Lookup into Kernel. More...

#include <yat/classifier/KernelLookup.h>

Public Types
typedef double	value_type

typedef const double	const_reference

typedef utility::StrideIterator < utility::Container2DIterator < const KernelLookup, const double, const_reference > >	const_iterator
	'Read Only' iterator

typedef const_iterator	const_column_iterator

typedef const_iterator	const_row_iterator

Public Member Functions
	KernelLookup (const Kernel &kernel, const bool owner=false)
	Constructor a Lookup into a Kernel. More...

	KernelLookup (const Kernel &kernel, const utility::Index &row, const utility::Index &column, const bool owner=false)
	Constructing a Lookup into a subKernel. More...

	KernelLookup (const KernelLookup &kl)
	Copy constructor. More...

	KernelLookup (const KernelLookup &kl, const utility::Index &row, const utility::Index &column)
	Contructing a sub-KernelLookup. More...

	KernelLookup (const KernelLookup &kl, const utility::Index &, const bool row=false)

virtual	~KernelLookup (void)
	Destructor. More...

const_iterator	begin (void) const

const_column_iterator	begin_column (size_t) const

const_row_iterator	begin_row (size_t) const

size_t	columns (void) const

MatrixLookup	data (void) const

MatrixLookupWeighted	data_weighted (void) const

double	element (const DataLookup1D &vec, size_t i) const

double	element (const DataLookupWeighted1D &vec, size_t i) const

const_iterator	end (void) const

const_column_iterator	end_column (size_t) const

const_row_iterator	end_row (size_t) const

size_t	rows (void) const

KernelLookup	selected (const utility::Index &index) const

KernelLookup	test_kernel (const MatrixLookup &data) const

KernelLookup	test_kernel (const MatrixLookupWeighted &data) const

bool	weighted (void) const

const_reference	operator() (size_t row, size_t column) const

Detailed Description

Lookup into Kernel.

This is the KernelLookup class to be used together with kernel methods such as Support Vector Machines (SVM). The class does not contain any data or values, but rather is a lookup into a Kernel object. Each row and each column corresponds to a row and a column in the Kernel, respectively. This design allows for fast creation of sub-kernels, which is a common operation in most traning/validation procedures.

A KernelLookup can be created directly from a Kernel or from another KernelLookup. In the latter case, the resulting KernelLookup is looking directly into the underlying Kernel to avoid multiple lookups.

There is a possibility to set the KernelLookup as owner of the underlying kernel. This implies that underlying kernel is deleted in destructor of MatrixLookup, but only if there is no other owner of the underlying kernel. A reference counter is used to keep track of number of owners. Ownership is copied in copy constructors and assignments.

Member Typedef Documentation

typedef const_iterator theplu::yat::classifier::KernelLookup::const_column_iterator

'Read only' iterator intended to iterate over a column

typedef const double theplu::yat::classifier::KernelLookup::const_reference

const_reference type is const double

Since: New in yat 0.5

typedef const_iterator theplu::yat::classifier::KernelLookup::const_row_iterator

'Read only' iterator intended to iterate over a row

typedef double theplu::yat::classifier::KernelLookup::value_type

value_type is double

Since: New in yat 0.5

Constructor & Destructor Documentation

theplu::yat::classifier::KernelLookup::KernelLookup	(	const Kernel &	kernel,
		const bool	owner = `false`
	)

Constructor a Lookup into a Kernel.

Constructs a KernelLookup corresponding to the Kernel kernel. By default owner is set to false, which means KernelLookup does not own the underlying Kernel.

Note: If underlying Kernel goes out of scope or is deleted, the KernelLookup becomes invalid and the result of further use is undefined.; Do not construct two KernelLookups from the same kernel with owner set to true because that will cause multiple deletion of kernel.

theplu::yat::classifier::KernelLookup::KernelLookup	(	const Kernel &	kernel,
		const utility::Index &	row,
		const utility::Index &	column,
		const bool	owner = `false`
	)

Constructing a Lookup into a subKernel.

Creating a Lookup into parts of the Kernel. In the created Lookup the element in the $ i $ th row in the $ j $ th column is identical to the element in row row[i] and columns column[j] in the underlying kernel. If owner is set to true yhe underlying kernel is destroyed in the destructor.

Note: If kernel goes out of scope or is deleted, the returned pointer becomes invalid and the result of further use is undefined.; For training usage row index shall always be equal to column index.

theplu::yat::classifier::KernelLookup::KernelLookup ( const KernelLookup & kl )

Copy constructor.

A Lookup is created looking into the same underlying Kernel as kl is looking into.

If kl is owner of underlying data, constructed KernelLookup will also be set as owner of underlying data.

theplu::yat::classifier::KernelLookup::KernelLookup	(	const KernelLookup &	kl,
		const utility::Index &	row,
		const utility::Index &	column
	)

Contructing a sub-KernelLookup.

Contructor building a sub-KernelLookup from a KernelLookup defined by row index vector and column index vector. In the created Lookup the element in the $ i $ th row in the $ j $ th column is identical to the element in row row[i] and columns column[j] in the copied kl. The resulting KernelLookup is independent of the old KernelLookup, but is undefined in case underlying Kernel is destroyed.

If kl is owner of underlying data, constructed KernelLookup will also be set as owner of underlying data.

Note: For training usage row index shall always be equal to column index.

theplu::yat::classifier::KernelLookup::KernelLookup	(	const KernelLookup &	kl,
		const utility::Index &	,
		const bool	row = `false`
	)

Constructor taking the column (default) or row index as input. If row is false the created KernelLookup will have equally many rows as kernel.

If kl is owner of underlying data, constructed KernelLookup will also be set as owner of underlying data.

Note: If underlying kernel goes out of scope or is deleted, the KernelLookup becomes invalid and the result of further use is undefined.

virtual theplu::yat::classifier::KernelLookup::~KernelLookup ( void )

virtual

Destructor.

Deletes underlying Kernel if KernelLookup owns it and there is no other owner.

Member Function Documentation

const_iterator theplu::yat::classifier::KernelLookup::begin ( void ) const

Iterator iterates along a row. When end of row is reached it jumps to beginning of next row.

Returns: const_iterator pointing to upper-left element.

const_column_iterator theplu::yat::classifier::KernelLookup::begin_column ( size_t ) const

Iterator iterates along a column.

Returns: iterator pointing to first element of column i.

const_row_iterator theplu::yat::classifier::KernelLookup::begin_row ( size_t ) const

Iterator iterates along a column.

Returns: const_iterator pointing to first element of column i.

size_t theplu::yat::classifier::KernelLookup::columns ( void ) const

Returns: number of columns

MatrixLookup theplu::yat::classifier::KernelLookup::data ( void ) const

Each column in returned DataLookup corresponds to the column in KernelLookup.

Returns: data that KernelLookup is built upon.

Exceptions

if	KernelLookup is weighted

MatrixLookupWeighted theplu::yat::classifier::KernelLookup::data_weighted ( void ) const

Each column in returned DataLookup corresponds to the column in KernelLookup.

Returns: data that KernelLookup is built upon.

Exceptions

if	KernelLookup is unweighted

double theplu::yat::classifier::KernelLookup::element	(	const DataLookup1D &	vec,
		size_t	i
	)		const

Function to calculate a new Kernel element using the underlying KernelFunction. The value is calculated between vec and the data vector of the i th sample, in other words, the sample corresponding to the i th row.

double theplu::yat::classifier::KernelLookup::element	(	const DataLookupWeighted1D &	vec,
		size_t	i
	)		const

Function to calculate a new Kernel element using the underlying KernelFunction. The value is calulated between vec and the data vector of the $ i $ th sample, in other words, the sample corresponding to the $ i $ th row or $ i $ th column. In case KernelLookup is a sub-Kernel and not symmetric, the kernel value is calculated between vec and the data vector corresponding to $ i $ th row.

const_iterator theplu::yat::classifier::KernelLookup::end ( void ) const

Returns: const_iterator pointing to end of matrix

const_column_iterator theplu::yat::classifier::KernelLookup::end_column ( size_t ) const

Returns: const_iterator pointing to end of column i

const_row_iterator theplu::yat::classifier::KernelLookup::end_row ( size_t ) const

Returns: const_iterator pointing to end of row i

const_reference theplu::yat::classifier::KernelLookup::operator()	(	size_t	row,
		size_t	column
	)		const

Returns: element at position (row, column) in the Kernel matrix

size_t theplu::yat::classifier::KernelLookup::rows ( void ) const

Returns: number of rows

KernelLookup theplu::yat::classifier::KernelLookup::selected ( const utility::Index & index ) const

Each element in returned KernelLookup is calculated using only selected features (defined by index). Each element corresponds to the same pair of samples as in the original KernelLookup.

KernelLookup theplu::yat::classifier::KernelLookup::test_kernel ( const MatrixLookup & data ) const

This function is useful when predicting on an independent data set using a kernel-based classifier. In returned KernelLookup column $ i $ corresponds to column $ i $ in data. Row $ i $ in returned KernelLookup corresponds to same sample as row $ i $ in this. In other words, this function returns a KernelLookup containing the kernel elements between the passed data and the internal underlying data this was built from.

KernelLookup theplu::yat::classifier::KernelLookup::test_kernel ( const MatrixLookupWeighted & data ) const

This function is useful when predicting on an independent data set using a kernel-based classifier. In returned KernelLookup column $ i $ corresponds to column $ i $ in data. Row $ i $ in returned KernelLookup corresponds to same sample as row $ i $ in this. In other words, this function returns a KernelLookup containing the kernel elements between the passed data and the internal underlying data this was built from.

bool theplu::yat::classifier::KernelLookup::weighted ( void ) const

Returns: true if underlying Kernel is weighted

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

yat/classifier/KernelLookup.h

Public Types

Public Member Functions

Detailed Description

Member Typedef Documentation

Constructor & Destructor Documentation

Member Function Documentation