MKLMultiClassGradient Class Reference

---

Detailed Description

MKLMultiClassGradient is a helper class for MKLMultiClass.

it solves the corresponding linear problem arising in SIP formulation for MKL using a gradient based approach

Definition at line 28 of file MKLMultiClassGradient.h.

Inheritance diagram for MKLMultiClassGradient:

List of all members.

Public Member Functions
	MKLMultiClassGradient ()
virtual	~MKLMultiClassGradient ()
	MKLMultiClassGradient (MKLMultiClassGradient &gl)
MKLMultiClassGradient	operator= (MKLMultiClassGradient &gl)
virtual void	setup (const int32_t numkernels2)
virtual void	addconstraint (const ::std::vector< float64_t > &normw2, const float64_t sumofpositivealphas)
virtual void	computeweights (std::vector< float64_t > &weights2)
virtual const char *	get_name () const
virtual void	set_mkl_norm (float64_t norm)
Protected Member Functions
void	genbetas (::std::vector< float64_t > &weights, const ::std::vector< float64_t > &gammas)
void	gengammagradient (::std::vector< float64_t > &gammagradient, const ::std::vector< float64_t > &gammas, const int32_t dim)
float64_t	objectives (const ::std::vector< float64_t > &weights, const int32_t index)
void	linesearch (std::vector< float64_t > &finalbeta, const std::vector< float64_t > &oldweights)
Protected Attributes
int32_t	numkernels
::std::vector< ::std::vector < float64_t > >	normsofsubkernels
::std::vector< float64_t >	sumsofalphas
float64_t	pnorm

---

Constructor & Destructor Documentation

MKLMultiClassGradient

(

)

Class default Constructor

Definition at line 15 of file MKLMultiClassGradient.cpp.

~MKLMultiClassGradient

(

)

[virtual]

Class default Destructor

Definition at line 21 of file MKLMultiClassGradient.cpp.

MKLMultiClassGradient

(

MKLMultiClassGradient &

gl

)

Class Copy Constructor

Definition at line 33 of file MKLMultiClassGradient.cpp.

---

Member Function Documentation

void addconstraint	(	const ::std::vector< float64_t > &	normw2,
		const float64_t	sumofpositivealphas
	)			[virtual]

adds a constraint to the LP arising in L1 MKL based on two parameters

Parameters:

	normw2	is the vector of for all kernels
	sumofpositivealphas	is a term depending on alphas, labels and biases, see in the function float64_t getsumofsignfreealphas() from MKLMultiClass.h, it depends on the formulation of the underlying GMNPSVM.

Reimplemented from MKLMultiClassOptimizationBase.

Definition at line 60 of file MKLMultiClassGradient.cpp.

void computeweights

(

std::vector< float64_t > &

weights2

)

[virtual]

computes MKL weights

Parameters:

weights2

stores the new weights

Reimplemented from MKLMultiClassOptimizationBase.

Definition at line 355 of file MKLMultiClassGradient.cpp.

void genbetas	(	::std::vector< float64_t > &	weights,
		const ::std::vector< float64_t > &	gammas
	)			[protected]

helper routine for MKL optimization, computes form manifold coordinates the point on the manifold

Parameters:

	gammas	- manifold coordinates
	weights	- the point on the manifold

Definition at line 67 of file MKLMultiClassGradient.cpp.

void gengammagradient	(	::std::vector< float64_t > &	gammagradient,
		const ::std::vector< float64_t > &	gammas,
		const int32_t	dim
	)			[protected]

helper routine for MKL optimization, computes greadient of manifold parametrization for one coordinate

Parameters:

	gammagradient	- gradient
	gammas	- manifold coordinates
	dim	- the coordinate for which thegradient is to be computed

Definition at line 98 of file MKLMultiClassGradient.cpp.

virtual const char* get_name

(

void

)

const [virtual]

Returns:

object name

Reimplemented from MKLMultiClassOptimizationBase.

Definition at line 77 of file MKLMultiClassGradient.h.

void linesearch	(	std::vector< float64_t > &	finalbeta,
		const std::vector< float64_t > &	oldweights
	)			[protected]

helper routine for MKL optimization, performs linesearch

Parameters:

	finalbeta
	oldweights

Definition at line 154 of file MKLMultiClassGradient.cpp.

float64_t objectives	(	const ::std::vector< float64_t > &	weights,
		const int32_t	index
	)			[protected]

helper routine for MKL optimization, computes optimization objective for one contraint

Parameters:

	weights	- MKL weights
	index	- index of constraint

Definition at line 138 of file MKLMultiClassGradient.cpp.

MKLMultiClassGradient operator=

(

MKLMultiClassGradient &

gl

)

Class Assignment operator

Definition at line 26 of file MKLMultiClassGradient.cpp.

void set_mkl_norm

(

float64_t

norm

)

[virtual]

sets p-norm parameter for MKL

Parameters:

norm

the MKL norm

Reimplemented from MKLMultiClassOptimizationBase.

Definition at line 52 of file MKLMultiClassGradient.cpp.

void setup

(

const int32_t

numkernels2

)

[virtual]

initializes solver

Parameters:

numkernels2

is the number of kernels

Reimplemented from MKLMultiClassOptimizationBase.

Definition at line 40 of file MKLMultiClassGradient.cpp.

---

Member Data Documentation

::std::vector< ::std::vector<float64_t> > normsofsubkernels [protected]

stores normsofsubkernels which is a constraint, normsofsubkernels[i] belongs to the i-th constraint

Definition at line 125 of file MKLMultiClassGradient.h.

int32_t numkernels [protected]

stores the number of kernels which acts as a parameter for the LP

Definition at line 122 of file MKLMultiClassGradient.h.

float64_t pnorm [protected]

stores the L^p norm which acts as a parameter for the LP

Definition at line 129 of file MKLMultiClassGradient.h.

::std::vector< float64_t > sumsofalphas [protected]

stores the bias type term of constraints, sumsofalphas[i] belongs to the i-th constraint

Definition at line 127 of file MKLMultiClassGradient.h.

---

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

• MKLMultiClassGradient.h
• MKLMultiClassGradient.cpp