CMAOptimizer Class Reference

The CMA-ES as a ReClaM Optimizer. More...

#include <CMAOptimizer.h>

Inheritance diagram for CMAOptimizer:

List of all members.

Classes

class ModelFitness

Public Types

enum eMode { modeRankMuUpdate = 1, modeRankOneUpdate = 2, modeOnePlusOne = 4 }

Public Member Functions

CMAOptimizer (int verbosity=0)

Constructor.

~CMAOptimizer ()

Destructor.

void init (Model &model)

basic initialization with default parameters

void init (Model &model, double sigma, eMode mode=modeRankMuUpdate, bool best=true, int lambda=0, int mu=0)

initialization with additional parameters

void init (Model &model, const Array< double > &sigma, eMode mode=modeRankMuUpdate, bool best=true, int lambda=0, int mu=0)

initialization with additional parameters

void initUncertainty (Model &model, double sigma=0.01, unsigned int maxEvals=1000, double alpha=1.5, double theta=0.1, eMode mode=modeRankMuUpdate, int lambda=0, int mu=0)

initialization with additional parameter for CMA with uncertainty handling

void initUncertainty (Model &model, const Array< double > &sigma, unsigned int maxEvals=1000, double alpha=1.5, double theta=0.1, eMode mode=modeRankMuUpdate, int lambda=0, int mu=0)

initialization with additional parameter

double optimize (Model &model, ErrorFunction &errorfunction, const Array< double > &input, const Array< double > &target)

create and select one CMA-ES generation

int getLambda ()

return the offspring size, that is, the number of fitness evaluations per generation.

Protected Member Functions

void Ind2Model (Individual &ind, Model &model)

Protected Attributes

ModelFitness objective

bool bFirstIteration

is this the first iteration?

CMA cma

CMA object from EALib.

ElitistCMA ecma

ElitistCMA object from EALib.

Population * parents

parent population

Population * offspring

offspring population

eMode cmaMode

CMA mode.

bool returnBestIndividual

if true, always return the best known individual

double bestFitness

best fitness

Array< double > bestParameters

parameters leading to the best fitness

int verbosity

verbosity level

bool uncertaintyHandling

enable uncertainty handling?

unsigned int maxEvals

maximum number of function evaluations

double alpha

speed of strategy adaptation for uncertainty handling

double theta

uncertainty threshold parameter

Detailed Description

The CMA-ES as a ReClaM Optimizer.

For a detailed description please refer to the EALib module.

Examples:: KernelOptimization.cpp.

Definition at line 59 of file CMAOptimizer.h.

Member Enumeration Documentation

enum CMAOptimizer::eMode

Enumerator:

modeRankMuUpdate
modeRankOneUpdate
modeOnePlusOne

Definition at line 62 of file CMAOptimizer.h.

Constructor & Destructor Documentation

CMAOptimizer::CMAOptimizer ( int verbosity = 0 )

Constructor.

Definition at line 46 of file CMAOptimizer.cpp.

References offspring, and parents.

CMAOptimizer::~CMAOptimizer ( )

Destructor.

Definition at line 54 of file CMAOptimizer.cpp.

References offspring, and parents.

Member Function Documentation

int CMAOptimizer::getLambda ( )

return the offspring size, that is, the number of fitness evaluations per generation.

Definition at line 283 of file CMAOptimizer.cpp.

References offspring.

void CMAOptimizer::Ind2Model	(	Individual &	ind,
		Model &	model
	)			`[protected]`

Definition at line 288 of file CMAOptimizer.cpp.

References Model::getParameterDimension(), i, and Model::setParameter().

Referenced by optimize().

void CMAOptimizer::init	(	Model &	model,
		const Array< double > &	sigma,
		eMode	mode = `modeRankMuUpdate`,
		bool	best = `true`,
		int	lambda = `0`,
		int	mu = `0`
	)

initialization with additional parameters

Parameters:

	model	model to optimize
	sigma	initial step size for each coordinate
	mode	CMA mode, see EALib for details
	best	if true, return the best individual ever evaluated
	lambda	number of offspring per generation, a value of 0 indicates the default depending on the problem dimension - not used in 1+1-mode
	mu	number of parents, a value of 0 indicates the default depending on lambda - not used in 1+1-mode

Definition at line 73 of file CMAOptimizer.cpp.

References bestFitness, bestParameters, bFirstIteration, cma, cmaMode, ecma, Model::getParameter(), Model::getParameterDimension(), i, modeOnePlusOne, modeRankMuUpdate, modeRankOneUpdate, offspring, parents, returnBestIndividual, uncertaintyHandling, and verbosity.

void CMAOptimizer::init	(	Model &	model,
		double	sigma,
		eMode	mode = `modeRankMuUpdate`,
		bool	best = `true`,
		int	lambda = `0`,
		int	mu = `0`
	)

initialization with additional parameters

Parameters:

	model	model to optimize
	sigma	initial step size
	mode	CMA mode, see EALib for details
	best	if true, return the best individual ever evaluated
	lambda	number of offspring per generation, a value of 0 indicates the default depending on the problem dimension - not used in 1+1-mode
	mu	number of parents, a value of 0 indicates the default depending on lambda - not used in 1+1-mode

Definition at line 66 of file CMAOptimizer.cpp.

References Model::getParameterDimension(), and init().

void CMAOptimizer::init ( Model & model ) [virtual]

basic initialization with default parameters

Parameters:

model

model to optimize

Implements Optimizer.

Definition at line 61 of file CMAOptimizer.cpp.

Referenced by init(), and initUncertainty().

void CMAOptimizer::initUncertainty	(	Model &	model,
		const Array< double > &	sigma,
		unsigned int	maxEvals = `1000`,
		double	alpha = `1.5`,
		double	theta = `0.1`,
		eMode	mode = `modeRankMuUpdate`,
		int	lambda = `0`,
		int	mu = `0`
	)

initialization with additional parameter

Parameters:

	model	model to optimize
	sigma	initial step size
	maxEvals	maximum number of function evaluation for one fitness computation
	alpha	speed of adaptation of number of function evaluations
	theta	uncertainty threshold
	mode	CMA mode, see EALib for details
	lambda	number of offspring per generation, a value of 0 indicates the default depending on the problem dimension - not used in 1+1-mode
	mu	number of parents, a value of 0 indicates the default depending on lambda - not used in 1+1-mode

Definition at line 146 of file CMAOptimizer.cpp.

References init(), objective, and uncertaintyHandling.

void CMAOptimizer::initUncertainty	(	Model &	model,
		double	sigma = `0.01`,
		unsigned int	maxEvals = `1000`,
		double	alpha = `1.5`,
		double	theta = `0.1`,
		eMode	mode = `modeRankMuUpdate`,
		int	lambda = `0`,
		int	mu = `0`
	)

initialization with additional parameter for CMA with uncertainty handling

Parameters:

	model	model to optimize
	sigma	initial step size
	maxEvals	maximum number of function evaluation for one fitness computation
	alpha	speed of adaptation of number of function evaluations
	theta	uncertainty threshold
	mode	CMA mode, see EALib for details
	lambda	number of offspring per generation, a value of 0 indicates the default depending on the problem dimension - not used in 1+1-mode
	mu	number of parents, a value of 0 indicates the default depending on lambda - not used in 1+1-mode

Definition at line 133 of file CMAOptimizer.cpp.

References init(), objective, and uncertaintyHandling.

double CMAOptimizer::optimize	(	Model &	model,
		ErrorFunction &	errorfunction,
		const Array< double > &	input,
		const Array< double > &	target
	)			`[virtual]`

create and select one CMA-ES generation

Implements Optimizer.

Definition at line 159 of file CMAOptimizer.cpp.

References alpha, bestFitness, bestParameters, bFirstIteration, cma, cmaMode, ecma, CMAOptimizer::ModelFitness::fitness(), Model::getParameterDimension(), i, Ind2Model(), Model::isFeasible(), maxEvals, modeOnePlusOne, modeRankMuUpdate, modeRankOneUpdate, objective, offspring, parents, returnBestIndividual, CMAOptimizer::ModelFitness::Set(), Model::setParameter(), theta, uncertaintyHandling, and verbosity.

Member Data Documentation

double CMAOptimizer::alpha [protected]

speed of strategy adaptation for uncertainty handling

Definition at line 184 of file CMAOptimizer.h.

Referenced by optimize().

double CMAOptimizer::bestFitness [protected]

best fitness

Definition at line 169 of file CMAOptimizer.h.

Referenced by init(), and optimize().

Array<double> CMAOptimizer::bestParameters [protected]

parameters leading to the best fitness

Definition at line 172 of file CMAOptimizer.h.

Referenced by init(), and optimize().

bool CMAOptimizer::bFirstIteration [protected]

is this the first iteration?

Definition at line 148 of file CMAOptimizer.h.

Referenced by init(), and optimize().

CMA CMAOptimizer::cma [protected]

CMA object from EALib.

Definition at line 151 of file CMAOptimizer.h.

Referenced by init(), and optimize().

eMode CMAOptimizer::cmaMode [protected]

CMA mode.

Definition at line 163 of file CMAOptimizer.h.

Referenced by init(), and optimize().

ElitistCMA CMAOptimizer::ecma [protected]

ElitistCMA object from EALib.

Definition at line 154 of file CMAOptimizer.h.

Referenced by init(), and optimize().

unsigned int CMAOptimizer::maxEvals [protected]

maximum number of function evaluations

Definition at line 181 of file CMAOptimizer.h.

Referenced by optimize().

ModelFitness CMAOptimizer::objective [protected]

Definition at line 143 of file CMAOptimizer.h.

Referenced by initUncertainty(), and optimize().

Population* CMAOptimizer::offspring [protected]

offspring population

Definition at line 160 of file CMAOptimizer.h.

Referenced by CMAOptimizer(), getLambda(), init(), optimize(), and ~CMAOptimizer().

Population* CMAOptimizer::parents [protected]

parent population

Definition at line 157 of file CMAOptimizer.h.

Referenced by CMAOptimizer(), init(), optimize(), and ~CMAOptimizer().

bool CMAOptimizer::returnBestIndividual [protected]

if true, always return the best known individual

Definition at line 166 of file CMAOptimizer.h.

Referenced by init(), and optimize().

double CMAOptimizer::theta [protected]

uncertainty threshold parameter

Definition at line 187 of file CMAOptimizer.h.

Referenced by optimize().

bool CMAOptimizer::uncertaintyHandling [protected]

enable uncertainty handling?

Definition at line 178 of file CMAOptimizer.h.

Referenced by init(), initUncertainty(), and optimize().

int CMAOptimizer::verbosity [protected]

verbosity level

Definition at line 175 of file CMAOptimizer.h.

Referenced by init(), and optimize().

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


Classes
class	ModelFitness
Public Types
enum	eMode { modeRankMuUpdate = 1, modeRankOneUpdate = 2, modeOnePlusOne = 4 }
Public Member Functions
	CMAOptimizer (int verbosity=0)
	Constructor.
	~CMAOptimizer ()
	Destructor.
void	init (Model &model)
	basic initialization with default parameters
void	init (Model &model, double sigma, eMode mode=modeRankMuUpdate, bool best=true, int lambda=0, int mu=0)
	initialization with additional parameters
void	init (Model &model, const Array< double > &sigma, eMode mode=modeRankMuUpdate, bool best=true, int lambda=0, int mu=0)
	initialization with additional parameters
void	initUncertainty (Model &model, double sigma=0.01, unsigned int maxEvals=1000, double alpha=1.5, double theta=0.1, eMode mode=modeRankMuUpdate, int lambda=0, int mu=0)
	initialization with additional parameter for CMA with uncertainty handling
void	initUncertainty (Model &model, const Array< double > &sigma, unsigned int maxEvals=1000, double alpha=1.5, double theta=0.1, eMode mode=modeRankMuUpdate, int lambda=0, int mu=0)
	initialization with additional parameter
double	optimize (Model &model, ErrorFunction &errorfunction, const Array< double > &input, const Array< double > &target)
	create and select one CMA-ES generation
int	getLambda ()
	return the offspring size, that is, the number of fitness evaluations per generation.
Protected Member Functions
void	Ind2Model (Individual &ind, Model &model)
Protected Attributes
ModelFitness	objective
bool	bFirstIteration
	is this the first iteration?
CMA	cma
	CMA object from EALib.
ElitistCMA	ecma
	ElitistCMA object from EALib.
Population *	parents
	parent population
Population *	offspring
	offspring population
eMode	cmaMode
	CMA mode.
bool	returnBestIndividual
	if true, always return the best known individual
double	bestFitness
	best fitness
Array< double >	bestParameters
	parameters leading to the best fitness
int	verbosity
	verbosity level
bool	uncertaintyHandling
	enable uncertainty handling?
unsigned int	maxEvals
	maximum number of function evaluations
double	alpha
	speed of strategy adaptation for uncertainty handling
double	theta
	uncertainty threshold parameter