Partly Precomputed version of a matrix for quadratic programming. More...

#include <shark/LinAlg/PartlyPrecomputedMatrix.h>

Public Types
typedef Matrix::QpFloatType	QpFloatType

Public Member Functions
	PartlyPrecomputedMatrix (Matrix *base, std::size_t cachesize=0x4000000)

bool	isCached (std::size_t k) const

void	row (std::size_t k, blas::vector< QpFloatType > &storage) const

QpFloatType	operator() (std::size_t i, std::size_t j) const

QpFloatType	entry (std::size_t i, std::size_t j) const

std::size_t	size () const

std::size_t	getMaxCacheSize ()

std::size_t	getCacheRowSize () const

Protected Attributes
blas::matrix< QpFloatType >	m_cachedMatrix
	container for precomputed values More...

size_t	m_cacheSize

Matrix *	m_baseMatrix

size_t	m_originalNumberOfRows

Detailed Description

template<class Matrix>
class shark::PartlyPrecomputedMatrix< Matrix >

Partly Precomputed version of a matrix for quadratic programming.

: The PartlyPrecomputedMatrix class computes all pairs of kernel evaluations that fits the given cachesize in its constructor and stores them im memory.

: Use of this class may be beneficial for certain model selection strategies, where the whole matrix does not fit into memory, and the LRU cache will produce too much hit rates, so that at least partially caching the kernel matrix will help. In particular this will help the KernelSGD/Pegasos algorithm.

Definition at line 68 of file PartlyPrecomputedMatrix.h.

Member Typedef Documentation

◆ QpFloatType

template<class Matrix >

typedef Matrix::QpFloatType shark::PartlyPrecomputedMatrix< Matrix >::QpFloatType

Definition at line 71 of file PartlyPrecomputedMatrix.h.

Constructor & Destructor Documentation

◆ PartlyPrecomputedMatrix()

template<class Matrix >

shark::PartlyPrecomputedMatrix< Matrix >::PartlyPrecomputedMatrix	(	Matrix *	base,
		std::size_t	cachesize = `0x4000000`
	)

inline

Constructor

Parameters

[in]	base	matrix to be cached. it is assumed that this matrix is not precomputed, but the (costy) computation takes place every time an entry is queried.
[in]	cachesize	size of the cache to use in bytes. the size of the cached matrix will

Definition at line 78 of file PartlyPrecomputedMatrix.h.

References shark::PartlyPrecomputedMatrix< Matrix >::m_baseMatrix, shark::PartlyPrecomputedMatrix< Matrix >::m_cachedMatrix, shark::PartlyPrecomputedMatrix< Matrix >::m_cacheSize, shark::PartlyPrecomputedMatrix< Matrix >::m_originalNumberOfRows, SHARK_RUNTIME_CHECK, and shark::PartlyPrecomputedMatrix< Matrix >::size().

Member Function Documentation

◆ entry()

template<class Matrix >

QpFloatType shark::PartlyPrecomputedMatrix< Matrix >::entry	(	std::size_t	i,
		std::size_t	j
	)		const

inline

return a single matrix entry param[in] i row of entry param[in] j column entry

Returns: value of matrix at given position

Definition at line 168 of file PartlyPrecomputedMatrix.h.

References shark::PartlyPrecomputedMatrix< Matrix >::isCached(), shark::PartlyPrecomputedMatrix< Matrix >::m_baseMatrix, and shark::PartlyPrecomputedMatrix< Matrix >::m_cachedMatrix.

Referenced by shark::PartlyPrecomputedMatrix< Matrix >::operator()().

◆ getCacheRowSize()

template<class Matrix >

std::size_t shark::PartlyPrecomputedMatrix< Matrix >::getCacheRowSize ( ) const

inline

return the dimension of a row in the cache (as we do not shorten our rows, this must be the same as the dimension of a row in the original kernel matrix).

Returns: dimension of any cached row

Definition at line 201 of file PartlyPrecomputedMatrix.h.

References shark::PartlyPrecomputedMatrix< Matrix >::m_cachedMatrix.

◆ getMaxCacheSize()

template<class Matrix >

std::size_t shark::PartlyPrecomputedMatrix< Matrix >::getMaxCacheSize ( )

inline

return size of cached matrix in QpFloatType units

Returns: the capacity of the cached matrix in QpFloatType units

Definition at line 191 of file PartlyPrecomputedMatrix.h.

References shark::PartlyPrecomputedMatrix< Matrix >::m_cachedMatrix.

◆ isCached()

template<class Matrix >

bool shark::PartlyPrecomputedMatrix< Matrix >::isCached ( std::size_t k ) const

inline

return, if a given row is cached

Parameters

[in] k row to check

Returns: is given row in cached matrix or not?

Definition at line 116 of file PartlyPrecomputedMatrix.h.

References shark::PartlyPrecomputedMatrix< Matrix >::m_cachedMatrix.

Referenced by shark::PartlyPrecomputedMatrix< Matrix >::entry(), and shark::PartlyPrecomputedMatrix< Matrix >::row().

◆ operator()()

template<class Matrix >

QpFloatType shark::PartlyPrecomputedMatrix< Matrix >::operator()	(	std::size_t	i,
		std::size_t	j
	)		const

inline

return a single matrix entry param[in] i row of entry param[in] j column entry

Returns: value of matrix at given position

Definition at line 157 of file PartlyPrecomputedMatrix.h.

References shark::PartlyPrecomputedMatrix< Matrix >::entry().

◆ row()

template<class Matrix >

void shark::PartlyPrecomputedMatrix< Matrix >::row	(	std::size_t	k,
		blas::vector< QpFloatType > &	storage
	)		const

inline

return a complete row of the matrix. if the row is cached, it will be returned from there, if not, it will be recomputed on-the-fly and not stored. param[in] k row to compute param[in] storage vector to store the row. must be the same size as a row!

Definition at line 130 of file PartlyPrecomputedMatrix.h.

References shark::PartlyPrecomputedMatrix< Matrix >::isCached(), shark::PartlyPrecomputedMatrix< Matrix >::m_cachedMatrix, shark::PartlyPrecomputedMatrix< Matrix >::m_originalNumberOfRows, RANGE_CHECK, and SIZE_CHECK.