sgpp::datadriven::DBMatOnlineDEOrthoAdapt Class Reference

#include <DBMatOnlineDEOrthoAdapt.hpp>

Inheritance diagram for sgpp::datadriven::DBMatOnlineDEOrthoAdapt:

Public Member Functions
void	add_new_refine_point (sgpp::base::DataVector &x)
	Adds new DataVector to list of refined points For testing purposes only. More...
	DBMatOnlineDEOrthoAdapt (DBMatOffline &offline, Grid &grid, double lambda, double beta=0.)
	Constructor. More...
sgpp::base::DataMatrix &	getB ()
	Returns the additive component of the sherman-morrison-formula, which yields all the information about the refined points. More...
std::vector< sgpp::base::DataVector > *	getRefinedPointsPointer ()
	Gets pointer to the container of refined points, only for testing purposes. More...
void	sherman_morrison_adapt (size_t newPoints, bool refine, std::vector< size_t > coarsen_indices={})
	Rank-one updates/downdates the system matrix, based on the Sherman-Morrison-formula In the current version of the function, the refinePts already are adapted to the regularization parameter lambda. More...
std::vector< size_t >	updateSystemMatrixDecomposition (DensityEstimationConfiguration &densityEstimationConfig, Grid &grid, size_t numAddedGridPoints, std::list< size_t > deletedGridPointIndices, double lambda) override
Public Member Functions inherited from sgpp::datadriven::DBMatOnlineDE
void	computeDensityFunction (DataVector &alpha, Grid &grid, DensityEstimationConfiguration &densityEstimationConfig, bool do_cv=false)
	Computes the density function again based on the saved b's (only applicable for streaming) More...
void	computeDensityFunction (DataVector &alpha, DataMatrix &m, Grid &grid, DensityEstimationConfiguration &densityEstimationConfig, bool save_b=false, bool do_cv=false, std::list< size_t > *deletedPoints=nullptr, size_t newPoints=0)
	Computes the density function for a certain data matrix. More...
	DBMatOnlineDE (DBMatOffline &offline, Grid &grid, double lambda, double beta=0.)
	Constructor. More...
double	eval (DataVector &alpha, const DataVector &p, Grid &grid, bool force=false)
	Evaluates the density function at a certain point. More...
void	eval (DataVector &alpha, DataMatrix &values, DataVector &results, Grid &grid, bool force=false)
	Evaluates the density function on multiple points. More...
double	getBeta ()
	Returns the current weighting factor. More...
bool	isComputed ()
	Returns if the surplus has already been computed. More...
double	normalize (DataVector &alpha, Grid &grid, size_t samples=1000)
	Normalize the Density. More...
double	normalizeQuadrature (DataVector &alpha, Grid &grid)
	Normalize the Density using Quadrature. More...
void	setBeta (double beta)
	Sets the weighting factor. More...
void	updateRhs (size_t gridSize, std::list< size_t > *deletedPoints)
	Restructures the rhs (b vector) of the system matrix. More...
Public Member Functions inherited from sgpp::datadriven::DBMatOnline
	DBMatOnline (DBMatOffline &o)
	Constructor. More...
	DBMatOnline (const DBMatOnline &rhs)=delete
	DBMatOnline (DBMatOnline &&rhs)=default
DBMatOffline &	getOfflineObject ()
	Returns a reference to the offline object. More...
const DBMatOffline &	getOfflineObject () const
DBMatOnline &	operator= (const DBMatOnline &rhs)=delete
DBMatOnline &	operator= (DBMatOnline &&rhs)=default
void	setLambda (double lambda)
	Changes the weighting factor for the regularization term, if possible (might depend on the kind of decomposition for classification) More...
virtual	~DBMatOnline ()=default
	Destructor. More...

Protected Member Functions
void	solveSLE (DataVector &alpha, DataVector &b, Grid &grid, DensityEstimationConfiguration &densityEstimationConfig, bool do_cv) override
	Solves the system (R + lambda*I) * alpha = b, and obtains alpha The solving is done after offline and online phase and works as follows: (R + lambda*I)^{-1} * b = ((Q*T^{-1}*Q^{t} + B) * b = alpha, where B holds the refine/coarsen information. More...
Protected Member Functions inherited from sgpp::datadriven::DBMatOnlineDE
double	computeL2Error (DataVector &alpha, Grid &grid)
double	resDensity (DataVector &alpha, Grid &grid)

Protected Attributes
sgpp::base::DataMatrix	b_adapt_matrix_
bool	b_is_refined
size_t	current_refine_index
std::vector< sgpp::base::DataVector >	refined_points_
Protected Attributes inherited from sgpp::datadriven::DBMatOnlineDE
double	beta
DataVector	bSave
DataVector	bTotalPoints
bool	functionComputed
double	lambda
double	normFactor
size_t	oDim
DataMatrix *	testMat
DataMatrix *	testMatRes
size_t	totalPoints
Protected Attributes inherited from sgpp::datadriven::DBMatOnline
DBMatOffline &	offlineObject

Constructor & Destructor Documentation

DBMatOnlineDEOrthoAdapt()

sgpp::datadriven::DBMatOnlineDEOrthoAdapt::DBMatOnlineDEOrthoAdapt ( DBMatOffline &  offline, Grid &  grid, double  lambda, double  beta = 0.  )

explicit

Constructor.

Parameters

offline	The offline object we base our evaluations on.
lambda	The regularization strength (TODO(fuchsgruber) remove this)
grid	The underlying grid (TODO(fuchsgruber) do we need this?)
beta	The initial weighting factor

References b_adapt_matrix_, b_is_refined, current_refine_index, sgpp::datadriven::DBMatOffline::getDecompositionType(), sgpp::datadriven::OrthoAdapt, and refined_points_.

Member Function Documentation

add_new_refine_point()

void sgpp::datadriven::DBMatOnlineDEOrthoAdapt::add_new_refine_point ( sgpp::base::DataVector &  x )

inline

Adds new DataVector to list of refined points For testing purposes only.

Parameters

x	The DataVector to add

References refined_points_.

getB()

sgpp::base::DataMatrix& sgpp::datadriven::DBMatOnlineDEOrthoAdapt::getB ( )

inline

Returns the additive component of the sherman-morrison-formula, which yields all the information about the refined points.

References b_adapt_matrix_.

Referenced by sgpp::datadriven::DBMatOnlineDE::computeDensityFunction(), and solveSLE().

getRefinedPointsPointer()

std::vector<sgpp::base::DataVector>* sgpp::datadriven::DBMatOnlineDEOrthoAdapt::getRefinedPointsPointer ( )

inline

Gets pointer to the container of refined points, only for testing purposes.

References python.uq.operations.discretizeProduct::refine(), refined_points_, sherman_morrison_adapt(), and updateSystemMatrixDecomposition().

sherman_morrison_adapt()

void sgpp::datadriven::DBMatOnlineDEOrthoAdapt::sherman_morrison_adapt	(	size_t	newPoints,
		bool	refine,
		std::vector< size_t >	coarsen_indices = {}
	)

Rank-one updates/downdates the system matrix, based on the Sherman-Morrison-formula In the current version of the function, the refinePts already are adapted to the regularization parameter lambda.

Parameters

newPoints	number of refined points
refine	decides: true for refine, false for coarsen
coarsen_indices	the indices of points to coarsen

References b_adapt_matrix_, b_is_refined, current_refine_index, sgpp::base::DataVector::get(), sgpp::base::DataMatrix::get(), sgpp::base::DataMatrix::getColumn(), sgpp::base::HashGridStorage::getDimension(), sgpp::datadriven::DBMatOffline::getGridSize(), sgpp::base::HashGridStorage::getLevelIndexArraysForEval(), sgpp::base::DataMatrix::getNcols(), sgpp::base::DataMatrix::getNrows(), sgpp::base::DataMatrix::getPointer(), sgpp::datadriven::DBMatOfflineOrthoAdapt::getQ(), sgpp::base::DataMatrix::getRow(), sgpp::base::HashGridStorage::getSize(), sgpp::base::Grid::getStorage(), sgpp::datadriven::DBMatOfflineOrthoAdapt::getTinv(), grid(), python.statsfileInfo::i, python.utils.statsfile2gnuplot::j, level, sgpp::datadriven::DBMatOnline::offlineObject, refined_points_, sgpp::base::DataMatrix::resizeQuadratic(), sgpp::base::DataMatrix::set(), sgpp::base::DataMatrix::setColumn(), and sgpp::base::DataMatrix::setRow().

Referenced by getRefinedPointsPointer(), and updateSystemMatrixDecomposition().

solveSLE()

void sgpp::datadriven::DBMatOnlineDEOrthoAdapt::solveSLE ( DataVector &  alpha, DataVector &  b, Grid &  grid, DensityEstimationConfiguration &  densityEstimationConfig, bool  do_cv  )

overrideprotectedvirtual

Solves the system (R + lambda*I) * alpha = b, and obtains alpha The solving is done after offline and online phase and works as follows: (R + lambda*I)^{-1} * b = ((Q*T^{-1}*Q^{t} + B) * b = alpha, where B holds the refine/coarsen information.

Parameters

alpha	the datavetor where surplusses will be stored
b	The right hand side of the system
grid	the underlying grid
densityEstimationConfig	configuration for the density estimation
do_cv	Specifies, if cross-validation should be done (todo: currently not implemented)

Implements sgpp::datadriven::DBMatOnlineDE.

References alpha, chess::b, getB(), sgpp::datadriven::DBMatOfflineOrthoAdapt::getQ(), sgpp::base::DataVector::getSize(), sgpp::datadriven::DBMatOfflineOrthoAdapt::getTinv(), sgpp::datadriven::DBMatOnline::offlineObject, sgpp::base::DataVector::resizeZero(), and sgpp::datadriven::DBMatDMSOrthoAdapt::solve().

updateSystemMatrixDecomposition()

std::vector< size_t > sgpp::datadriven::DBMatOnlineDEOrthoAdapt::updateSystemMatrixDecomposition ( DensityEstimationConfiguration &  densityEstimationConfig, Grid &  grid, size_t  numAddedGridPoints, std::list< size_t >  deletedGridPointIndices, double  lambda  )

overridevirtual

Parameters

densityEstimationConfig	configuration for the density estimation
grid	the underlying grid
numAddedGridPoints	Number of grid points inserted at the end of the grid storage
deletedGridPointIndices	Indices of grid points that were deleted
lambda	The last best lambda value

Returns

list of grid points, that cannot be coarsened

Reimplemented from sgpp::datadriven::DBMatOnline.

References sgpp::datadriven::DBMatOffline::getGridSize(), sgpp::datadriven::DBMatOnline::offlineObject, and sherman_morrison_adapt().

Referenced by getRefinedPointsPointer().

Member Data Documentation

b_adapt_matrix_

sgpp::base::DataMatrix sgpp::datadriven::DBMatOnlineDEOrthoAdapt::b_adapt_matrix_

protected

Referenced by DBMatOnlineDEOrthoAdapt(), getB(), and sherman_morrison_adapt().

b_is_refined

bool sgpp::datadriven::DBMatOnlineDEOrthoAdapt::b_is_refined

protected

Referenced by DBMatOnlineDEOrthoAdapt(), and sherman_morrison_adapt().

current_refine_index

size_t sgpp::datadriven::DBMatOnlineDEOrthoAdapt::current_refine_index

protected

Referenced by DBMatOnlineDEOrthoAdapt(), and sherman_morrison_adapt().

refined_points_

std::vector<sgpp::base::DataVector> sgpp::datadriven::DBMatOnlineDEOrthoAdapt::refined_points_

protected

Referenced by add_new_refine_point(), DBMatOnlineDEOrthoAdapt(), getRefinedPointsPointer(), and sherman_morrison_adapt().

---

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

• datadriven/src/sgpp/datadriven/algorithm/DBMatOnlineDEOrthoAdapt.hpp
• datadriven/src/sgpp/datadriven/algorithm/DBMatOnlineDEOrthoAdapt.cpp