cla3p::lra::RankModerator< T_Matrix > Class Template Reference

The rank moderator class. More...

Constructors
	RankModerator ()
	Default constructor.
	RankModerator (T_RScalar tol, lraMethod_t method, T_RScalar ref=autoReference())
	The parameterized constructor.
	~RankModerator ()
	Destructor.

Arguments
T_RScalar	cutTolerance () const
	The cutoff tolerance.
lraMethod_t	reductionMethod () const
	The reduction method.
T_RScalar	maxReference () const
	The reference value.
void	setCutTolerance (T_RScalar tol)
	Sets the cutoff tolerance.
void	setReductionMethod (lraMethod_t method)
	Sets the reduction method.
void	setMaxReference (T_RScalar ref)
	The reference value.

Public Member Functions
void	clear ()
	Clears the rank moderator.
XxMatrix< T_Matrix >	reduce (const T_Matrix &mat) const
	Full-rank to low-rank convertor.

Advanced Public Member Functions
XxMatrix< T_Matrix >	reduceUsingDefaultSVD (const T_Matrix &mat, DefaultSVD< T_Matrix > &svd) const
	Full-rank to low-rank convertor.
XxMatrix< T_Matrix >	reduceUsingFastQR (const T_Matrix &mat, PartialQR< T_Matrix > &pqr, DefaultSVD< T_Matrix > &svd) const
	Full-rank to low-rank convertor.

Global Options
static constexpr T_RScalar	autoReference ()
	Reference value will be set automaticaly.

Detailed Description

template<typename T_Matrix>
class cla3p::lra::RankModerator< T_Matrix >

The rank moderator class.

Controls the rank truncation strategy for low-rank approximations based on singular values.

Template Parameters

T_Matrix

The matrix type (e.g., dns::XxMatrix<T_Scalar>).

Constructor & Destructor Documentation

RankModerator() [1/2]

template<typename T_Matrix>

cla3p::lra::RankModerator< T_Matrix >::RankModerator

(

)

Default constructor.

Creates a rank moderator with default parameter values.

RankModerator() [2/2]

template<typename T_Matrix>

cla3p::lra::RankModerator< T_Matrix >::RankModerator ( T_RScalar tol, lraMethod_t method, T_RScalar ref = autoReference() )

explicit

The parameterized constructor.

Constructs a rank moderator with user defined parameters. Ranks kept must satisfy \( \sigma_j > \mathrm{ref} \cdot \mathrm{tol} \).

Parameters

[in]	tol	The cutoff tolerance.
[in]	method	The reduction method.
[in]	ref	Use an auto/fixed value as reference. If negative, the maximum singular value will be used instead.

~RankModerator()

template<typename T_Matrix>

cla3p::lra::RankModerator< T_Matrix >::~RankModerator

(

)

Destructor.

Destroys the rank moderator object.

Member Function Documentation

cutTolerance()

template<typename T_Matrix>

T_RScalar cla3p::lra::RankModerator< T_Matrix >::cutTolerance

(

)

const

The cutoff tolerance.

Controls the level of detail in the low-rank matrix. Ranks kept must satisfy \( \sigma_j > \mathrm{maxReference}() \cdot \mathrm{cutTolerance}() \).

Returns

The cutoff tolerance value.

reductionMethod()

template<typename T_Matrix>

lraMethod_t cla3p::lra::RankModerator< T_Matrix >::reductionMethod

(

)

const

The reduction method.

Controls the full-rank to low-rank conversion method.

Returns

The reduction method.

maxReference()

template<typename T_Matrix>

T_RScalar cla3p::lra::RankModerator< T_Matrix >::maxReference

(

)

const

The reference value.

Controls the level of detail in the low-rank matrix. Ranks kept must satisfy \( \sigma_j > \mathrm{maxReference}() \cdot \mathrm{cutTolerance}() \). If negative, the maximum singular value will be used instead.

Returns

The reference value.

setCutTolerance()

template<typename T_Matrix>

void cla3p::lra::RankModerator< T_Matrix >::setCutTolerance

(

T_RScalar

tol

)

Sets the cutoff tolerance.

Controls the level of detail in the low-rank matrix. Ranks kept must satisfy \( \sigma_j > \mathrm{maxReference}() \cdot \mathrm{cutTolerance}() \).

Parameters

[in]

tol

The desired cutoff tolerance.

setReductionMethod()

template<typename T_Matrix>

void cla3p::lra::RankModerator< T_Matrix >::setReductionMethod

(

lraMethod_t

method

)

Sets the reduction method.

Controls the full-rank to low-rank conversion method.

Parameters

[in]

method

The desired reduction method.

setMaxReference()

template<typename T_Matrix>

void cla3p::lra::RankModerator< T_Matrix >::setMaxReference

(

T_RScalar

ref

)

The reference value.

Controls the level of detail in the low-rank matrix. Ranks kept must satisfy \( \sigma_j > \mathrm{maxReference}() \cdot \mathrm{cutTolerance}() \). If negative, the maximum singular value will be used instead.

Parameters

[in]

ref

The desired auto/fixed reference value.

clear()

template<typename T_Matrix>

void cla3p::lra::RankModerator< T_Matrix >::clear

(

)

Clears the rank moderator.

Resets all members to match those of the default rank moderator.

reduce()

template<typename T_Matrix>

XxMatrix< T_Matrix > cla3p::lra::RankModerator< T_Matrix >::reduce

(

const T_Matrix &

mat

)

const

Full-rank to low-rank convertor.

Converts a full rank dense matrix to low-rank using internal parameterization.

Parameters

[in]

mat

The dense matrix to be reduced.

Returns

The low-rank approximation of mat.

reduceUsingDefaultSVD()

template<typename T_Matrix>

XxMatrix< T_Matrix > cla3p::lra::RankModerator< T_Matrix >::reduceUsingDefaultSVD	(	const T_Matrix &	mat,
		DefaultSVD< T_Matrix > &	svd ) const

Full-rank to low-rank convertor.

Converts a full rank dense matrix to low-rank using the standard SVD method. A user parameterized svd object is supplied in order to avoid repeated (de-)allocations.

Parameters

[in]	mat	The dense matrix to be reduced.
[in,out]	svd	The Singular Value Decomposition object to be used.

Returns

The low-rank approximation of mat.

reduceUsingFastQR()

template<typename T_Matrix>

XxMatrix< T_Matrix > cla3p::lra::RankModerator< T_Matrix >::reduceUsingFastQR	(	const T_Matrix &	mat,
		PartialQR< T_Matrix > &	pqr,
		DefaultSVD< T_Matrix > &	svd ) const

Full-rank to low-rank convertor.

Converts a full rank dense matrix to low-rank using the hybrid QR/SVD method. User parameterized pqr/svd objects are supplied in order to avoid repeated (de-)allocations.

Parameters

[in]	mat	The dense matrix to be reduced.
[in,out]	pqr	The Partial QR Decomposition object to be used.
[in,out]	svd	The Singular Value Decomposition object to be used.

Returns

The low-rank approximation of mat.

autoReference()

template<typename T_Matrix>

constexpr T_RScalar cla3p::lra::RankModerator< T_Matrix >::autoReference ( )

inlinestaticconstexpr

Reference value will be set automaticaly.

The default value for automatic deduction of the reference value \( \sigma_{\mathrm{max}} \).

Returns

The automatic reference sentinel value.