	Margin of difference between correct class and other classes.
fitIntercept	Intercept term flag.

Member Function Documentation

◆ Evaluate() [1/2]

template<typename MatType >

double mlpack::svm::LinearSVMFunction< MatType >::Evaluate ( const arma::mat & parameters )

Evaluate the hinge loss function for all the datapoints.

Parameters

parameters The parameters of the SVM.

Returns: The value of the loss function for the entire dataset.

◆ Evaluate() [2/2]

template<typename MatType >

double mlpack::svm::LinearSVMFunction< MatType >::Evaluate	(	const arma::mat &	parameters,
		const size_t	firstId,
		const size_t	batchSize = `1`
	)

Evaluate the hinge loss function on the specified datapoints.

Parameters

parameters	The parameters of the SVM.
firstId	Index of the datapoints to use for function evaluation.
batchSize	Size of batch to process.

Returns: The value of the loss function for the given parameters.

◆ EvaluateWithGradient() [1/2]

template<typename MatType >

template<typename GradType >

double mlpack::svm::LinearSVMFunction< MatType >::EvaluateWithGradient	(	const arma::mat &	parameters,
		GradType &	gradient
	)		const

Evaluate the gradient of the hinge loss function, following the LinearFunctionType requirements on the Gradient function followed by evaluation of the hinge loss function on all the datapoints.

Template Parameters

GradType Type of the gradient matrix.

Parameters

parameters	The parameters of the SVM.
gradient	Linear matrix to output the gradient into.

Returns: The value of the loss function at the given parameters.

◆ EvaluateWithGradient() [2/2]

template<typename MatType >

template<typename GradType >

double mlpack::svm::LinearSVMFunction< MatType >::EvaluateWithGradient	(	const arma::mat &	parameters,
		const size_t	firstId,
		GradType &	gradient,
		const size_t	batchSize = `1`
	)		const

Evaluate the gradient of the hinge loss function, following the LinearFunctionType requirements on the Gradient function followed by evaluation of the hinge loss function on the specified datapoints.

Template Parameters

GradType Type of the gradient matrix.

Parameters

parameters	The parameters of the SVM.
firstId	Index of the datapoint to use for the gradient and function evaluation.
gradient	Linear matrix to output the gradient into.
batchSize	Size of the batch to process.

Returns: The value of the loss function at the given parameters.

◆ GetGroundTruthMatrix()

template<typename MatType >

void mlpack::svm::LinearSVMFunction< MatType >::GetGroundTruthMatrix	(	const arma::Row< size_t > &	labels,
		arma::sp_mat &	groundTruth
	)

Constructs the ground truth label matrix with the passed labels.

This is equivalent to applying the indicator function to the training labels.

Parameters

labels	Labels associated with the training data.
groundTruth	Pointer to arma::mat which stores the computed matrix.

The output is in the form of a matrix, which leads to simpler calculations in the Evaluate() and Gradient() methods.

◆ Gradient() [1/2]

template<typename MatType >

template<typename GradType >

void mlpack::svm::LinearSVMFunction< MatType >::Gradient	(	const arma::mat &	parameters,
		GradType &	gradient
	)

Evaluate the gradient of the hinge loss function following the LinearFunctionType requirements on the Gradient function.

Template Parameters

GradType Type of the gradient matrix.

Parameters

parameters	The parameters of the SVM.
gradient	Linear matrix to output the gradient into.

◆ Gradient() [2/2]

template<typename MatType >

template<typename GradType >

void mlpack::svm::LinearSVMFunction< MatType >::Gradient	(	const arma::mat &	parameters,
		const size_t	firstId,
		GradType &	gradient,
		const size_t	batchSize = `1`
	)

Evaluate the gradient of the hinge loss function, following the LinearFunctionType requirements on the Gradient function.

Template Parameters

GradType Type of the gradient matrix.

Parameters

parameters	The parameters of the SVM.
firstId	Index of the datapoint to use for the gradient evaluation.
gradient	Linear matrix to output the gradient into.
batchSize	Size of the batch to process.

◆ InitializeWeights()

template<typename MatType >

void mlpack::svm::LinearSVMFunction< MatType >::InitializeWeights	(	arma::mat &	weights,
		const size_t	featureSize,
		const size_t	numClasses,
		const bool	fitIntercept = `false`
	)

static

Initialize Linear SVM weights (trainable parameters) with the given parameters.

Initializes parameter weights to random values taken from a scaled standard normal distribution.

Parameters

weights	This will be filled with the initialized model weights.
featureSize	The number of features in the training set.
numClasses	Number of classes for classification.
fitIntercept	If true, an intercept is fitted.

The weights cannot be initialized to zero, as that will lead to each class output being the same.

◆ Shuffle()

template<typename MatType >

void mlpack::svm::LinearSVMFunction< MatType >::Shuffle ( )

Shuffle the dataset.

Shuffle the data.

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

src/mlpack/methods/linear_svm/linear_svm_function.hpp
src/mlpack/methods/linear_svm/linear_svm_function_impl.hpp

Public Member Functions

Static Public Member Functions

Detailed Description

template<typename MatType = arma::mat> class mlpack::svm::LinearSVMFunction< MatType >

Constructor & Destructor Documentation

◆ LinearSVMFunction()

Member Function Documentation

◆ Evaluate() [1/2]

◆ Evaluate() [2/2]

◆ EvaluateWithGradient() [1/2]

◆ EvaluateWithGradient() [2/2]

◆ GetGroundTruthMatrix()

◆ Gradient() [1/2]

◆ Gradient() [2/2]

◆ InitializeWeights()

◆ Shuffle()

template<typename MatType = arma::mat>
class mlpack::svm::LinearSVMFunction< MatType >