sgpp::datadriven::LearnerBase Class Reference

Abstract class that implements a regression/classification learner based on spatial adaptive Sparse Grids. More...

#include <LearnerBase.hpp>

Inheritance diagram for sgpp::datadriven::LearnerBase:

Public Member Functions
void	dumpFunction (std::string tFilename, size_t resolution)
	simple dump of sparse grid function into file, e.g. More...
void	dumpGrid (std::string tFilename)
	simple dump of grid points into file, e.g. More...
virtual double	getAccuracy (sgpp::base::DataMatrix &testDataset, const sgpp::base::DataVector &classesReference, const double threshold=0.0)
	compute the accuracy for given testDataset. More...
virtual double	getAccuracy (const sgpp::base::DataVector &classesComputed, const sgpp::base::DataVector &classesReference, const double threshold=0.0)
	compute the accuracy for given testDataset. More...
sgpp::base::DataVector &	getAlpha ()
virtual ClassificatorQuality	getCassificatorQuality (sgpp::base::DataMatrix &testDataset, const sgpp::base::DataVector &classesReference, const double threshold=0.0)
	compute the quality for given testDataset, classification ONLY! test is automatically called in order to determine the regression values of the current learner More...
virtual ClassificatorQuality	getCassificatorQuality (const sgpp::base::DataVector &classesComputed, const sgpp::base::DataVector &classesReference, const double threshold=0.0)
	compute the quality for given testDataset, classification ONLY! More...
sgpp::base::Grid &	getGrid ()
bool	getIsRegression () const
	determines the current mode More...
bool	getIsVerbose () const
	determines the current verbose mode of learner More...
std::vector< std::pair< size_t, double > >	getRefinementExecTimes ()
	LearnerBase (const bool isRegression, const bool isVerbose=true)
	Constructor. More...
	LearnerBase (const LearnerBase &copyMe)
	Copy-Constructor. More...
virtual void	multTranspose (sgpp::base::DataMatrix &dataset, sgpp::base::DataVector &multiplier, sgpp::base::DataVector &result)
virtual void	predict (sgpp::base::DataMatrix &testDataset, sgpp::base::DataVector &classesComputed)
	executes a Regression test for a given dataset and returns the result More...
void	setIsVerbose (const bool isVerbose)
	sets the current verbose mode of learner More...
void	setReuseCoefficients (bool reuseCoefficients)
void	setSolverVerbose (bool solverVerbose)
void	store (std::string tGridFilename, std::string tAlphaFilename)
	store the grid and its current coefficients into files for further usage. More...
virtual LearnerTiming	train (sgpp::base::DataMatrix &trainDataset, sgpp::base::DataVector &classes, const sgpp::base::RegularGridConfiguration &GridConfig, const sgpp::solver::SLESolverConfiguration &SolverConfigRefine, const sgpp::solver::SLESolverConfiguration &SolverConfigFinal, const sgpp::base::AdaptivityConfiguration &AdaptConfig, bool testAccDuringAdapt, const double lambdaRegularization, sgpp::base::DataMatrix *testDataset=nullptr, sgpp::base::DataVector *testClasses=nullptr)
	Learning a dataset with spatially adaptive sparse grids. More...
LearnerTiming	train (sgpp::base::DataMatrix &trainDataset, sgpp::base::DataVector &classes, const sgpp::base::RegularGridConfiguration &GridConfig, const sgpp::solver::SLESolverConfiguration &SolverConfig, const double lambdaRegularization)
	Learning a dataset with regular sparse grids. More...
virtual	~LearnerBase ()
	Destructor. More...

Protected Member Functions
virtual std::unique_ptr< sgpp::datadriven::DMSystemMatrixBase >	createDMSystem (sgpp::base::DataMatrix &trainDataset, double lambda)=0
	abstract method that constructs the corresponding system of linear equations Derived classes MUST overwrite this functions! More...
virtual void	InitializeGrid (const sgpp::base::RegularGridConfiguration &GridConfig)
	Initialize the grid and its coefficients. More...
virtual void	postProcessing (const sgpp::base::DataMatrix &trainDataset, const sgpp::solver::SLESolverType &solver, const size_t numNeededIterations)
	Hook-Method for post-processing after each refinement learning. More...
virtual void	preProcessing ()
	Hook-Method for pre-processing before starting learning. More...

Protected Attributes
std::unique_ptr< sgpp::base::DataVector >	alpha
	the grid's coefficients More...
size_t	currentRefinementStep
	the current refinment step during training More...
double	execTime
	execution time More...
std::vector< std::pair< size_t, double > >	ExecTimeOnStep
double	GByte
	number of transferred Gbytes More...
double	GFlop
	number of executed Floating Point operations More...
std::unique_ptr< sgpp::base::Grid >	grid
	sparse grid object More...
bool	isRegression
	is regression selected More...
bool	isTrained
	is the grid trained More...
bool	isVerbose
	is verbose output enabled More...
bool	reuseCoefficients
	shall the coefficients be reused between refinement steps More...
bool	solverVerbose
	sets the verbose option for the solver More...
double	stepExecTime
	execution time for current refinement to calculate the GFlops at the current timestep only otherwise accumulated GFlops (all refinement steps) are calculated More...
double	stepGByte
	number of transferred Gbytes in the current refinement step More...
double	stepGFlop
	number of executed Floating Point operations in the current refinement step More...

Detailed Description

Abstract class that implements a regression/classification learner based on spatial adaptive Sparse Grids.

Furthermore this class is intended to provide a framework for adavanded regression and classification methods by allowing to override basic routines like train and test.

Constructor & Destructor Documentation

LearnerBase() [1/2]

sgpp::datadriven::LearnerBase::LearnerBase ( const bool  isRegression, const bool  isVerbose = true  )

explicit

Constructor.

Parameters

isRegression	flag for regression
isVerbose	flag for verbose output

LearnerBase() [2/2]

sgpp::datadriven::LearnerBase::LearnerBase

(

const LearnerBase &

copyMe

)

Copy-Constructor.

Parameters

copyMe

LearnerBase that should be duplicated

References alpha, currentRefinementStep, execTime, GByte, GFlop, grid, isRegression, isTrained, isVerbose, reuseCoefficients, solverVerbose, stepExecTime, stepGByte, stepGFlop, and sgpp::base::Grid::unserialize().

~LearnerBase()

sgpp::datadriven::LearnerBase::~LearnerBase ( )

virtual

Destructor.

Member Function Documentation

createDMSystem()

virtual std::unique_ptr<sgpp::datadriven::DMSystemMatrixBase> sgpp::datadriven::LearnerBase::createDMSystem ( sgpp::base::DataMatrix &  trainDataset, double  lambda  )

protectedpure virtual

abstract method that constructs the corresponding system of linear equations Derived classes MUST overwrite this functions!

Parameters

trainDataset	training dataset
lambda	lambda regularization parameter

Implemented in sgpp::datadriven::LearnerLeastSquaresIdentity, and sgpp::datadriven::Learner.

Referenced by train().

dumpFunction()

void sgpp::datadriven::LearnerBase::dumpFunction	(	std::string	tFilename,
		size_t	resolution
	)

simple dump of sparse grid function into file, e.g.

used to plot with gnuplot.

only executed if grid is trained and number of dimensions <= 2.

Parameters

tFilename	filename to store the dump to
resolution	resolution of function plot

References alpha, grid, isTrained, and sgpp::base::GridPrinter::printGrid().

dumpGrid()

void sgpp::datadriven::LearnerBase::dumpGrid

(

std::string

tFilename

)

simple dump of grid points into file, e.g.

used to plot with gnuplot

only executed if grid is trained

Parameters

tFilename

filename to store the dump to

References alpha, grid, isTrained, and sgpp::base::GridPrinter::printSparseGrid().

getAccuracy() [1/2]

double sgpp::datadriven::LearnerBase::getAccuracy ( sgpp::base::DataMatrix &  testDataset, const sgpp::base::DataVector &  classesReference, const double  threshold = 0.0  )

virtual

compute the accuracy for given testDataset.

test is automatically called in order to determine the regression values of the current learner

In case if classification (isRegression == false) this routine returns the learner's accuracy In case of regressions (isRegression == true) this routine returns the learner's MSE

Parameters

testDataset	dataset to be tested
classesReference	reference labels of the test dataset
threshold	threshold used for classification, ignored when performing regressions

Returns

accuracy, percent or MSE, depending on the execution mode

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

Referenced by train().

getAccuracy() [2/2]

double sgpp::datadriven::LearnerBase::getAccuracy ( const sgpp::base::DataVector &  classesComputed, const sgpp::base::DataVector &  classesReference, const double  threshold = 0.0  )

virtual

compute the accuracy for given testDataset.

In case if classification (isRegression == false) this routine returns the learner's accuracy In case of regressions (isRegression == true) this routine returns the learner's MSE

Parameters

classesComputed	regression results of the test dataset
classesReference	reference labels of the test dataset
threshold	threshold used for classification, ignored when performing regressions

Returns

accuracy, percent or MSE, depending on the execution mode

References sgpp::base::DataVector::get(), sgpp::base::DataVector::getSize(), python.statsfileInfo::i, isRegression, sgpp::base::DataVector::sqr(), sgpp::base::DataVector::sub(), sgpp::base::DataVector::sum(), and analyse_erg::tmp.

getAlpha()

sgpp::base::DataVector & sgpp::datadriven::LearnerBase::getAlpha

(

)

References alpha.

getCassificatorQuality() [1/2]

ClassificatorQuality sgpp::datadriven::LearnerBase::getCassificatorQuality ( sgpp::base::DataMatrix &  testDataset, const sgpp::base::DataVector &  classesReference, const double  threshold = 0.0  )

virtual

compute the quality for given testDataset, classification ONLY! test is automatically called in order to determine the regression values of the current learner

Parameters

testDataset	dataset to be tested
classesReference	reference labels of the test dataset
threshold	threshold used for classification, ignored when performing regressions

Returns

quality structure containing tp, tn, fp, fn counts

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

getCassificatorQuality() [2/2]

ClassificatorQuality sgpp::datadriven::LearnerBase::getCassificatorQuality ( const sgpp::base::DataVector &  classesComputed, const sgpp::base::DataVector &  classesReference, const double  threshold = 0.0  )

virtual

compute the quality for given testDataset, classification ONLY!

Parameters

classesComputed	regression results of the test dataset
classesReference	reference labels of the test dataset
threshold	threshold used for classification, ignored when performing regressions

Returns

quality structure containing tp, tn, fp, fn counts

References sgpp::datadriven::ClassificatorQuality::falseNegative_, sgpp::datadriven::ClassificatorQuality::falsePositive_, sgpp::base::DataVector::get(), sgpp::base::DataVector::getSize(), python.statsfileInfo::i, isRegression, sgpp::datadriven::ClassificatorQuality::trueNegative_, and sgpp::datadriven::ClassificatorQuality::truePositive_.

getGrid()

sgpp::base::Grid & sgpp::datadriven::LearnerBase::getGrid

(

)

References grid.

getIsRegression()

bool sgpp::datadriven::LearnerBase::getIsRegression

(

)

const

determines the current mode

Returns

returns whether the current mode is regression or not

References isRegression.

getIsVerbose()

bool sgpp::datadriven::LearnerBase::getIsVerbose

(

)

const

determines the current verbose mode of learner

Returns

returns whether the current learner has verbose output

References isVerbose.

getRefinementExecTimes()

std::vector< std::pair< size_t, double > > sgpp::datadriven::LearnerBase::getRefinementExecTimes

(

)

References ExecTimeOnStep.

InitializeGrid()

void sgpp::datadriven::LearnerBase::InitializeGrid ( const sgpp::base::RegularGridConfiguration &  GridConfig )

protectedvirtual

Initialize the grid and its coefficients.

Parameters

GridConfig

structure which describes the regular start grid

References alpha, sgpp::base::GeneralGridConfiguration::dim_, grid, sgpp::base::GeneralGridConfiguration::level_, sgpp::base::Linear, sgpp::base::LinearBoundary, sgpp::base::ModLinear, and sgpp::base::GeneralGridConfiguration::type_.

Referenced by sgpp::datadriven::LearnerLeastSquaresIdentity::testRegular(), and train().

multTranspose()

void sgpp::datadriven::LearnerBase::multTranspose ( sgpp::base::DataMatrix &  dataset, sgpp::base::DataVector &  multiplier, sgpp::base::DataVector &  result  )

virtual

Reimplemented in sgpp::datadriven::LearnerLeastSquaresIdentity.

References sgpp::op_factory::createOperationMultipleEval(), grid, and sgpp::base::OperationMultipleEval::multTranspose().

Referenced by train().

postProcessing()

void sgpp::datadriven::LearnerBase::postProcessing ( const sgpp::base::DataMatrix &  trainDataset, const sgpp::solver::SLESolverType &  solver, const size_t  numNeededIterations  )

protectedvirtual

Hook-Method for post-processing after each refinement learning.

can be overwritten by derived classes

Parameters

trainDataset	matrix with training data
solver	solver
numNeededIterations	number of required iterations

Reimplemented in sgpp::datadriven::LearnerLeastSquaresIdentity.

References execTime, and isVerbose.

Referenced by train().

predict()

void sgpp::datadriven::LearnerBase::predict ( sgpp::base::DataMatrix &  testDataset, sgpp::base::DataVector &  classesComputed  )

virtual

executes a Regression test for a given dataset and returns the result

Parameters

testDataset	dataset that is evaluated with the current learner
classesComputed	result of the evaluation of the data set

Returns

regression values of testDataset

Reimplemented in sgpp::datadriven::LearnerLeastSquaresIdentity.

References alpha, sgpp::op_factory::createOperationMultipleEval(), sgpp::base::DataMatrix::getNrows(), grid, and sgpp::base::OperationMultipleEval::mult().

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

preProcessing()

void sgpp::datadriven::LearnerBase::preProcessing ( )

protectedvirtual

Hook-Method for pre-processing before starting learning.

can be overwritten by derived classes

Referenced by train().

setIsVerbose()

void sgpp::datadriven::LearnerBase::setIsVerbose

(

const bool

isVerbose

)

sets the current verbose mode of learner

Parameters

isVerbose

the current learner's verbose output

References isVerbose.

setReuseCoefficients()

void sgpp::datadriven::LearnerBase::setReuseCoefficients

(

bool

reuseCoefficients

)

References reuseCoefficients.

setSolverVerbose()

void sgpp::datadriven::LearnerBase::setSolverVerbose

(

bool

solverVerbose

)

References solverVerbose.

store()

void sgpp::datadriven::LearnerBase::store	(	std::string	tGridFilename,
		std::string	tAlphaFilename
	)

store the grid and its current coefficients into files for further usage.

Parameters

tGridFilename	filename of grid file
tAlphaFilename	filename of coefficient file

train() [1/2]

LearnerTiming sgpp::datadriven::LearnerBase::train ( sgpp::base::DataMatrix &  trainDataset, sgpp::base::DataVector &  classes, const sgpp::base::RegularGridConfiguration &  GridConfig, const sgpp::solver::SLESolverConfiguration &  SolverConfigRefine, const sgpp::solver::SLESolverConfiguration &  SolverConfigFinal, const sgpp::base::AdaptivityConfiguration &  AdaptConfig, bool  testAccDuringAdapt, const double  lambdaRegularization, sgpp::base::DataMatrix *  testDataset = nullptr, sgpp::base::DataVector *  testClasses = nullptr  )

virtual

Learning a dataset with spatially adaptive sparse grids.

Parameters

trainDataset	the training dataset
classes	classes corresponding to the training dataset
GridConfig	configuration of the regular start grid
SolverConfigRefine	configuration of the SLE solver during the adaptive refinements of the grid
SolverConfigFinal	configuration of the final SLE solving step on the refined grid
AdaptConfig	configuration of the adaptivity strategy
testAccDuringAdapt	set to true if the training accuracy should be determined in evert refinement step
lambdaRegularization	regularization parameter lambda
testDataset	the test dataset (for accuracy output, nullptr if no output is wished)
testClasses	classes corresponding to the testing dataset (for accuracy output, nullptr if no output is wished)

References alpha, chess::b, sgpp::solver::BiCGSTAB, sgpp::solver::CG, sgpp::base::DataVector::componentwise_mult(), createDMSystem(), currentRefinementStep, sgpp::solver::SLESolverConfiguration::eps_, sgpp::base::AdaptivityConfiguration::errorBasedRefinement, execTime, GByte, sgpp::datadriven::LearnerTiming::GByte_, sgpp::datadriven::DMSystemMatrixBase::generateb(), getAccuracy(), sgpp::base::DataMatrix::getNrows(), sgpp::solver::SGSolver::getNumberIterations(), sgpp::solver::SGSolver::getResiduum(), sgpp::base::DataVector::getSize(), sgpp::datadriven::DMSystemMatrixBase::getTimers(), GFlop, sgpp::datadriven::LearnerTiming::GFlop_, grid, python.statsfileInfo::i, InitializeGrid(), isRegression, isTrained, isVerbose, sgpp::solver::SLESolverConfiguration::maxIterations_, multTranspose(), sgpp::base::AdaptivityConfiguration::noPoints_, sgpp::base::AdaptivityConfiguration::numRefinements_, postProcessing(), predict(), sgpp::datadriven::DMSystemMatrixBase::prepareGrid(), preProcessing(), reuseCoefficients, sgpp::solver::SGSolver::setEpsilon(), sgpp::solver::SGSolver::setMaxIterations(), sgpp::solver::SLESolver::solve(), solverVerbose, sgpp::base::DataVector::sqr(), sgpp::base::SGppStopwatch::start(), stepExecTime, sgpp::base::SGppStopwatch::stop(), sgpp::base::DataVector::sub(), sgpp::base::AdaptivityConfiguration::threshold_, sgpp::datadriven::LearnerTiming::timeComplete_, sgpp::datadriven::LearnerTiming::timeMultComplete_, sgpp::datadriven::LearnerTiming::timeMultCompute_, sgpp::datadriven::LearnerTiming::timeMultTransComplete_, sgpp::datadriven::LearnerTiming::timeMultTransCompute_, sgpp::datadriven::LearnerTiming::timeRegularization_, and sgpp::solver::SLESolverConfiguration::type_.

Referenced by train().

train() [2/2]

LearnerTiming sgpp::datadriven::LearnerBase::train	(	sgpp::base::DataMatrix &	trainDataset,
		sgpp::base::DataVector &	classes,
		const sgpp::base::RegularGridConfiguration &	GridConfig,
		const sgpp::solver::SLESolverConfiguration &	SolverConfig,
		const double	lambdaRegularization
	)

Learning a dataset with regular sparse grids.

Parameters

trainDataset	the training dataset
classes	classes corresponding to the training dataset
GridConfig	configuration of the regular grid
SolverConfig	configuration of the SLE solver
lambdaRegularization	regularization parameter lambda

References sgpp::base::AdaptivityConfiguration::maxLevelType_, sgpp::base::AdaptivityConfiguration::noPoints_, sgpp::base::AdaptivityConfiguration::numRefinements_, sgpp::base::AdaptivityConfiguration::percent_, sgpp::base::AdaptivityConfiguration::threshold_, and train().

Member Data Documentation

alpha

std::unique_ptr<sgpp::base::DataVector> sgpp::datadriven::LearnerBase::alpha

protected

the grid's coefficients

Referenced by dumpFunction(), dumpGrid(), getAlpha(), InitializeGrid(), LearnerBase(), sgpp::datadriven::LearnerLeastSquaresIdentity::predict(), predict(), python.learner.Classifier.Classifier::refineGrid(), sgpp::datadriven::LearnerLeastSquaresIdentity::testRegular(), and train().

currentRefinementStep

size_t sgpp::datadriven::LearnerBase::currentRefinementStep

protected

the current refinment step during training

Referenced by LearnerBase(), and train().

execTime

double sgpp::datadriven::LearnerBase::execTime

protected

execution time

Referenced by LearnerBase(), sgpp::datadriven::LearnerLeastSquaresIdentity::postProcessing(), postProcessing(), sgpp::datadriven::LearnerLeastSquaresIdentity::testRegular(), and train().

ExecTimeOnStep

std::vector<std::pair<size_t, double> > sgpp::datadriven::LearnerBase::ExecTimeOnStep

protected

Referenced by getRefinementExecTimes().

GByte

double sgpp::datadriven::LearnerBase::GByte

protected

number of transferred Gbytes

Referenced by LearnerBase(), sgpp::datadriven::LearnerLeastSquaresIdentity::postProcessing(), and train().

GFlop

double sgpp::datadriven::LearnerBase::GFlop

protected

number of executed Floating Point operations

Referenced by LearnerBase(), sgpp::datadriven::LearnerLeastSquaresIdentity::postProcessing(), and train().

grid

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

protected

sparse grid object

Referenced by sgpp::datadriven::Learner::createDMSystem(), sgpp::datadriven::LearnerLeastSquaresIdentity::createDMSystem(), python.uq.learner.Interpolant.Interpolant::doLearningIteration(), dumpFunction(), dumpGrid(), python.learner.Classifier.Classifier::evalError(), python.uq.learner.Interpolant.Interpolant::evalError(), python.uq.learner.SimulationLearner.SimulationLearner::getCollocationNodes(), python.uq.learner.SimulationLearner.SimulationLearner::getGrid(), getGrid(), python.uq.learner.SimulationLearner.SimulationLearner::getLearner(), InitializeGrid(), python.uq.learner.Regressor.Regressor::learnData(), python.uq.learner.Regressor.Regressor::learnDataWithFolding(), python.uq.learner.Regressor.Regressor::learnDataWithTest(), LearnerBase(), sgpp::datadriven::LearnerLeastSquaresIdentity::multTranspose(), multTranspose(), sgpp::datadriven::LearnerLeastSquaresIdentity::postProcessing(), sgpp::datadriven::LearnerLeastSquaresIdentity::predict(), predict(), python.learner.Classifier.Classifier::refineGrid(), python.learner.Regressor.Regressor::refineGrid(), python.uq.learner.Regressor.Regressor::refineGrid(), python.uq.learner.SimulationLearner.SimulationLearner::refineGrid(), sgpp::datadriven::LearnerLeastSquaresIdentity::testRegular(), train(), python.learner.Classifier.Classifier::updateResults(), python.learner.Regressor.Regressor::updateResults(), and python.uq.learner.Regressor.Regressor::updateResults().

isRegression

bool sgpp::datadriven::LearnerBase::isRegression

protected

is regression selected

Referenced by getAccuracy(), getCassificatorQuality(), getIsRegression(), LearnerBase(), and train().

isTrained

bool sgpp::datadriven::LearnerBase::isTrained

protected

is the grid trained

Referenced by dumpFunction(), dumpGrid(), LearnerBase(), and train().

isVerbose

bool sgpp::datadriven::LearnerBase::isVerbose

protected

is verbose output enabled

Referenced by getIsVerbose(), LearnerBase(), sgpp::datadriven::LearnerLeastSquaresIdentity::postProcessing(), postProcessing(), setIsVerbose(), and train().

reuseCoefficients

bool sgpp::datadriven::LearnerBase::reuseCoefficients

protected

shall the coefficients be reused between refinement steps

Referenced by LearnerBase(), sgpp::datadriven::LearnerLeastSquaresIdentity::postProcessing(), setReuseCoefficients(), and train().

solverVerbose

bool sgpp::datadriven::LearnerBase::solverVerbose

protected

sets the verbose option for the solver

Referenced by LearnerBase(), setSolverVerbose(), and train().

stepExecTime

double sgpp::datadriven::LearnerBase::stepExecTime

protected

execution time for current refinement to calculate the GFlops at the current timestep only otherwise accumulated GFlops (all refinement steps) are calculated

Referenced by LearnerBase(), and train().

stepGByte

double sgpp::datadriven::LearnerBase::stepGByte

protected

number of transferred Gbytes in the current refinement step

Referenced by LearnerBase().

stepGFlop

double sgpp::datadriven::LearnerBase::stepGFlop

protected

number of executed Floating Point operations in the current refinement step

Referenced by LearnerBase().

---

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

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