Detailed Description

The Kernel base class.

Non-mathematically spoken, a kernel is a function that given two input objects ${\bf x}$ and ${\bf x'}$ returns a score describing the similarity of the vectors. The score should be larger when the objects are more similar.

It can be defined as

$k({\bf x},{\bf x'})= \Phi_k({\bf x})\cdot \Phi_k({\bf x'})$

where $\Phi$ maps the objects into some potentially high dimensional feature space.

Apart from the input features, the base kernel takes only one argument (the size of the kernel cache) that is used to efficiently train kernel-machines like e.g. SVMs.

In case you would like to define your own kernel, you only have to define a new compute() function (and the kernel name via get_name() and the kernel type get_kernel_type()). A good example to look at is the GaussianKernel.

Definition at line 150 of file Kernel.h.

Inheritance diagram for CKernel:

[legend]

List of all members.

Public Member Functions
	CKernel ()
	CKernel (int32_t size)
	CKernel (CFeatures l, CFeatures r, int32_t size)
virtual	~CKernel ()
float64_t	kernel (int32_t idx_a, int32_t idx_b)
void	get_kernel_matrix (float64_t *dst, int32_t m, int32_t *n)
virtual std::vector< float64_t >	get_kernel_col (int32_t j)
virtual std::vector< float64_t >	get_kernel_row (int32_t i)
template<class T >
T *	get_kernel_matrix (int32_t &m, int32_t &n, T *target)
virtual bool	init (CFeatures lhs, CFeatures rhs)
virtual bool	set_normalizer (CKernelNormalizer *normalizer)
virtual CKernelNormalizer *	get_normalizer ()
virtual bool	init_normalizer ()
virtual void	cleanup ()
void	load (CFile *loader)
void	save (CFile *writer)
CFeatures *	get_lhs ()
CFeatures *	get_rhs ()
virtual int32_t	get_num_vec_lhs ()
virtual int32_t	get_num_vec_rhs ()
virtual bool	has_features ()
bool	get_lhs_equals_rhs ()
virtual void	remove_lhs_and_rhs ()
virtual void	remove_lhs ()
virtual void	remove_rhs ()
	takes all necessary steps if the rhs is removed from kernel
virtual EKernelType	get_kernel_type ()=0
virtual EFeatureType	get_feature_type ()=0
virtual EFeatureClass	get_feature_class ()=0
void	set_cache_size (int32_t size)
int32_t	get_cache_size ()
void	list_kernel ()
bool	has_property (EKernelProperty p)
virtual void	clear_normal ()
virtual void	add_to_normal (int32_t vector_idx, float64_t weight)
EOptimizationType	get_optimization_type ()
virtual void	set_optimization_type (EOptimizationType t)
bool	get_is_initialized ()
virtual bool	init_optimization (int32_t count, int32_t IDX, float64_t weights)
virtual bool	delete_optimization ()
bool	init_optimization_svm (CSVM *svm)
virtual float64_t	compute_optimized (int32_t vector_idx)
virtual void	compute_batch (int32_t num_vec, int32_t vec_idx, float64_t target, int32_t num_suppvec, int32_t IDX, float64_t alphas, float64_t factor=1.0)
float64_t	get_combined_kernel_weight ()
void	set_combined_kernel_weight (float64_t nw)
virtual int32_t	get_num_subkernels ()
virtual void	compute_by_subkernel (int32_t vector_idx, float64_t *subkernel_contrib)
virtual const float64_t *	get_subkernel_weights (int32_t &num_weights)
virtual void	set_subkernel_weights (float64_t *weights, int32_t num_weights)
Protected Member Functions
void	set_property (EKernelProperty p)
void	unset_property (EKernelProperty p)
void	set_is_initialized (bool p_init)
virtual float64_t	compute (int32_t x, int32_t y)=0
int32_t	compute_row_start (int64_t offs, int32_t n, bool symmetric)
virtual void	load_serializable_post () throw (ShogunException)
virtual void	save_serializable_pre () throw (ShogunException)
virtual void	save_serializable_post () throw (ShogunException)
Static Protected Member Functions
template<class T >
static void *	get_kernel_matrix_helper (void *p)
Protected Attributes
int32_t	cache_size
	cache_size in MB
KERNELCACHE_ELEM *	kernel_matrix
CFeatures *	lhs
	feature vectors to occur on left hand side
CFeatures *	rhs
	feature vectors to occur on right hand side
bool	lhs_equals_rhs
	lhs
int32_t	num_lhs
	number of feature vectors on left hand side
int32_t	num_rhs
	number of feature vectors on right hand side
float64_t	combined_kernel_weight
bool	optimization_initialized
EOptimizationType	opt_type
uint64_t	properties
CKernelNormalizer *	normalizer
Friends
class	CVarianceKernelNormalizer
class	CSqrtDiagKernelNormalizer
class	CAvgDiagKernelNormalizer
class	CRidgeKernelNormalizer
class	CFirstElementKernelNormalizer
class	CMultitaskKernelNormalizer
class	CMultitaskKernelMklNormalizer
class	CMultitaskKernelMaskNormalizer
class	CMultitaskKernelMaskPairNormalizer
class	CTanimotoKernelNormalizer
class	CDiceKernelNormalizer
class	CZeroMeanCenterKernelNormalizer

Constructor & Destructor Documentation

CKernel ( )

default constructor

Definition at line 38 of file Kernel.cpp.

CKernel ( int32_t size )

constructor

Parameters:

size	cache size

Definition at line 43 of file Kernel.cpp.

CKernel	(	CFeatures *	l,
		CFeatures *	r,
		int32_t	size
	)

constructor

Parameters:

l	features for left-hand side
r	features for right-hand side
size	cache size

Definition at line 54 of file Kernel.cpp.

~CKernel ( ) [virtual]

Definition at line 67 of file Kernel.cpp.

Member Function Documentation

void add_to_normal	(	int32_t	vector_idx,
		float64_t	weight
	)		`[virtual]`

add vector*factor to 'virtual' normal vector

Parameters:

vector_idx	index
weight	weight

Reimplemented in CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CLinearKernel, CLinearStringKernel, CWeightedCommWordStringKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 334 of file Kernel.cpp.

void cleanup ( ) [virtual]

clean up your kernel

base method only removes lhs and rhs overload to add further cleanup but make sure CKernel::cleanup() is called

Reimplemented in CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CCustomKernel, CFixedDegreeStringKernel, CGaussianKernel, CGaussianMatchStringKernel, CHistogramWordStringKernel, CLinearKernel, CLinearStringKernel, CLocalAlignmentStringKernel, COligoStringKernel, CPolyKernel, CPolyMatchStringKernel, CPolyMatchWordStringKernel, CPyramidChi2, CSalzbergWordStringKernel, CSigmoidKernel, CSimpleLocalityImprovedStringKernel, CSNPStringKernel, CSparseSpatialSampleStringKernel, CSpectrumMismatchRBFKernel, CSpectrumRBFKernel, CWeightedCommWordStringKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 162 of file Kernel.cpp.

void clear_normal ( ) [virtual]

for optimizable kernels, i.e. kernels where the weight vector can be computed explicitly (if it fits into memory)

Reimplemented in CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CLinearKernel, CLinearStringKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 339 of file Kernel.cpp.

virtual float64_t compute	(	int32_t	x,
		int32_t	y
	)		`[protected, pure virtual]`

compute kernel function for features a and b idx_{a,b} denote the index of the feature vectors in the corresponding feature object

abstract base method

Parameters:

x	index a
y	index b

Returns:: computed kernel function at indices a,b

Implemented in CAUCKernel, CChi2Kernel, CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CConstKernel, CCustomKernel, CDiagKernel, CDistanceKernel, CDotKernel, CFixedDegreeStringKernel, CGaussianKernel, CGaussianMatchStringKernel, CGaussianShiftKernel, CGaussianShortRealKernel, CHistogramIntersectionKernel, CHistogramWordStringKernel, CLinearStringKernel, CLocalAlignmentStringKernel, CLocalityImprovedStringKernel, CMatchWordStringKernel, COligoStringKernel, CPolyKernel, CPolyMatchStringKernel, CPolyMatchWordStringKernel, CPyramidChi2, CRegulatoryModulesStringKernel, CSalzbergWordStringKernel, CSigmoidKernel, CSimpleLocalityImprovedStringKernel, CSNPStringKernel, CSparseSpatialSampleStringKernel, CSpectrumMismatchRBFKernel, CSpectrumRBFKernel, CTensorProductPairKernel, CWeightedDegreePositionStringKernel, CWeightedDegreeRBFKernel, and CWeightedDegreeStringKernel.

void compute_batch	(	int32_t	num_vec,
		int32_t *	vec_idx,
		float64_t *	target,
		int32_t	num_suppvec,
		int32_t *	IDX,
		float64_t *	alphas,
		float64_t	factor = `1.0`
	)		`[virtual]`

computes output for a batch of examples in an optimized fashion (favorable if kernel supports it, i.e. has KP_BATCHEVALUATION. to the outputvector target (of length num_vec elements) the output for the examples enumerated in vec_idx are added. therefore make sure that it is initialized with ZERO. the following num_suppvec, IDX, alphas arguments are the number of support vectors, their indices and weights

Reimplemented in CCombinedKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 327 of file Kernel.cpp.

void compute_by_subkernel	(	int32_t	vector_idx,
		float64_t *	subkernel_contrib
	)		`[virtual]`

compute by subkernel

Parameters:

vector_idx	index
subkernel_contrib	subkernel contribution

Reimplemented in CCombinedKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 349 of file Kernel.cpp.

float64_t compute_optimized ( int32_t vector_idx ) [virtual]

compute optimized

Parameters:

vector_idx index to compute

Returns:: optimized value at given index

Reimplemented in CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CLinearKernel, CLinearStringKernel, CWeightedCommWordStringKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 321 of file Kernel.cpp.

int32_t compute_row_start	(	int64_t	offs,
		int32_t	n,
		bool	symmetric
	)		`[protected]`

compute row start offset for parallel kernel matrix computation

Parameters:

offs	offset
n	number of columns
symmetric	whether matrix is symmetric

Definition at line 673 of file Kernel.h.

bool delete_optimization ( ) [virtual]

delete optimization

Returns:: if deleting was successful

Reimplemented in CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CLinearKernel, CLinearStringKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 315 of file Kernel.cpp.

int32_t get_cache_size ( )

return the size of the kernel cache

Returns:: size of kernel cache

Definition at line 502 of file Kernel.h.

float64_t get_combined_kernel_weight ( )

get combined kernel weight

Returns:: combined kernel weight

Definition at line 593 of file Kernel.h.

virtual EFeatureClass get_feature_class ( ) [pure virtual]

return feature class the kernel can deal with

abstract base method

Returns:: feature class

Implemented in CAUCKernel, CChi2Kernel, CCombinedKernel, CConstKernel, CCustomKernel, CDiagKernel, CDistanceKernel, CDotKernel, CSparseKernel< ST >, CStringKernel< ST >, CTensorProductPairKernel, CWeightedDegreeRBFKernel, CStringKernel< char >, CStringKernel< uint16_t >, and CStringKernel< uint64_t >.

virtual EFeatureType get_feature_type ( ) [pure virtual]

return feature type the kernel can deal with

abstract base method

Returns:: feature type

Implemented in CAUCKernel, CChi2Kernel, CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CConstKernel, CCustomKernel, CDiagKernel, CDistanceKernel, CDotKernel, CSparseKernel< ST >, CStringKernel< ST >, CTensorProductPairKernel, CWeightedCommWordStringKernel, CWeightedDegreeRBFKernel, CStringKernel< char >, CStringKernel< uint16_t >, CStringKernel< uint64_t >, CSparseKernel< ST >, CSparseKernel< ST >, CSparseKernel< ST >, CSparseKernel< ST >, CSparseKernel< ST >, CSparseKernel< ST >, CSparseKernel< ST >, CStringKernel< ST >, CStringKernel< ST >, CStringKernel< ST >, CStringKernel< ST >, CStringKernel< ST >, CStringKernel< ST >, and CStringKernel< ST >.

bool get_is_initialized ( )

check if optimization is initialized

Returns:: if optimization is initialized

Definition at line 544 of file Kernel.h.

virtual std::vector<float64_t> get_kernel_col ( int32_t j ) [virtual]

get column j

Returns:: the jth column of the kernel matrix

Definition at line 221 of file Kernel.h.

T* get_kernel_matrix	(	int32_t &	m,
		int32_t &	n,
		T *	target
	)

get kernel matrix real

Parameters:

m	dimension m of matrix
n	dimension n of matrix
target	the kernel matrix

Returns:: the kernel matrix

Definition at line 264 of file Kernel.h.

void get_kernel_matrix	(	float64_t **	dst,
		int32_t *	m,
		int32_t *	n
	)

get kernel matrix

Parameters:

dst	destination where matrix will be stored
m	dimension m of matrix
n	dimension n of matrix

Definition at line 78 of file Kernel.cpp.

static void* get_kernel_matrix_helper ( void * p ) [static, protected]

helper for computing the kernel matrix in a parallel way

Parameters:

p	thread parameters

Definition at line 690 of file Kernel.h.

virtual std::vector<float64_t> get_kernel_row ( int32_t i ) [virtual]

get row i

Returns:: the ith row of the kernel matrix

Definition at line 241 of file Kernel.h.

virtual EKernelType get_kernel_type ( ) [pure virtual]

return what type of kernel we are, e.g. Linear,Polynomial, Gaussian,...

abstract base method

Returns:: kernel type

Implemented in CAUCKernel, CChi2Kernel, CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CConstKernel, CCustomKernel, CDiagKernel, CDistanceKernel, CDotKernel, CFixedDegreeStringKernel, CGaussianKernel, CGaussianMatchStringKernel, CGaussianShiftKernel, CGaussianShortRealKernel, CHistogramIntersectionKernel, CHistogramWordStringKernel, CLinearKernel, CLinearStringKernel, CLocalAlignmentStringKernel, CLocalityImprovedStringKernel, CMatchWordStringKernel, COligoStringKernel, CPolyKernel, CPolyMatchStringKernel, CPolyMatchWordStringKernel, CPyramidChi2, CRegulatoryModulesStringKernel, CSalzbergWordStringKernel, CSigmoidKernel, CSimpleLocalityImprovedStringKernel, CSNPStringKernel, CSparseKernel< ST >, CSparseSpatialSampleStringKernel, CSpectrumMismatchRBFKernel, CSpectrumRBFKernel, CStringKernel< ST >, CTensorProductPairKernel, CWeightedCommWordStringKernel, CWeightedDegreePositionStringKernel, CWeightedDegreeRBFKernel, CWeightedDegreeStringKernel, CStringKernel< char >, CStringKernel< uint16_t >, and CStringKernel< uint64_t >.

CFeatures* get_lhs ( )

get left-hand side of features used in kernel

Returns:: features of left-hand side

Definition at line 410 of file Kernel.h.

bool get_lhs_equals_rhs ( )

test whether features on lhs and rhs are the same

Returns:: true if features are the same

Definition at line 449 of file Kernel.h.

CKernelNormalizer * get_normalizer ( ) [virtual]

obtain the current kernel normalizer

Returns:: the kernel normalizer

Definition at line 151 of file Kernel.cpp.

int32_t get_num_subkernels ( ) [virtual]

get number of subkernels

Returns:: number of subkernels

Reimplemented in CCombinedKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 344 of file Kernel.cpp.

virtual int32_t get_num_vec_lhs ( ) [virtual]

get number of vectors of lhs features

Returns:: number of vectors of left-hand side

Reimplemented in CCustomKernel.

Definition at line 422 of file Kernel.h.

virtual int32_t get_num_vec_rhs ( ) [virtual]

get number of vectors of rhs features

Returns:: number of vectors of right-hand side

Reimplemented in CCustomKernel.

Definition at line 431 of file Kernel.h.

EOptimizationType get_optimization_type ( )

get optimization type

Returns:: optimization type

Definition at line 532 of file Kernel.h.

CFeatures* get_rhs ( )

get right-hand side of features used in kernel

Returns:: features of right-hand side

Definition at line 416 of file Kernel.h.

const float64_t * get_subkernel_weights ( int32_t & num_weights ) [virtual]

get subkernel weights

Parameters:

num_weights number of weights will be stored here

Returns:: subkernel weights

Reimplemented in CCombinedKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 355 of file Kernel.cpp.

virtual bool has_features ( ) [virtual]

test whether features have been assigned to lhs and rhs

Returns:: true if features are assigned

Reimplemented in CCombinedKernel, and CCustomKernel.

Definition at line 440 of file Kernel.h.

bool has_property ( EKernelProperty p )

check if kernel has given property

Parameters:

p	kernel property

Returns:: if kernel has given property

Definition at line 514 of file Kernel.h.

bool init	(	CFeatures *	lhs,
		CFeatures *	rhs
	)		`[virtual]`

initialize kernel e.g. setup lhs/rhs of kernel, precompute normalization constants etc. make sure to check that your kernel can deal with the supplied features (!)

Parameters:

lhs	features for left-hand side
rhs	features for right-hand side

Returns:: if init was successful

Reimplemented in CAUCKernel, CChi2Kernel, CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CConstKernel, CCustomKernel, CDiagKernel, CDistanceKernel, CDotKernel, CFixedDegreeStringKernel, CGaussianKernel, CGaussianMatchStringKernel, CGaussianShortRealKernel, CHistogramIntersectionKernel, CHistogramWordStringKernel, CLinearKernel, CLinearStringKernel, CLocalAlignmentStringKernel, CLocalityImprovedStringKernel, CMatchWordStringKernel, COligoStringKernel, CPolyKernel, CPolyMatchStringKernel, CPolyMatchWordStringKernel, CPyramidChi2, CRegulatoryModulesStringKernel, CSalzbergWordStringKernel, CSigmoidKernel, CSimpleLocalityImprovedStringKernel, CSNPStringKernel, CSparseKernel< ST >, CSparseSpatialSampleStringKernel, CSpectrumMismatchRBFKernel, CSpectrumRBFKernel, CStringKernel< ST >, CTensorProductPairKernel, CWeightedCommWordStringKernel, CWeightedDegreePositionStringKernel, CWeightedDegreeRBFKernel, CWeightedDegreeStringKernel, CStringKernel< char >, CStringKernel< uint16_t >, and CStringKernel< uint64_t >.

Definition at line 107 of file Kernel.cpp.

bool init_normalizer ( ) [virtual]

initialize the current kernel normalizer

Returns:: if init was successful

Definition at line 157 of file Kernel.cpp.

bool init_optimization	(	int32_t	count,
		int32_t *	IDX,
		float64_t *	weights
	)		`[virtual]`

initialize optimization

Parameters:

count	count
IDX	index
weights	weights

Returns:: if initializing was successful

Reimplemented in CCombinedKernel, CCommUlongStringKernel, CCommWordStringKernel, CLinearKernel, CLinearStringKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 308 of file Kernel.cpp.

bool init_optimization_svm ( CSVM * svm )

initialize optimization

Parameters:

svm svm model

Returns:: if initializing was successful

Definition at line 368 of file Kernel.cpp.

float64_t kernel	(	int32_t	idx_a,
		int32_t	idx_b
	)

get kernel function for lhs feature vector a and rhs feature vector b

Parameters:

idx_a	index of feature vector a
idx_b	index of feature vector b

Returns:: computed kernel function

Definition at line 196 of file Kernel.h.

void list_kernel ( )

list kernel

Definition at line 225 of file Kernel.cpp.

void load ( CFile * loader )

load the kernel matrix

Parameters:

loader File object via which to load data

Definition at line 169 of file Kernel.cpp.

void load_serializable_post ( void ) throw (ShogunException) [protected, virtual]

Can (optionally) be overridden to post-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::LOAD_SERIALIZABLE_POST is called.

Exceptions:

ShogunException Will be thrown if an error occurres.

Reimplemented from CSGObject.

Reimplemented in CGaussianKernel, and CWeightedDegreePositionStringKernel.

Definition at line 386 of file Kernel.cpp.

void remove_lhs ( ) [virtual]

remove lhs from kernel

Reimplemented in CCombinedKernel, CCommUlongStringKernel, CSpectrumMismatchRBFKernel, CSpectrumRBFKernel, CWeightedDegreePositionStringKernel, and CWeightedDegreeStringKernel.

Definition at line 200 of file Kernel.cpp.

void remove_lhs_and_rhs ( ) [virtual]

remove lhs and rhs from kernel

Reimplemented in CCombinedKernel.

Definition at line 185 of file Kernel.cpp.

void remove_rhs ( ) [virtual]

takes all necessary steps if the rhs is removed from kernel

remove rhs from kernel

Reimplemented in CCombinedKernel, and CCommUlongStringKernel.

Definition at line 212 of file Kernel.cpp.

void save ( CFile * writer )

save kernel matrix

Parameters:

writer File object via which to save data

Definition at line 175 of file Kernel.cpp.

void save_serializable_post ( ) throw (ShogunException) [protected, virtual]

Can (optionally) be overridden to post-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::SAVE_SERIALIZABLE_POST is called.

Exceptions:

ShogunException Will be thrown if an error occurres.

Reimplemented from CSGObject.

Definition at line 401 of file Kernel.cpp.

void save_serializable_pre ( ) throw (ShogunException) [protected, virtual]

Can (optionally) be overridden to pre-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::SAVE_SERIALIZABLE_PRE is called.

Exceptions:

ShogunException Will be thrown if an error occurres.

Reimplemented from CSGObject.

Definition at line 393 of file Kernel.cpp.

void set_cache_size ( int32_t size )

set the size of the kernel cache

Parameters: