This is an implementation of CMA-ES. It uses the reference algorithm and termination criteria of the following paper: Hansen, N. (2009). Benchmarking a BI-Population CMA-ES on the BBOB-2009 Function Testbed. Workshop Proceedings of the GECCO Genetic and Evolutionary Computation Conference, ACM, pp. 2389-2395 See https://www.lri.fr/~hansen/publications.html for more information. More...

#include <libcmaes/cmastrategy.h>

Inheritance diagram for libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >:

Public Member Functions
	CMAStrategy ()
	dummy constructor

	CMAStrategy (FitFunc &func, CMAParameters< TGenoPheno > &parameters)
	constructor.

	CMAStrategy (FitFunc &func, CMAParameters< TGenoPheno > &parameters, const CMASolutions &cmasols)
	constructor for starting from an existing solution.

dMat	ask ()
	generates nsols new candidate solutions, sampled from a multivariate normal distribution. return A matrix whose rows contain the candidate points.

void	tell ()
	Updates the covariance matrix and prepares for the next iteration.

bool	stop ()
	Stops search on a set of termination criterias, see reference paper.

int	optimize (const EvalFunc &evalf, const AskFunc &askf, const TellFunc &tellf)
	Finds the minimum of the objective function. It makes alternate calls to ask(), tell() and stop() until one of the termination criteria triggers.

int	optimize ()
	Finds the minimum of the objective function. It makes alternate calls to ask(), tell() and stop() until one of the termination criteria triggers.

void	plot ()
	Stream the internal state of the search into an output file, as defined in the _parameters object.

Public Member Functions inherited from libcmaes::ESOStrategy< TParameters, TSolutions, TStopCriteria >
	ESOStrategy ()
	dummy constructor.

	ESOStrategy (FitFunc &func, TParameters &parameters)
	constructor

	ESOStrategy (FitFunc &func, TParameters &parameters, const TSolutions &solutions)
	constructor for starting from an existing solution.

dMat	ask ()
	Generates a set of candidate points.

void	eval (const dMat &candidates, const dMat &phenocandidates=dMat(0, 0))
	Evaluates a set of candidates against the objective function. The procedure is multithreaded and stores both the candidates and their f-value into the _solutions object that bears the current set of potential solutions to the optimization problem.

void	tell ()
	Updates the state of the stochastic search, and prepares for the next iteration.

bool	stop ()
	Decides whether to stop the search for solutions.

int	optimize (const EvalFunc &evalf, const AskFunc &askf, const TellFunc &tellf)
	Finds the minimum of the objective function. It makes alternative calls to ask(), tell() and stop() until one of the termination criteria triggers.

void	inc_iter ()
	increment iteration count.

void	update_fevals (const int &evals)
	updates the consumed budget of objective function evaluations.

void	set_gradient_func (GradFunc &gfunc)
	sets the gradient function, if available.

void	set_progress_func (ProgressFunc< TParameters, TSolutions > &pfunc)
	Sets the possibly custom progress function, that is called in between every search step, and gives an outside user a simple way to witness progress of the algorithm, as well as to add custom termination criteria.

void	start_from_solution (const TSolutions &sol)
	starts optimization from a given solution object.

void	set_plot_func (PlotFunc< TParameters, TSolutions > &pffunc)
	Sets the possibly custom plot to file function, that is useful for storing into file various possibly custom variable values for each step until termination.

dVec	gradf (const dVec &x)
	returns numerical gradient of objective function at x.

dVec	gradgp (const dVec &x) const
	returns the numerical gradient of the objective function in phenotype space

double	edm ()
	computes expected distance to minimum (EDM).

TSolutions &	get_solutions ()
	returns reference to current solution object

TParameters &	get_parameters ()
	returns reference to current optimization parameters object

double	fitfunc (const double *x, const int N)
	execute objective function

void	uncertainty_handling ()
	uncertainty handling scheme that computes and uncertainty level based on a dual candidate ranking.

void	perform_uh (const dMat &candidates, const dMat &phenocandidates, int &nfcalls)
	uncertainty handling scheme that perform completely the reevaluation of solutions.

void	select_candidates_uh (const dMat &candidates, const dMat &phenocandidates, dMat &candidates_uh)
	part of the ucertainty handling scheme that select which candidates should be reevaluated.

void	eval_candidates_uh (const dMat &candidates, const dMat &candidates_uh, std::vector< RankedCandidate > &nvcandidates, int &nfcalls)
	part of the ucertainty handling scheme that evaluate the candidates to be reevaluated.

void	set_candidates_uh (const std::vector< RankedCandidate > &nvcandidates)
	part of the ucertainty handling scheme that set the results of evaluation to the solutions.

void	tpa_update ()
	updates the two-point adaptation average rank difference for the step-size adaptation mechanism

Candidate	best_solution () const

void	set_initial_elitist (const bool &e)

Static Public Attributes
static ProgressFunc< CMAParameters< TGenoPheno >, CMASolutions >	_defaultPFunc = &pfuncdef_impl<TCovarianceUpdate,TGenoPheno>

static PlotFunc< CMAParameters< TGenoPheno >, CMASolutions >	_defaultFPFunc = &fpfuncdef_impl<TCovarianceUpdate,TGenoPheno>

Protected Attributes
Eigen::EigenMultivariateNormal< double >	_esolver

CMAStopCriteria< TGenoPheno >	_stopcriteria

std::ofstream *	_fplotstream = nullptr

Protected Attributes inherited from libcmaes::ESOStrategy< TParameters, TSolutions, TStopCriteria >
FitFunc	_func

int	_nevals

int	_niter

TSolutions	_solutions

TParameters	_parameters

ProgressFunc< TParameters, TSolutions >	_pfunc

GradFunc	_gfunc = nullptr

PlotFunc< TParameters, TSolutions >	_pffunc

FitFunc	_funcaux

bool	_initial_elitist = false

Detailed Description

template<class TCovarianceUpdate, class TGenoPheno = GenoPheno<NoBoundStrategy>>
class libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >

This is an implementation of CMA-ES. It uses the reference algorithm and termination criteria of the following paper: Hansen, N. (2009). Benchmarking a BI-Population CMA-ES on the BBOB-2009 Function Testbed. Workshop Proceedings of the GECCO Genetic and Evolutionary Computation Conference, ACM, pp. 2389-2395 See https://www.lri.fr/~hansen/publications.html for more information.

Constructor & Destructor Documentation

◆ CMAStrategy() [1/2]

template<class TCovarianceUpdate , class TGenoPheno >

libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::CMAStrategy	(	FitFunc &	func,
		CMAParameters< TGenoPheno > &	parameters
	)

constructor.

Parameters

func	objective function to minimize
parameters	stochastic search parameters

◆ CMAStrategy() [2/2]

template<class TCovarianceUpdate , class TGenoPheno >

libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::CMAStrategy	(	FitFunc &	func,
		CMAParameters< TGenoPheno > &	parameters,
		const CMASolutions &	cmasols
	)

constructor for starting from an existing solution.

Parameters

func	objective function to minimize
parameters	stochastic search parameters
cmasols	solution object to start from

Member Function Documentation

◆ optimize() [1/2]

template<class TCovarianceUpdate , class TGenoPheno = GenoPheno<NoBoundStrategy>>

int libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::optimize ( )

inline

Finds the minimum of the objective function. It makes alternate calls to ask(), tell() and stop() until one of the termination criteria triggers.

Returns: success or error code, as defined in opti_err.h Note: the termination criteria code is held by _solutions._run_status

◆ optimize() [2/2]

template<class TCovarianceUpdate , class TGenoPheno >

int libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::optimize	(	const EvalFunc &	evalf,
		const AskFunc &	askf,
		const TellFunc &	tellf
	)

Finds the minimum of the objective function. It makes alternate calls to ask(), tell() and stop() until one of the termination criteria triggers.

Parameters

evalf	custom eval function
askf	custom ask function
tellf	custom tell function

Returns: success or error code, as defined in opti_err.h Note: the termination criteria code is held by _solutions._run_status

◆ stop()

template<class TCovarianceUpdate , class TGenoPheno >

bool libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::stop ( )

Stops search on a set of termination criterias, see reference paper.

Returns: true if search must stop, false otherwise.

Member Data Documentation

◆ _defaultFPFunc

template<class TCovarianceUpdate , class TGenoPheno >

PlotFunc< CMAParameters< TGenoPheno >, CMASolutions > libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::_defaultFPFunc = &fpfuncdef_impl<TCovarianceUpdate,TGenoPheno>

static

the default plot to file function.

◆ _defaultPFunc

template<class TCovarianceUpdate , class TGenoPheno >

ProgressFunc< CMAParameters< TGenoPheno >, CMASolutions > libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::_defaultPFunc = &pfuncdef_impl<TCovarianceUpdate,TGenoPheno>

static

the default progress function.

◆ _esolver

template<class TCovarianceUpdate , class TGenoPheno = GenoPheno<NoBoundStrategy>>

Eigen::EigenMultivariateNormal<double> libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::_esolver

protected

multivariate normal distribution sampler, and eigendecomposition solver.

◆ _fplotstream

template<class TCovarianceUpdate , class TGenoPheno = GenoPheno<NoBoundStrategy>>

std::ofstream* libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::_fplotstream = nullptr

protected

plotting file stream, not in parameters because of copy-constructor hell.

◆ _stopcriteria

template<class TCovarianceUpdate , class TGenoPheno = GenoPheno<NoBoundStrategy>>

CMAStopCriteria<TGenoPheno> libcmaes::CMAStrategy< TCovarianceUpdate, TGenoPheno >::_stopcriteria

protected

holds the set of termination criteria, see reference paper.

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

include/libcmaes/cmastrategy.h
src/cmastrategy.cc

Public Member Functions

Static Public Attributes

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ CMAStrategy() [1/2]

◆ CMAStrategy() [2/2]

Member Function Documentation

◆ optimize() [1/2]

◆ optimize() [2/2]

◆ stop()

Member Data Documentation

◆ _defaultFPFunc

◆ _defaultPFunc

◆ _esolver

◆ _fplotstream

◆ _stopcriteria