Implements a Limited-Memory-CMA. More...

#include <shark/Algorithms/DirectSearch/LMCMA.h>

Inheritance diagram for shark::LMCMA:

Public Member Functions
	LMCMA (random::rng_type &rng=random::globalRng)
	Default c'tor.

std::string	name () const
	From INameable: return the class name.

unsigned	suggestLambda (unsigned int dimension)
	Calculates lambda for the supplied dimensionality n.

double	suggestMu (unsigned int lambda)
	Calculates mu for the supplied lambda and the recombination strategy.

void	init (ObjectiveFunctionType const &function, SearchPointType const &p)
	Initializes the algorithm for the supplied objective function.

void	init (ObjectiveFunctionType const &function, SearchPointType const &initialSearchPoint, unsigned int lambda, double mu, double initialSigma)
	Initializes the algorithm for the supplied objective function.

void	step (ObjectiveFunctionType const &function)
	Executes one iteration of the algorithm.

double	sigma () const
	Accesses the current step size.

RealVector const &	mean () const
	Accesses the current population mean.

RealVector const &	weights () const
	Accesses the current weighting vector.

RealVector const &	evolutionPath () const
	Accesses the evolution path for the covariance matrix update.

unsigned int	mu () const
	Returns the size of the parent population \(\mu\).

unsigned int &	mu ()
	Returns a mutabl rference to the size of the parent population \(\mu\).

unsigned int	lambda () const
	Returns a immutable reference to the size of the offspring population \(\mu\).

unsigned int &	lambda ()
	Returns a mutable reference to the size of the offspring population \(\mu\).

Public Member Functions inherited from shark::AbstractSingleObjectiveOptimizer< RealVector >
std::size_t	numInitPoints () const
	By default most single objective optimizers only require a single point.

virtual void	init (ObjectiveFunctionType const &function, std::vector< SearchPointType > const &initPoints)
	Initialize the optimizer for the supplied objective function using a set of initialisation points.

virtual const SolutionType &	solution () const
	returns the current solution of the optimizer

Public Member Functions inherited from shark::AbstractOptimizer< PointType, ResultT, SolutionTypeT >
const Features &	features () const

virtual void	updateFeatures ()

bool	requiresValue () const

bool	requiresFirstDerivative () const

bool	requiresSecondDerivative () const

bool	canSolveConstrained () const

bool	requiresClosestFeasible () const

virtual	~AbstractOptimizer ()

virtual void	init (ObjectiveFunctionType const &function)
	Initialize the optimizer for the supplied objective function.

virtual void	init (ObjectiveFunctionType const &function, std::vector< SearchPointType > const &initPoints)=0
	Initialize the optimizer for the supplied objective function using a set of initialisation points.

virtual void	step (ObjectiveFunctionType const &function)=0
	Carry out one step of the optimizer for the supplied objective function.

Public Member Functions inherited from shark::INameable
virtual	~INameable ()

Public Member Functions inherited from shark::ISerializable
virtual	~ISerializable ()
	Virtual d'tor.

virtual void	read (InArchive &archive)
	Read the component from the supplied archive.

virtual void	write (OutArchive &archive) const
	Write the component to the supplied archive.

void	load (InArchive &archive, unsigned int version)
	Versioned loading of components, calls read(...).

void	save (OutArchive &archive, unsigned int version) const
	Versioned storing of components, calls write(...).

	BOOST_SERIALIZATION_SPLIT_MEMBER ()

Additional Inherited Members
Public Types inherited from shark::AbstractSingleObjectiveOptimizer< RealVector >
typedef base_type::SearchPointType	SearchPointType

typedef base_type::SolutionType	SolutionType

typedef base_type::ResultType	ResultType

typedef base_type::ObjectiveFunctionType	ObjectiveFunctionType

Public Types inherited from shark::AbstractOptimizer< PointType, ResultT, SolutionTypeT >
enum	Feature { REQUIRES_VALUE = 1 , REQUIRES_FIRST_DERIVATIVE = 2 , REQUIRES_SECOND_DERIVATIVE = 4 , CAN_SOLVE_CONSTRAINED = 8 , REQUIRES_CLOSEST_FEASIBLE = 16 }
	Models features that the optimizer requires from the objective function. More...

typedef PointType	SearchPointType

typedef ResultT	ResultType

typedef SolutionTypeT	SolutionType

typedef AbstractObjectiveFunction< PointType, ResultType >	ObjectiveFunctionType

typedef TypedFlags< Feature >	Features

typedef TypedFeatureNotAvailableException< Feature >	FeatureNotAvailableException

Protected Member Functions inherited from shark::AbstractOptimizer< PointType, ResultT, SolutionTypeT >
void	checkFeatures (ObjectiveFunctionType const &objectiveFunction)
	Convenience function that checks whether the features of the supplied objective function match with the required features of the optimizer.

Protected Attributes inherited from shark::AbstractSingleObjectiveOptimizer< RealVector >
SolutionType	m_best
	Current solution of the optimizer.

Protected Attributes inherited from shark::AbstractOptimizer< PointType, ResultT, SolutionTypeT >
Features	m_features

Detailed Description

Implements a Limited-Memory-CMA.

This is the algorithm as proposed in Ilya Loshchilov, "A Computationally Efficient Limited Memory CMA-ES for Large Scale Optimization" with a few corrections regarding the covariance matrix update.

The algorithm stores a subset of previous evolution path vectors and approximates the covariance matrix based on this. This algorithm only requires O(nm) memory, where n is the dimensionality and n the problem dimensionality. To be more exact, 2*m vectors of size n are stored to calculate the matrix-vector product with the choelsky factor of the covariance matrix in O(mn).

The algorithm uses the population based step size adaptation strategy as proposed in the same paper.

Definition at line 198 of file LMCMA.h.

Constructor & Destructor Documentation

◆ LMCMA()

shark::LMCMA::LMCMA ( random::rng_type & rng = random::globalRng )

inline

Default c'tor.

Definition at line 202 of file LMCMA.h.

References shark::AbstractOptimizer< PointType, ResultT, SolutionTypeT >::m_features, and shark::AbstractOptimizer< PointType, ResultT, SolutionTypeT >::REQUIRES_VALUE.

Member Function Documentation

◆ evolutionPath()

RealVector const & shark::LMCMA::evolutionPath ( ) const

inline

Accesses the evolution path for the covariance matrix update.

Definition at line 317 of file LMCMA.h.

◆ init() [1/2]

void shark::LMCMA::init	(	ObjectiveFunctionType const &	function,
		SearchPointType const &	initialSearchPoint,
		unsigned int	lambda,
		double	mu,
		double	initialSigma
	)

inline

Initializes the algorithm for the supplied objective function.

Definition at line 239 of file LMCMA.h.

References shark::AbstractOptimizer< PointType, ResultT, SolutionTypeT >::checkFeatures(), shark::PopulationBasedStepSizeAdaptation::init(), lambda(), shark::AbstractSingleObjectiveOptimizer< RealVector >::m_best, mu(), and shark::sqr().

◆ init() [2/2]

void shark::LMCMA::init	(	ObjectiveFunctionType const &	function,
		SearchPointType const &	p
	)

inlinevirtual

Initializes the algorithm for the supplied objective function.

Implements shark::AbstractSingleObjectiveOptimizer< RealVector >.

Definition at line 227 of file LMCMA.h.

References init(), lambda(), mu(), suggestLambda(), and suggestMu().

Referenced by init().

◆ lambda() [1/2]

unsigned int & shark::LMCMA::lambda ( )

inline

Returns a mutable reference to the size of the offspring population \(\mu\).

Definition at line 337 of file LMCMA.h.

◆ lambda() [2/2]

unsigned int shark::LMCMA::lambda ( ) const

inline

Returns a immutable reference to the size of the offspring population \(\mu\).

Definition at line 332 of file LMCMA.h.

Referenced by init(), init(), step(), suggestLambda(), and suggestMu().

◆ mean()

RealVector const & shark::LMCMA::mean ( ) const

inline

Accesses the current population mean.

Definition at line 307 of file LMCMA.h.

◆ mu() [1/2]

unsigned int & shark::LMCMA::mu ( )

inline

Returns a mutabl rference to the size of the parent population \(\mu\).

Definition at line 327 of file LMCMA.h.

◆ mu() [2/2]

unsigned int shark::LMCMA::mu ( ) const

inline

Returns the size of the parent population \(\mu\).

Definition at line 322 of file LMCMA.h.

Referenced by init(), and init().

◆ name()

std::string shark::LMCMA::name ( ) const

inlinevirtual

From INameable: return the class name.

Reimplemented from shark::INameable.

Definition at line 208 of file LMCMA.h.

◆ sigma()

double shark::LMCMA::sigma ( ) const

inline

Accesses the current step size.

Definition at line 302 of file LMCMA.h.

References shark::PopulationBasedStepSizeAdaptation::stepSize().

◆ step()

void shark::LMCMA::step ( ObjectiveFunctionType const & function )

inline

Executes one iteration of the algorithm.

Definition at line 278 of file LMCMA.h.

References lambda(), shark::AbstractSingleObjectiveOptimizer< RealVector >::m_best, and shark::searchPoint().

◆ suggestLambda()

unsigned shark::LMCMA::suggestLambda ( unsigned int dimension )

inline

Calculates lambda for the supplied dimensionality n.

Definition at line 212 of file LMCMA.h.

References lambda().

Referenced by init().

◆ suggestMu()

double shark::LMCMA::suggestMu ( unsigned int lambda )

inline

Calculates mu for the supplied lambda and the recombination strategy.

Definition at line 220 of file LMCMA.h.

References lambda().

Referenced by init().

◆ weights()

RealVector const & shark::LMCMA::weights ( ) const

inline

Accesses the current weighting vector.

Definition at line 312 of file LMCMA.h.

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

include/shark/Algorithms/DirectSearch/LMCMA.h

Public Member Functions

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ LMCMA()

Member Function Documentation

◆ evolutionPath()

◆ init() [1/2]

◆ init() [2/2]

◆ lambda() [1/2]

◆ lambda() [2/2]

◆ mean()

◆ mu() [1/2]

◆ mu() [2/2]

◆ name()

◆ sigma()

◆ step()

◆ suggestLambda()

◆ suggestMu()

◆ weights()