Levenberg Marquardt optimizer class. More...

#include <GOptimizerLM.hpp>

Inheritance diagram for GOptimizerLM:

Public Member Functions
	GOptimizerLM (void)
	Void constructor. More...

	GOptimizerLM (GLog *log)
	Constructor with logger. More...

	GOptimizerLM (const GOptimizerLM &opt)
	Copy constructor. More...

virtual	~GOptimizerLM (void)
	Destructor. More...

GOptimizerLM &	operator= (const GOptimizerLM &opt)
	Assignment operator. More...

virtual void	clear (void)
	Clear object. More...

virtual GOptimizerLM *	clone (void) const
	Clone object. More...

virtual std::string	classname (void) const
	Return class name. More...

virtual void	optimize (GOptimizerFunction &fct, GOptimizerPars &pars)
	Optimize function parameters. More...

virtual void	errors (GOptimizerFunction &fct, GOptimizerPars &pars)
	Compute parameter uncertainties. More...

virtual double	value (void) const
	Return function value. More...

virtual int	status (void) const
	Return optimizer status. More...

virtual int	iter (void) const
	Return number of iterations. More...

virtual std::string	print (const GChatter &chatter=NORMAL) const
	Print optimizer information. More...

void	logger (GLog *log)
	Set logger. More...

void	max_iter (const int &max_iter)
	Set maximum number of iterations. More...

void	max_stalls (const int &max_stalls)
	Set maximum number of allowed subsequent stalls. More...

void	max_boundary_hits (const int &max_hit)
	Set maximum number of parameter boundary hits. More...

void	lambda_start (const double &value)
	Set lambda starting value. More...

void	lambda_inc (const double &value)
	Set lambda increment value. More...

void	lambda_dec (const double &value)
	Set lambda decrement value. More...

void	eps (const double &eps)
	Set requested absolute convergence precision. More...

void	accept_dec (const double &value)
	Set acceptable function decrease. More...

const int &	npars (void) const
	Return number of model parameters. More...

const int &	nfree (void) const
	Return number of free model parameters. More...

const int &	max_iter (void) const
	Return maximum number of iterations. More...

const int &	max_stalls (void) const
	Return maximum number of allowed subsequent stalls. More...

const int &	max_boundary_hits (void) const
	Return maximum number of parameter boundary hits. More...

const double &	lambda_start (void) const
	Return lambda starting value. More...

const double &	lambda_inc (void) const
	Return lambda increment value. More...

const double &	lambda_dec (void) const
	Return lambda decrement value. More...

const double &	lambda (void) const
	Return lambda value. More...

const double &	eps (void) const
	Return requested absolute convergence precision. More...

const double &	accept_dec (void) const
	Return acceptable function decrease. More...

std::string	status_string (void) const
	Set fit status string. More...

Public Member Functions inherited from GOptimizer
	GOptimizer (void)
	Constructor. More...

	GOptimizer (const GOptimizer &opt)
	Copy constructor. More...

virtual	~GOptimizer (void)
	Destructor. More...

virtual GOptimizer &	operator= (const GOptimizer &opt)
	Assignment operator. More...

Public Member Functions inherited from GBase
virtual	~GBase (void)
	Destructor. More...

Protected Member Functions
void	init_members (void)
	Initialise class members. More...

void	copy_members (const GOptimizerLM &opt)
	Copy class members. More...

void	free_members (void)
	Delete class members. More...

double	iteration (GOptimizerFunction &fct, GOptimizerPars &pars)
	Perform one LM iteration. More...

double	step_size (const GVector &grad, const GOptimizerPars &pars)
	Return LM step size. More...

Protected Member Functions inherited from GOptimizer
void	init_members (void)
	Initialise class members. More...

void	copy_members (const GOptimizer &opt)
	Copy class members. More...

void	free_members (void)
	Delete class members. More...

Protected Attributes
int	m_npars
	Number of parameters. More...

int	m_nfree
	Number of free parameters. More...

double	m_lambda_start
	Initial start value. More...

double	m_lambda_inc
	Lambda increase. More...

double	m_lambda_dec
	Lambda decrease. More...

double	m_eps
	Absolute precision. More...

double	m_accept_dec
	Acceptable function decrease. More...

int	m_max_iter
	Maximum number of iterations. More...

int	m_max_stall
	Maximum number of stalls. More...

int	m_max_hit
	Maximum number of successive hits. More...

int	m_max_dec
	Maximum number of function decrease. More...

bool	m_step_adjust
	Adjust step size to boundaries. More...

std::vector< bool >	m_hit_boundary
	Bookkeeping array for boundary hits. More...

std::vector< int >	m_hit_minimum
	Bookkeeping of successive minimum hits. More...

std::vector< int >	m_hit_maximum
	Bookkeeping of successive maximum hits. More...

std::vector< bool >	m_par_freeze
	Bookkeeping of parameter freeze. More...

std::vector< bool >	m_par_remove
	Bookkeeping of parameter removal. More...

double	m_lambda
	Actual lambda. More...

double	m_value
	Actual function value. More...

double	m_delta
	Function improvement. More...

int	m_status
	Fit status. More...

int	m_iter
	Iteration. More...

int	m_num_dec
	Number of function decreases. More...

GLog *	m_logger
	Pointer to optional logger. More...

Detailed Description

Levenberg Marquardt optimizer class.

This method implements an Levenberg Marquardt optimizer.

Definition at line 51 of file GOptimizerLM.hpp.

Constructor & Destructor Documentation

GOptimizerLM::GOptimizerLM ( void )

Void constructor.

Definition at line 56 of file GOptimizerLM.cpp.

References init_members().

Referenced by clone().

GOptimizerLM::GOptimizerLM ( GLog * log )

explicit

Constructor with logger.

Parameters

[in] log Logger to use in optimizer.

Definition at line 71 of file GOptimizerLM.cpp.

References init_members(), log(), and m_logger.

GOptimizerLM::GOptimizerLM ( const GOptimizerLM & opt )

Copy constructor.

Parameters

[in] opt Optimizer from which the instance should be built.

Definition at line 89 of file GOptimizerLM.cpp.

References copy_members(), and init_members().

GOptimizerLM::~GOptimizerLM ( void )

virtual

Destructor.

Definition at line 105 of file GOptimizerLM.cpp.

References free_members().

Member Function Documentation

void GOptimizerLM::accept_dec ( const double & value )

inline

Set acceptable function decrease.

Parameters

[in] value Acceptable function decrease.

Sets the acceptable function decrease value for which the new solution will be kept and the iterations continue. This strategy provides better convergence in case that a function decrease is encountered.

Definition at line 299 of file GOptimizerLM.hpp.

References m_accept_dec, and value().

const double & GOptimizerLM::accept_dec ( void ) const

inline

Return acceptable function decrease.

Returns: Acceptable function decrease.

Definition at line 432 of file GOptimizerLM.hpp.

References m_accept_dec.

std::string GOptimizerLM::classname ( void ) const

inlinevirtual

Return class name.

Returns: String containing the class name ("GOptimizerLM").

Implements GOptimizer.

Definition at line 141 of file GOptimizerLM.hpp.

void GOptimizerLM::clear ( void )

virtual

Clear object.

This method properly resets the object to an initial state.

Implements GOptimizer.

Definition at line 161 of file GOptimizerLM.cpp.

References GOptimizer::free_members(), free_members(), GOptimizer::init_members(), and init_members().

GOptimizerLM * GOptimizerLM::clone ( void ) const

virtual

Clone object.

Implements GOptimizer.

Definition at line 179 of file GOptimizerLM.cpp.

References GOptimizerLM().

void GOptimizerLM::copy_members ( const GOptimizerLM & opt )

protected

Copy class members.

Parameters

[in] opt GOptimizerLM members to be copied.

Definition at line 687 of file GOptimizerLM.cpp.

References m_accept_dec, m_delta, m_eps, m_hit_boundary, m_hit_maximum, m_hit_minimum, m_iter, m_lambda, m_lambda_dec, m_lambda_inc, m_lambda_start, m_logger, m_max_dec, m_max_iter, m_max_stall, m_nfree, m_npars, m_num_dec, m_par_freeze, m_par_remove, m_status, m_step_adjust, and m_value.

Referenced by GOptimizerLM(), and operator=().

void GOptimizerLM::eps ( const double & eps )

inline

Set requested absolute convergence precision.

Parameters

[in] eps Requested absolute convergence precision.

Definition at line 282 of file GOptimizerLM.hpp.

References eps(), and m_eps.

Referenced by GCOMDris::vetorate_fit().

const double & GOptimizerLM::eps ( void ) const

inline

Return requested absolute convergence precision.

Returns: Requested absolute convergence precision.

Definition at line 420 of file GOptimizerLM.hpp.

References m_eps.

Referenced by eps().

void GOptimizerLM::errors	(	GOptimizerFunction &	fct,
		GOptimizerPars &	pars
	)

virtual

Compute parameter uncertainties.

Parameters

[in]	fct	Optimizer function.
[in]	pars	Function parameters.

Compute parameter uncertainties from the diagonal elements of the curvature matrix.

Implements GOptimizer.

Definition at line 454 of file GOptimizerLM.cpp.

References GMatrixSparse::cholesky_decompose(), GMatrixSparse::cholesky_solver(), GOptimizerFunction::curvature(), GOptimizerFunction::eval(), G_LM_BAD_ERRORS, G_LM_NOT_POSTIVE_DEFINITE, m_logger, m_status, m_value, npars(), GOptimizerPars::size(), sqrt(), and GOptimizerFunction::value().

Referenced by GCOMDris::vetorate_fit().

void GOptimizerLM::free_members ( void )

protected

Delete class members.

Definition at line 722 of file GOptimizerLM.cpp.

Referenced by clear(), operator=(), and ~GOptimizerLM().

void GOptimizerLM::init_members ( void )

protected

Initialise class members.

< Changed on 30/10/2014 from 1.0e-6

< Do not allow to decrease

< Changed on 30/10/2014 from 1000

< Maximum successive boundary hits before freeze

< Maximum number of function decreases

Definition at line 643 of file GOptimizerLM.cpp.

References m_accept_dec, m_delta, m_eps, m_hit_boundary, m_hit_maximum, m_hit_minimum, m_iter, m_lambda, m_lambda_dec, m_lambda_inc, m_lambda_start, m_logger, m_max_dec, m_max_hit, m_max_iter, m_max_stall, m_nfree, m_npars, m_num_dec, m_par_freeze, m_par_remove, m_status, m_step_adjust, and m_value.

Referenced by clear(), GOptimizerLM(), and operator=().

int GOptimizerLM::iter ( void ) const

inlinevirtual

Return number of iterations.

Returns: Number of iterations.

Implements GOptimizer.

Definition at line 177 of file GOptimizerLM.hpp.

References m_iter.

Referenced by GCOMDris::vetorate_fit().

double GOptimizerLM::iteration	(	GOptimizerFunction &	fct,
		GOptimizerPars &	pars
	)

protected

Perform one LM iteration.

Parameters

[in]	fct	Optimizer function.
[in]	pars	Function parameters.

Returns: Step size for iteration.

This method performs one LM iteration. Note that the method only acts on the parameter value, i.e. it does not worry about the true scale of the parameter. It calls the eval() method of the optimizer function which is assumed to return the gradient and curvature matrix with respect to the parameter values (and not the scaled true values).

Definition at line 742 of file GOptimizerLM.cpp.

References abs(), GOptimizerFunction::curvature(), GOptimizerFunction::eval(), G_LM_NOT_POSTIVE_DEFINITE, GOptimizerFunction::gradient(), m_accept_dec, m_delta, m_hit_maximum, m_hit_minimum, m_lambda, m_lambda_dec, m_lambda_inc, m_logger, m_max_dec, m_max_hit, m_npars, m_num_dec, m_par_freeze, m_par_remove, m_status, m_value, GMatrixSparse::solve(), step_size(), and GOptimizerFunction::value().

Referenced by optimize().

const double & GOptimizerLM::lambda ( void ) const

inline

Return lambda value.

Returns: Lambda value.

Definition at line 408 of file GOptimizerLM.hpp.

References m_lambda.

void GOptimizerLM::lambda_dec ( const double & value )

inline

Set lambda decrement value.

Parameters

[in] value Lambda decrement value.

Definition at line 269 of file GOptimizerLM.hpp.

References m_lambda_dec, and value().

const double & GOptimizerLM::lambda_dec ( void ) const

inline

Return lambda decrement value.

Returns: Lambda decrement value.

Definition at line 396 of file GOptimizerLM.hpp.

References m_lambda_dec.

void GOptimizerLM::lambda_inc ( const double & value )

inline

Set lambda increment value.

Parameters

[in] value Lambda increment value.

Definition at line 256 of file GOptimizerLM.hpp.

References m_lambda_inc, and value().

const double & GOptimizerLM::lambda_inc ( void ) const

inline

Return lambda increment value.

Returns: Lambda increment value.

Definition at line 384 of file GOptimizerLM.hpp.

References m_lambda_inc.

Referenced by optimize().

void GOptimizerLM::lambda_start ( const double & value )

inline

Set lambda starting value.

Parameters

[in] value Lambda starting value.

Definition at line 243 of file GOptimizerLM.hpp.

References m_lambda_start, and value().

const double & GOptimizerLM::lambda_start ( void ) const

inline

Return lambda starting value.

Returns: Lambda starting value.

Definition at line 372 of file GOptimizerLM.hpp.

References m_lambda_start.

void GOptimizerLM::logger ( GLog * log )

inline

Set logger.

Parameters

[in] log Logger to use in optimizer.

Set the logger into which the optimizer will dump any output.

Definition at line 191 of file GOptimizerLM.hpp.

References log(), and m_logger.

void GOptimizerLM::max_boundary_hits ( const int & max_hit )

inline

Set maximum number of parameter boundary hits.

Parameters

[in] max_hit Maximum number of parameter boundary hits.

Definition at line 230 of file GOptimizerLM.hpp.

References m_max_hit.

const int & GOptimizerLM::max_boundary_hits ( void ) const

inline

Return maximum number of parameter boundary hits.

Returns: Maximum number of parameter boundary hits.

Definition at line 360 of file GOptimizerLM.hpp.

References m_max_hit.

void GOptimizerLM::max_iter ( const int & max_iter )

inline

Set maximum number of iterations.

Parameters

[in] max_iter Maximum number of iterations.

Definition at line 204 of file GOptimizerLM.hpp.

References m_max_iter, and max_iter().

Referenced by GCOMDris::vetorate_fit().

const int & GOptimizerLM::max_iter ( void ) const

inline

Return maximum number of iterations.

Returns: Maximum number of iterations.

Definition at line 336 of file GOptimizerLM.hpp.

References m_max_iter.

Referenced by max_iter().

void GOptimizerLM::max_stalls ( const int & max_stalls )

inline

Set maximum number of allowed subsequent stalls.

Parameters

[in] max_stalls Maximum number of allowed subsequent stalls.

Definition at line 217 of file GOptimizerLM.hpp.

References m_max_stall, and max_stalls().

const int & GOptimizerLM::max_stalls ( void ) const

inline

Return maximum number of allowed subsequent stalls.

Returns: Maximum number of allowed subsequent stalls.

Definition at line 348 of file GOptimizerLM.hpp.

References m_max_stall.

Referenced by max_stalls().

const int & GOptimizerLM::nfree ( void ) const

inline

Return number of free model parameters.

Returns: Number of free model parameters.

Definition at line 324 of file GOptimizerLM.hpp.

References m_nfree.

const int & GOptimizerLM::npars ( void ) const

inline

Return number of model parameters.

Returns: Number of model parameters.

Definition at line 312 of file GOptimizerLM.hpp.

References m_npars.

Referenced by errors().

GOptimizerLM & GOptimizerLM::operator= ( const GOptimizerLM & opt )

Assignment operator.

Parameters

[in] opt Optimizer to be assigned.

Definition at line 126 of file GOptimizerLM.cpp.

References copy_members(), free_members(), init_members(), and GOptimizer::operator=().

void GOptimizerLM::optimize	(	GOptimizerFunction &	fct,
		GOptimizerPars &	pars
	)

virtual

Optimize function parameters.

Parameters

[in]	fct	Optimization function.
[in]	pars	Function parameters.

Implements GOptimizer.

Definition at line 191 of file GOptimizerLM.cpp.

References abs(), GOptimizerFunction::curvature(), GOptimizerFunction::eval(), G_LM_CONVERGED, G_LM_STALLED, GOptimizerFunction::gradient(), iteration(), lambda_inc(), m_delta, m_eps, m_hit_boundary, m_hit_maximum, m_hit_minimum, m_iter, m_lambda, m_lambda_start, m_logger, m_max_iter, m_max_stall, m_nfree, m_npars, m_num_dec, m_par_freeze, m_par_remove, m_status, m_value, GOptimizerPars::nfree(), GOptimizerPars::size(), status(), gammalib::str(), and GOptimizerFunction::value().

Referenced by GCOMDris::vetorate_fit().

std::string GOptimizerLM::print ( const GChatter & chatter = NORMAL ) const

virtual

Print optimizer information.

Parameters

[in] chatter Chattiness.

Returns: String containing optimizer information.

Implements GOptimizer.

Definition at line 594 of file GOptimizerLM.cpp.

References abs(), m_accept_dec, m_eps, m_iter, m_lambda, m_nfree, m_npars, m_value, gammalib::parformat(), SILENT, status_string(), and gammalib::str().

int GOptimizerLM::status ( void ) const

inlinevirtual

Return optimizer status.

Returns: Optimizer status.

Implements GOptimizer.

Definition at line 165 of file GOptimizerLM.hpp.

References m_status.

Referenced by optimize(), and status_string().

std::string GOptimizerLM::status_string ( void ) const

Set fit status string.

Returns: Fit status string.

Definition at line 556 of file GOptimizerLM.cpp.

References G_LM_BAD_ERRORS, G_LM_CONVERGED, G_LM_NOT_POSTIVE_DEFINITE, G_LM_SINGULAR, G_LM_STALLED, m_status, and status().

Referenced by print(), and GCOMDris::vetorate_fit().

double GOptimizerLM::step_size	(	const GVector &	grad,
		const GOptimizerPars &	pars
	)

protected

Return LM step size.

Parameters

[in]	grad	Function gradient.
[in]	pars	Function parameters.

Determine the size of the LM step. By default a step size of 1 is taken. If m_step_adjust=true then the step size will be estimated so that the next parameter vector should stay within the parameter boundaries (provided that boundaries exist).

Definition at line 1015 of file GOptimizerLM.cpp.

References m_hit_boundary, m_logger, m_step_adjust, and GOptimizerPars::size().

Referenced by iteration().

double GOptimizerLM::value ( void ) const

inlinevirtual

Return function value.