sgpp::datadriven::LearnerSVM Class Reference

LearnerSVM learns the data using support vector machines and sparse grid kernels. More...

#include <LearnerSVM.hpp>

Public Member Functions
double	getAccuracy (sgpp::base::DataMatrix &testDataset, const sgpp::base::DataVector &referenceLabels, const double threshold)
	Computes the classification accuracy on the given dataset. More...
double	getAccuracy (const sgpp::base::DataVector &referenceLabels, const double threshold, const sgpp::base::DataVector &predictedLabels)
	Computes the classification accuracy. More...
double	getError (sgpp::base::DataMatrix &data, sgpp::base::DataVector &labels, std::string errorType)
	Computes specified error type (e.g. More...
void	initialize (size_t budget)
	Initializes the SVM learner. More...
	LearnerSVM (base::RegularGridConfiguration &gridConfig, base::AdaptivityConfiguration &adaptConfig, base::DataMatrix &pTrainData, base::DataVector &pTrainLabels, base::DataMatrix &pTestData, base::DataVector &pTestLabels, base::DataMatrix *pValidData, base::DataVector *pValidLabels)
	Constructor. More...
void	predict (sgpp::base::DataMatrix &testData, sgpp::base::DataVector &predictedLabels)
	Predicts class labels based on the trained model. More...
void	storeResults (sgpp::base::DataMatrix &testDataset)
	Stores classified data, grids and function evaluations to csv files. More...
void	train (size_t maxDataPasses, double lambda, double betaRef, std::string refType, std::string refMonitor, size_t refPeriod, double errorDeclineThreshold, size_t errorDeclineBufferSize, size_t minRefInterval)
	Implements support vector learning with sparse grid kernels. More...
	~LearnerSVM ()
	Destructor. More...

Public Attributes
sgpp::base::DataVector	avgErrors
double	error

Protected Member Functions
std::unique_ptr< base::Grid >	createRegularGrid ()
	Generates a regular sparse grid. More...

Protected Attributes
base::AdaptivityConfiguration	adaptivityConfig
std::unique_ptr< base::Grid >	grid
base::RegularGridConfiguration	gridConfig
std::unique_ptr< PrimalDualSVM >	svm
base::DataMatrix &	testData
base::DataVector &	testLabels
base::DataMatrix &	trainData
base::DataVector &	trainLabels
base::DataMatrix *	validData
base::DataVector *	validLabels

Detailed Description

LearnerSVM learns the data using support vector machines and sparse grid kernels.

As learning algorithm the Pegasos-method is implemented.

Constructor & Destructor Documentation

LearnerSVM()

sgpp::datadriven::LearnerSVM::LearnerSVM	(	base::RegularGridConfiguration &	gridConfig,
		base::AdaptivityConfiguration &	adaptConfig,
		base::DataMatrix &	pTrainData,
		base::DataVector &	pTrainLabels,
		base::DataMatrix &	pTestData,
		base::DataVector &	pTestLabels,
		base::DataMatrix *	pValidData,
		base::DataVector *	pValidLabels
	)

Constructor.

Parameters

gridConfig	The grid configuration
adaptConfig	The refinement configuration
pTrainData	The training dataset
pTrainLabels	The corresponding training labels
pTestData	The test dataset
pTestLabels	The corresponding test labels
pValidData	The validation dataset
pValidLabels	The corresponding validation labels

~LearnerSVM()

sgpp::datadriven::LearnerSVM::~LearnerSVM

(

)

Destructor.

Member Function Documentation

createRegularGrid()

std::unique_ptr< base::Grid > sgpp::datadriven::LearnerSVM::createRegularGrid ( )

protected

Generates a regular sparse grid.

Returns

The created grid

References sgpp::base::Grid::createModLinearGrid(), sgpp::base::GeneralGridConfiguration::dim_, gridConfig, sgpp::base::GeneralGridConfiguration::level_, sgpp::base::ModLinear, and sgpp::base::GeneralGridConfiguration::type_.

Referenced by initialize().

getAccuracy() [1/2]

double sgpp::datadriven::LearnerSVM::getAccuracy	(	sgpp::base::DataMatrix &	testDataset,
		const sgpp::base::DataVector &	referenceLabels,
		const double	threshold
	)

Computes the classification accuracy on the given dataset.

Parameters

testDataset	The data for which class labels should be predicted
referenceLabels	The corresponding actual class labels
threshold	The decision threshold (e.g. for class labels -1, 1 -> threshold = 0)

Returns

The resulting accuracy

References sgpp::base::DataMatrix::getNrows(), and predict().

Referenced by train().

getAccuracy() [2/2]

double sgpp::datadriven::LearnerSVM::getAccuracy	(	const sgpp::base::DataVector &	referenceLabels,
		const double	threshold,
		const sgpp::base::DataVector &	predictedLabels
	)

Computes the classification accuracy.

Parameters

referenceLabels	The actual class labels
threshold	The decision threshold (e.g. for class labels -1, 1 -> threshold = 0)
predictedLabels	The predicted class labels

Returns

The resulting accuracy

References sgpp::base::DataVector::get(), sgpp::base::DataVector::getSize(), and python.statsfileInfo::i.

getError()

double sgpp::datadriven::LearnerSVM::getError	(	sgpp::base::DataMatrix &	data,
		sgpp::base::DataVector &	labels,
		std::string	errorType
	)

Computes specified error type (e.g.

MSE).

Parameters

data	The data points
labels	The corresponding class labels
errorType	The type of the error measurement (MSE or Hinge loss)

Returns

The error estimation

References chess::dim, error, sgpp::base::DataVector::get(), sgpp::base::DataMatrix::getNcols(), sgpp::base::DataMatrix::getNrows(), sgpp::base::DataMatrix::getRow(), grid, python.statsfileInfo::i, sgpp::base::DataVector::set(), sgpp::base::DataVector::setAll(), and svm.

Referenced by train().

initialize()

void sgpp::datadriven::LearnerSVM::initialize

(

size_t

budget

)

Initializes the SVM learner.

Parameters

budget

The max. number of stored support vectors

References createRegularGrid(), sgpp::base::DataMatrix::getNcols(), grid, svm, and trainData.

predict()

void sgpp::datadriven::LearnerSVM::predict	(	sgpp::base::DataMatrix &	testData,
		sgpp::base::DataVector &	predictedLabels
	)

Predicts class labels based on the trained model.

Parameters

testData	The data for which class labels should be predicted
predictedLabels	The predicted class labels

References chess::dim, sgpp::base::DataMatrix::getNcols(), sgpp::base::DataMatrix::getNrows(), sgpp::base::DataMatrix::getRow(), grid, python.statsfileInfo::i, sgpp::base::DataVector::set(), and svm.

Referenced by getAccuracy(), storeResults(), and train().

storeResults()

void sgpp::datadriven::LearnerSVM::storeResults

(

sgpp::base::DataMatrix &

testDataset

)

Stores classified data, grids and function evaluations to csv files.

Parameters

testDataset

Data points for which the model is evaluated

References sgpp::base::DataMatrix::appendRow(), sgpp::base::HashGridStorage::begin(), chess::dim, sgpp::base::HashGridStorage::end(), sgpp::base::DataVector::get(), sgpp::base::HashGridStorage::getCoordinates(), sgpp::base::DataMatrix::getNcols(), sgpp::base::DataMatrix::getNrows(), sgpp::base::DataMatrix::getRow(), sgpp::base::DataVector::getSize(), grid, python.statsfileInfo::i, python.utils.statsfile2gnuplot::j, predict(), sgpp::base::DataVector::set(), and svm.

train()

void sgpp::datadriven::LearnerSVM::train	(	size_t	maxDataPasses,
		double	lambda,
		double	betaRef,
		std::string	refType,
		std::string	refMonitor,
		size_t	refPeriod,
		double	errorDeclineThreshold,
		size_t	errorDeclineBufferSize,
		size_t	minRefInterval
	)

Implements support vector learning with sparse grid kernels.

Parameters

maxDataPasses	The number of passes over the whole training data
lambda	The regularization parameter
betaRef	Weighting factor for grid points; used within combined-measure refinement
refType	The refinement indicator (surplus, zero-crossings or data-based)
refMonitor	The refinement strategy (periodic or convergence-based)
refPeriod	The refinement interval (if periodic refinement is chosen)
errorDeclineThreshold	The convergence threshold (if convergence-based refinement is chosen)
errorDeclineBufferSize	The number of error measurements which are used to check convergence (if convergence-based refinement is chosen)
minRefInterval	The minimum number of data points which have to be processed before next refinement can be scheduled (if convergence-based refinement is chosen)

References adaptivityConfig, sgpp::base::DataVector::append(), avgErrors, chess::dim, error, sgpp::base::ImpurityRefinement::free_refine(), sgpp::base::ForwardSelectorRefinement::free_refine(), sgpp::base::DataVector::get(), getAccuracy(), getError(), sgpp::base::DataMatrix::getNcols(), sgpp::base::DataMatrix::getNrows(), sgpp::base::DataMatrix::getRow(), grid, python.utils.sg_projections::gridStorage, python.statsfileInfo::i, sgpp::base::AdaptivityConfiguration::noPoints_, predict(), sgpp::datadriven::RefinementMonitor::pushToBuffer(), sgpp::datadriven::RefinementMonitor::refinementsNecessary(), svm, testData, testLabels, sgpp::base::AdaptivityConfiguration::threshold_, trainData, trainLabels, validData, and validLabels.

Member Data Documentation

adaptivityConfig

base::AdaptivityConfiguration sgpp::datadriven::LearnerSVM::adaptivityConfig

protected

Referenced by train().

avgErrors

sgpp::base::DataVector sgpp::datadriven::LearnerSVM::avgErrors

Referenced by train().

error

double sgpp::datadriven::LearnerSVM::error

Referenced by getError(), and train().

grid

std::unique_ptr<base::Grid> sgpp::datadriven::LearnerSVM::grid

protected

Referenced by python.uq.learner.Interpolant.Interpolant::doLearningIteration(), python.learner.Classifier.Classifier::evalError(), python.uq.learner.Interpolant.Interpolant::evalError(), python.uq.learner.SimulationLearner.SimulationLearner::getCollocationNodes(), getError(), python.uq.learner.SimulationLearner.SimulationLearner::getGrid(), python.uq.learner.SimulationLearner.SimulationLearner::getLearner(), initialize(), python.uq.learner.Regressor.Regressor::learnData(), python.uq.learner.Regressor.Regressor::learnDataWithFolding(), python.uq.learner.Regressor.Regressor::learnDataWithTest(), predict(), python.learner.Classifier.Classifier::refineGrid(), python.learner.Regressor.Regressor::refineGrid(), python.uq.learner.Regressor.Regressor::refineGrid(), python.uq.learner.SimulationLearner.SimulationLearner::refineGrid(), storeResults(), train(), python.learner.Classifier.Classifier::updateResults(), python.learner.Regressor.Regressor::updateResults(), and python.uq.learner.Regressor.Regressor::updateResults().

gridConfig

base::RegularGridConfiguration sgpp::datadriven::LearnerSVM::gridConfig

protected

Referenced by createRegularGrid().

svm

std::unique_ptr<PrimalDualSVM> sgpp::datadriven::LearnerSVM::svm

protected

Referenced by getError(), initialize(), predict(), storeResults(), and train().

testData

base::DataMatrix& sgpp::datadriven::LearnerSVM::testData

protected

Referenced by train().

testLabels

base::DataVector& sgpp::datadriven::LearnerSVM::testLabels

protected

Referenced by train().

trainData

base::DataMatrix& sgpp::datadriven::LearnerSVM::trainData

protected

Referenced by python.uq.dists.KDEDist.KDEDist::__init__(), initialize(), python.uq.dists.KDEDist.KDEDist::marginalize(), python.uq.dists.KDEDist.KDEDist::marginalizeToDimX(), python.uq.dists.KDEDist.KDEDist::toJson(), and train().

trainLabels

base::DataVector& sgpp::datadriven::LearnerSVM::trainLabels

protected

Referenced by train().

validData

base::DataMatrix* sgpp::datadriven::LearnerSVM::validData

protected

Referenced by train().

validLabels

base::DataVector* sgpp::datadriven::LearnerSVM::validLabels

protected

Referenced by train().

---

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

• datadriven/src/sgpp/datadriven/application/LearnerSVM.hpp
• datadriven/src/sgpp/datadriven/application/LearnerSVM.cpp