Template class for performing cross-validation on the regularization parameter: derived-from-CV_base thorugh CRTP ('CV_alpha' its template D parameter). More...

#include <CV_alpha.hpp>

Inheritance diagram for CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >:

Public Member Functions
template<typename STOR_OBJ, typename STRATEGY>
	CV_alpha (STOR_OBJ &&Data, STRATEGY &&strategy, const std::vector< double > &params, int k, const pred_func_t< k_imp > &pred_f, int number_threads)
	Constructor if number of PPCs k is known (k_imp=K_IMP::YES), for derived class: constructs firstly CV_base<CV_alpha,...>

template<typename STOR_OBJ, typename STRATEGY>
	CV_alpha (STOR_OBJ &&Data, STRATEGY &&strategy, const std::vector< double > &params, double threshold_ppc, const pred_func_t< k_imp > &pred_f, int number_threads)
	Constructor if number of PPCs k is not known (k_imp=K_IMP::NO), but has to be retained using explanatory power criterion, for derived class: constructs firstly CV_base<CV_alpha,...>

double	param_best () const
	Getter for the best regularization parameter.

std::vector< double >	valid_errors () const
	Getter for validation errors.

double	best_valid_error () const
	Getter for the best validation error.

double	error_single_cv_iter (const double &param, const KO_Traits::StoringMatrix &training_set, const KO_Traits::StoringMatrix &validation_set) const
	Error for a single cross-validation iteration (parameter given), with fixed training and validation sets.

double	error_single_param (const double &param, const cv_strategy_t &strat, const std::size_t &number_cv_iter) const
	Validation error for a given parameter, as the mean of the errors on the various validation sets.

void	best_param_search ()
	Selecting the best regularization parameter, modifying it into the class.

Public Member Functions inherited from CV_base< CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >, cv_strat, err_eval, k_imp, valid_err_ret >
double	err_valid_set_eval (const KO_Traits::StoringVector &pred, const KO_Traits::StoringVector &valid, int number_threads) const
	Evaluation of the loss between prediction on validation set and validation set: estimate of L2 norm loss. Tag-dispacther.

	CV_base (STOR_OBJ &&Data, STRATEGY &&strategy, int number_threads)
	Constructor for the class.

KO_Traits::StoringMatrix	Data () const
	Getter for the data matrix.

cv_strategy< cv_strat >	strategy () const
	Getter for the training/validation set splitting.

int	number_threads () const
	Getter for the number of threads for OMP.

Detailed Description

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>
class CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >

Template class for performing cross-validation on the regularization parameter: derived-from-CV_base thorugh CRTP ('CV_alpha' its template D parameter).

Template Parameters

cv_strat	strategy for splitting training/validation sets
err_eval	how to evaluate the loss between prediction on validation set and validation set
k_imp	if k is imposed or has to be found through explanatory power criterion
valid_err_ret	if validation error are stored

Derived class. Polymorphism is known at compile time thanks to Curiously Recursive Template Pattern (CRTP)

Constructor & Destructor Documentation

◆ CV_alpha() [1/2]

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>

template<typename STOR_OBJ, typename STRATEGY>

CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >::CV_alpha	(	STOR_OBJ &&	Data,
		STRATEGY &&	strategy,
		const std::vector< double > &	params,
		int	k,
		const pred_func_t< k_imp > &	pred_f,
		int	number_threads )

inline

Constructor if number of PPCs k is known (k_imp=K_IMP::YES), for derived class: constructs firstly CV_base<CV_alpha,...>

Parameters

Data	fts data matrix
strategy	splitting training/validation strategy
params	input space for regularization parameter
k	number of retained PPCs
pred_f	function to make validation set prediction (overloading with k imposed)
number_threads	number of threads for OMP

Universal constructor: move semantic used to optimazing handling big size objects

◆ CV_alpha() [2/2]

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>

template<typename STOR_OBJ, typename STRATEGY>

CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >::CV_alpha	(	STOR_OBJ &&	Data,
		STRATEGY &&	strategy,
		const std::vector< double > &	params,
		double	threshold_ppc,
		const pred_func_t< k_imp > &	pred_f,
		int	number_threads )

inline

Constructor if number of PPCs k is not known (k_imp=K_IMP::NO), but has to be retained using explanatory power criterion, for derived class: constructs firstly CV_base<CV_alpha,...>

Parameters

Data	fts data matrix
strategy	splitting training/validation strategy
params	input space for regularization parameter
threshold_ppc	requested explanatory power for PPCs
pred_f	function to make validation set prediction (overloading with k not imposed)
number_threads	number of threads for OMP

Universal constructor: move semantic used to optimazing handling big size objects

Member Function Documentation

◆ best_param_search()

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>

void CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >::best_param_search ( )

inline

Selecting the best regularization parameter, modifying it into the class.

Note: eventual usage of 'pragma' directive for OMP

◆ best_valid_error()

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>

double CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >::best_valid_error ( ) const

inline

Getter for the best validation error.

Returns: the private m_best_valid_error

◆ error_single_cv_iter()

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>

double CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >::error_single_cv_iter	(	const double &	param,
		const KO_Traits::StoringMatrix &	training_set,
		const KO_Traits::StoringMatrix &	validation_set ) const

inline

Error for a single cross-validation iteration (parameter given), with fixed training and validation sets.

Parameters

param	parameter that is being evaluated through cross-validation
training_set	training set
validation_set	validation set

Returns: the error, according to 'err_eval' class template parameter, between prediction on validation set and validation set

◆ error_single_param()

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>

double CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >::error_single_param	(	const double &	param,
		const cv_strategy_t &	strat,
		const std::size_t &	number_cv_iter ) const

inline

Validation error for a given parameter, as the mean of the errors on the various validation sets.

Parameters

param	element of the input space for regularization parameter
strat	strategy for splitting training and validation set
number_cv_iter	total number of different splits

Returns: the average of the errors between prediction on validation set and validation set, for each split

Note: eventual usage of 'pragma' directive for OMP

◆ param_best()

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>

double CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >::param_best ( ) const

inline

Getter for the best regularization parameter.

Returns: the private m_param_best

◆ valid_errors()

template<CV_STRAT cv_strat, CV_ERR_EVAL err_eval, K_IMP k_imp, VALID_ERR_RET valid_err_ret>

std::vector< double > CV_alpha< cv_strat, err_eval, k_imp, valid_err_ret >::valid_errors ( ) const

inline

Getter for validation errors.

Returns: the private m_valid_errors

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

src/CV_alpha.hpp

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ CV_alpha() [1/2]

◆ CV_alpha() [2/2]

Member Function Documentation

◆ best_param_search()

◆ best_valid_error()

◆ error_single_cv_iter()

◆ error_single_param()

◆ param_best()

◆ valid_errors()