Ranking SVM algorithm with support for custom kernels. More...

#include <libcmaes/surrogates/rankingsvm.hpp>

Public Member Functions
void	train (dMat &x, const int &niter, const dMat &covinv, const dVec &xmean)
	trains a ranker from a set of points

void	predict (dVec &fit, dMat &x_test, dMat &x_train, const dMat &covinv, const dVec &xmean)
	predicts a ranking from a learnt ranker

void	encode (dMat &x, const dMat &covinv, const dVec &xmean)
	encoding a set of point in a transformed space

void	compute_training_kernel (dMat &x)
	pre-computation of the kernel values for every examples and coordinates

void	optimize (const dMat &x, const int &niter)
	optimizes a ranker's model given a training set x

double	error (dMat &x_test, dMat &x_train, const dVec &ref_fit, const dMat &covinv, const dVec &xmean)
	computes the ranker's error over a dataset

Public Attributes
bool	_encode = false

dMat	_K

dVec	_alpha

dMat	_dKij

dMat	_C

double	_Cval = 1e6

double	_epsilon = 1.0

TKernel	_kernel

std::mt19937	_rng

std::uniform_real_distribution	_udist

Detailed Description

template<class TKernel = RBFKernel>
class RankingSVM< TKernel >

Ranking SVM algorithm with support for custom kernels.

Member Function Documentation

◆ compute_training_kernel()

template<class TKernel = RBFKernel>

void RankingSVM< TKernel >::compute_training_kernel ( dMat & x )

inline

pre-computation of the kernel values for every examples and coordinates

Parameters

x	training set as a point per column of the matrix

◆ encode()

template<class TKernel = RBFKernel>

void RankingSVM< TKernel >::encode	(	dMat &	x,
		const dMat &	covinv,
		const dVec &	xmean
	)

inline

encoding a set of point in a transformed space

Parameters

x	the points to be transformed, one per column of the matrix
covinv	the inverse sqrt covariance of the points in the training set
training	set mean distribution if available along with covariance matrix

◆ error()

template<class TKernel = RBFKernel>

double RankingSVM< TKernel >::error	(	dMat &	x_test,
		dMat &	x_train,
		const dVec &	ref_fit,
		const dMat &	covinv,
		const dVec &	xmean
	)

inline

computes the ranker's error over a dataset

Parameters

x_test	testing dataset, with one point per column of the matrix
x_train	the initial training dataset, used by encode() if needed
ref_fit	the reference ranking fit against which to test the prediction
covinv	the inverse sqrt covariance of the points in the training set, if available and needed, encode() function.
training	set mean distribution if available along with covariance matrix, and needed, see encode() function

See also: encode

◆ optimize()

template<class TKernel = RBFKernel>

void RankingSVM< TKernel >::optimize	(	const dMat &	x,
		const int &	niter
	)

inline

optimizes a ranker's model given a training set x

Parameters

training	set as a set of points in column of the matrix
niter	the number of iterations allowed for optimization

◆ predict()

template<class TKernel = RBFKernel>

void RankingSVM< TKernel >::predict	(	dVec &	fit,
		dMat &	x_test,
		dMat &	x_train,
		const dMat &	covinv,
		const dVec &	xmean
	)

inline

predicts a ranking from a learnt ranker

Parameters

fit	the final ranking fitted by the ranker
x_test	points to be ranked, one per column of the matrix
x_train	the initial training set, possibly used by kernel computation
covinv	the inverse sqrt covariance of the points in the training set, if available and needed, encode() function.
training	set mean distribution if available along with covariance matrix, and needed, see encode() function

See also: encode

◆ train()

template<class TKernel = RBFKernel>

void RankingSVM< TKernel >::train	(	dMat &	x,
		const int &	niter,
		const dMat &	covinv,
		const dVec &	xmean
	)

inline

trains a ranker from a set of points

Parameters

x	matrix in which every column represents a point from the training set
covinv	the inverse sqrt covariance of the points in the training set, if available and needed, encode() function.
training	set mean distribution if available along with covariance matrix, and needed, see encode() function

See also: encode

Member Data Documentation

◆ _alpha

template<class TKernel = RBFKernel>

dVec RankingSVM< TKernel >::_alpha

vector of Ranking SVM parameters over ranking constraints.

◆ _C

template<class TKernel = RBFKernel>

dMat RankingSVM< TKernel >::_C

constraint violation weights.

◆ _Cval

template<class TKernel = RBFKernel>

double RankingSVM< TKernel >::_Cval = 1e6

constraing violation base weight value.

◆ _encode

template<class TKernel = RBFKernel>

bool RankingSVM< TKernel >::_encode = false

whether to use encoding from inverse sqrt covariance matrix of points.

◆ _K

template<class TKernel = RBFKernel>

dMat RankingSVM< TKernel >::_K

pre-computed matrix of kernel values for a given training set.

◆ _kernel

template<class TKernel = RBFKernel>

TKernel RankingSVM< TKernel >::_kernel

kernel class.

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

include/libcmaes/surrogates/rankingsvm.hpp

Public Member Functions

Public Attributes

Detailed Description

Member Function Documentation

◆ compute_training_kernel()

◆ encode()

◆ error()

◆ optimize()

◆ predict()

◆ train()

Member Data Documentation

◆ _alpha

◆ _C

◆ _Cval

◆ _encode

◆ _K

◆ _kernel