DBMatOfflineChol specialization that uses a parallel, iterative incomplete cholesky factorization on a dense matrix. More...

#include <DBMatOfflineDenseIChol.hpp>

Inheritance diagram for sgpp::datadriven::DBMatOfflineDenseIChol:

Public Member Functions
void	choleskyModification (Grid &grid, datadriven::DensityEstimationConfiguration &densityEstimationConfig, size_t newPoints, std::vector< size_t > &deletedPoints, double lambda) override
	Updates offline cholesky factorization based on coarsed (deletedPoints) and refined (newPoints) gridPoints.

DBMatOffline *	clone () const override
	Interface for the clone idiom.

	DBMatOfflineDenseIChol ()

	DBMatOfflineDenseIChol (const std::string &fileName)

void	decomposeMatrix (const RegularizationConfiguration &regularizationConfig, const DensityEstimationConfiguration &densityEstimationConfig) override
	Decomposes the matrix according to the chosen decomposition type.

sgpp::datadriven::MatrixDecompositionType	getDecompositionType () override
	Returns the decomposition type of the DBMatOffline object.

Public Member Functions inherited from sgpp::datadriven::DBMatOfflineChol
void	compute_inverse () override

void	compute_inverse_parallel (std::shared_ptr< BlacsProcessGrid > processGrid, const ParallelConfiguration &parallelConfig) override
	parallel/distributed version of compute_inverse()

	DBMatOfflineChol ()

	DBMatOfflineChol (const std::string &fileName)

void	decomposeMatrixParallel (RegularizationConfiguration &regularizationConfig, DensityEstimationConfiguration &densityEstimationConfig, std::shared_ptr< BlacsProcessGrid > processGrid, const ParallelConfiguration &parallelConfig) override
	The parallel/distributed version of decomposeMatrix(...)

const DataMatrix &	getUnmodifiedR () override
	Get the unmodified (without added lambda) system matrix R.

const DataMatrixDistributed &	getUnmodifiedRDistributed (std::shared_ptr< BlacsProcessGrid > processGrid, const ParallelConfiguration &parallelConfig) override
	Get the unmodified (without added lambda) system matrix R.

bool	isRefineable () override
	Only Offline objects based on Cholesky decomposition, or orthogonal adaptivity can be refined.

void	updateRegularization (double lambda) override
	Modifies the decomposition to update the regularization parameter lambda.

void	updateRegularizationParallel (double lambda, std::shared_ptr< BlacsProcessGrid > processGrid, const ParallelConfiguration &parallelConfig) override
	Modifies the parallel decomposition to update the regularization parameter lambda.

Public Member Functions inherited from sgpp::datadriven::DBMatOfflineGE
void	buildMatrix (Grid *grid, const RegularizationConfiguration &regularizationConfig) override
	Builds the right hand side matrix with identity regularization term.

	DBMatOfflineGE (const std::string &fileName)

Public Member Functions inherited from sgpp::datadriven::DBMatOffline
void	compute_L2_refine_vectors (DataMatrix mat_refine, Grid grid, size_t newPoints)
	computes vectors of L2 products of grid points for refinement

	DBMatOffline (const DBMatOffline &rhs)
	Copy Constructor.

	DBMatOffline (const std::string &fileName)
	Constructor Create offline object from serialized offline object.

	DBMatOffline (DBMatOffline &&rhs)=default
	Default move constructor.

DataMatrixDistributed &	getDecomposedInverseDistributed ()
	Get a reference to the distributed decomposed matrix, analogously to getDecomposedMatrixDistributed()

DataMatrix &	getDecomposedMatrix ()
	Get a reference to the decomposed matrix.

DataMatrixDistributed &	getDecomposedMatrixDistributed ()
	Get a reference to the distributed decomposed matrix.

virtual size_t	getGridSize ()
	Returns the dimensionality of the quadratic lhs matrix (i.e.

DataMatrix &	getInverseMatrix ()
	Get a reference to the inverse matrix.

DataMatrix &	getLhsMatrix_ONLY_FOR_TESTING ()
	Allows access to lhs matrix, which is meant ONLY FOR TESTING.

DBMatOffline &	operator= (const DBMatOffline &rhs)
	Default copy assign operator.

DBMatOffline &	operator= (DBMatOffline &&rhs)=default
	Default move assign operator.

void	printMatrix ()
	Prints the matrix onto standard output.

virtual void	store (const std::string &fileName)
	Serialize the DBMatOffline Object.

virtual void	syncDistributedDecomposition (std::shared_ptr< BlacsProcessGrid > processGrid, const ParallelConfiguration &parallelConfig)
	Synchronizes the decomposed matrix.

virtual void	syncDistributedInverse (std::shared_ptr< BlacsProcessGrid > processGrid, const ParallelConfiguration &parallelConfig)
	Synchronizes the inverse matrix.

virtual	~DBMatOffline ()=default
	Default virtual destructor.

Static Public Member Functions
static void	ichol (const DataMatrix &matrix, DataMatrix &result, size_t sweeps=4, size_t startRow=0)
	perform parlallel incomplete cholesky factorization of a matrix.

Additional Inherited Members
Public Attributes inherited from sgpp::datadriven::DBMatOffline
std::set< std::set< size_t > >	interactions

Protected Member Functions inherited from sgpp::datadriven::DBMatOfflineChol
void	choleskyAddPoint (DataVector &newCol, size_t size)
	Updates the cholesky factor when a new grid point is added (e.g.

void	choleskyPermutation (size_t k, size_t l, size_t job)
	Permutes the rows of the cholesky factor based on permutations of the system matrix (e.g.

Protected Member Functions inherited from sgpp::datadriven::DBMatOfflineGE
	DBMatOfflineGE ()

Protected Member Functions inherited from sgpp::datadriven::DBMatOffline
	DBMatOffline ()

void	parseInter (const std::string &fileName, std::set< std::set< size_t > > &interactions) const
	Read the Interactionsterms from a serialized DBMatOfflibe object.

Protected Attributes inherited from sgpp::datadriven::DBMatOffline
bool	isConstructed

bool	isDecomposed

DataMatrixDistributed	lhsDistributed

DataMatrixDistributed	lhsDistributedInverse

DataMatrix	lhsInverse

DataMatrix	lhsMatrix

Detailed Description

DBMatOfflineChol specialization that uses a parallel, iterative incomplete cholesky factorization on a dense matrix.

The current implementation is a proof of concept.

Constructor & Destructor Documentation

◆ DBMatOfflineDenseIChol() [1/2]

sgpp::datadriven::DBMatOfflineDenseIChol::DBMatOfflineDenseIChol ( )

◆ DBMatOfflineDenseIChol() [2/2]

sgpp::datadriven::DBMatOfflineDenseIChol::DBMatOfflineDenseIChol ( const std::string & fileName )

explicit

Member Function Documentation

◆ choleskyModification()

void sgpp::datadriven::DBMatOfflineDenseIChol::choleskyModification	(	Grid &	grid,
		datadriven::DensityEstimationConfiguration &	densityEstimationConfig,
		size_t	newPoints,
		std::vector< size_t > &	deletedPoints,
		double	lambda
	)

overridevirtual

Updates offline cholesky factorization based on coarsed (deletedPoints) and refined (newPoints) gridPoints.

We ignore coarsening.

Parameters

grid	the underlying grid
densityEstimationConfig	configuration for the density estimation
newPoints	amount of refined points
deletedPoints	list of indices of last coarsed points that are ignored.
lambda	the regularization parameter

Reimplemented from sgpp::datadriven::DBMatOfflineChol.

References sgpp::base::DataMatrix::get(), sgpp::base::DataMatrix::getNrows(), grid(), ichol(), sgpp::datadriven::DensityEstimationConfiguration::iCholSweepsRefine_, lambda, level, sgpp::datadriven::DBMatOffline::lhsMatrix, and sgpp::base::DataMatrix::set().

◆ clone()

DBMatOffline * sgpp::datadriven::DBMatOfflineDenseIChol::clone ( ) const

overridevirtual

Interface for the clone idiom.

Returns: a copy of this very object as a pointer to a new DBMatOffline object which is owned by the caller.

Reimplemented from sgpp::datadriven::DBMatOfflineChol.

◆ decomposeMatrix()

void sgpp::datadriven::DBMatOfflineDenseIChol::decomposeMatrix	(	const RegularizationConfiguration &	regularizationConfig,
		const DensityEstimationConfiguration &	densityEstimationConfig
	)

overridevirtual

Decomposes the matrix according to the chosen decomposition type.

The number of rows of the stored result depends on the decomposition type.

Parameters

regularizationConfig	the regularization configuration
densityEstimationConfig	the density estimation configuration

Reimplemented from sgpp::datadriven::DBMatOfflineChol.

References sgpp::base::DataMatrix::get(), sgpp::base::DataMatrix::getNcols(), sgpp::base::DataMatrix::getNrows(), ichol(), sgpp::datadriven::DensityEstimationConfiguration::iCholSweepsDecompose_, sgpp::datadriven::DBMatOffline::isConstructed, sgpp::datadriven::DBMatOffline::isDecomposed, sgpp::datadriven::DBMatOffline::lhsMatrix, and sgpp::base::DataMatrix::set().

◆ getDecompositionType()

sgpp::datadriven::MatrixDecompositionType sgpp::datadriven::DBMatOfflineDenseIChol::getDecompositionType ( )

overridevirtual

Returns the decomposition type of the DBMatOffline object.

Returns: the type of matrix decomposition

Reimplemented from sgpp::datadriven::DBMatOfflineChol.

References sgpp::datadriven::DenseIchol.

◆ ichol()

void sgpp::datadriven::DBMatOfflineDenseIChol::ichol	(	const DataMatrix &	matrix,
		DataMatrix &	result,
		size_t	sweeps = `4`,
		size_t	startRow = `0`
	)

static

perform parlallel incomplete cholesky factorization of a matrix.

This is an out of place operation.

Parameters

matrix	the matrix to be decomposed
result	data matrix that will hold the decomposed matrix
sweeps	how many iterations of the algorithm are required until the result is good enough?
startRow	on which row to start the decomposition (needed for refinement)

References sgpp::base::DataMatrix::get(), sgpp::base::DataMatrix::getNrows(), sgpp::base::DataMatrix::getSize(), and sgpp::base::DataMatrix::set().

Referenced by choleskyModification(), sgpp::datadriven::DBMatDMSDenseIChol::choleskyUpdateLambda(), and decomposeMatrix().

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

datadriven/src/sgpp/datadriven/algorithm/DBMatOfflineDenseIChol.hpp
datadriven/src/sgpp/datadriven/algorithm/DBMatOfflineDenseIChol.cpp

Public Member Functions

Static Public Member Functions

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ DBMatOfflineDenseIChol() [1/2]

◆ DBMatOfflineDenseIChol() [2/2]

Member Function Documentation

◆ choleskyModification()

◆ clone()

◆ decomposeMatrix()

◆ getDecompositionType()

◆ ichol()