#include <levenberg_marquardt_algorithm.h>

Inheritance diagram for OpenNN::LevenbergMarquardtAlgorithm:

Classes
struct	LevenbergMarquardtAlgorithmResults

Public Member Functions
	LevenbergMarquardtAlgorithm (void)

	LevenbergMarquardtAlgorithm (PerformanceFunctional *)

	LevenbergMarquardtAlgorithm (const tinyxml2::XMLDocument &)

virtual	~LevenbergMarquardtAlgorithm (void)

const double &	get_warning_parameters_norm (void) const

const double &	get_warning_gradient_norm (void) const

const double &	get_error_parameters_norm (void) const

const double &	get_error_gradient_norm (void) const

const double &	get_minimum_parameters_increment_norm (void) const

const double &	get_minimum_performance_increase (void) const

const double &	get_performance_goal (void) const

const double &	get_gradient_norm_goal (void) const

const size_t &	get_maximum_generalization_performance_decreases (void) const

const size_t &	get_maximum_iterations_number (void) const

const double &	get_maximum_time (void) const

const bool &	get_reserve_parameters_history (void) const

const bool &	get_reserve_parameters_norm_history (void) const

const bool &	get_reserve_performance_history (void) const

const bool &	get_reserve_gradient_history (void) const

const bool &	get_reserve_gradient_norm_history (void) const

const bool &	get_reserve_Hessian_approximation_history (void) const

const bool &	get_reserve_generalization_performance_history (void) const

const bool &	get_reserve_elapsed_time_history (void) const

const double &	get_damping_parameter (void) const

const double &	get_damping_parameter_factor (void) const

const double &	get_minimum_damping_parameter (void) const

const double &	get_maximum_damping_parameter (void) const

const bool &	get_reserve_damping_parameter_history (void) const

const Vector< double > &	get_damping_parameter_history (void) const

void	set_default (void)

void	set_damping_parameter (const double &)

void	set_damping_parameter_factor (const double &)

void	set_minimum_damping_parameter (const double &)

void	set_maximum_damping_parameter (const double &)

void	set_reserve_damping_parameter_history (const bool &)

void	set_warning_parameters_norm (const double &)

void	set_warning_gradient_norm (const double &)

void	set_error_parameters_norm (const double &)

void	set_error_gradient_norm (const double &)

void	set_minimum_parameters_increment_norm (const double &)

void	set_minimum_performance_increase (const double &)

void	set_performance_goal (const double &)

void	set_gradient_norm_goal (const double &)

void	set_maximum_generalization_performance_decreases (const size_t &)

void	set_maximum_iterations_number (const size_t &)

void	set_maximum_time (const double &)

void	set_reserve_parameters_history (const bool &)

void	set_reserve_parameters_norm_history (const bool &)

void	set_reserve_performance_history (const bool &)

void	set_reserve_gradient_history (const bool &)

void	set_reserve_gradient_norm_history (const bool &)

void	set_reserve_Hessian_approximation_history (const bool &)

void	set_reserve_generalization_performance_history (const bool &)

void	set_reserve_elapsed_time_history (const bool &)

virtual void	set_reserve_all_training_history (const bool &)

void	set_display_period (const size_t &)

void	check (void) const

double	calculate_performance (const Vector< double > &) const

Vector< double >	calculate_gradient (const Vector< double > &, const Matrix< double > &) const

Matrix< double >	calculate_Hessian_approximation (const Matrix< double > &) const

LevenbergMarquardtAlgorithmResults *	perform_training (void)

std::string	write_training_algorithm_type (void) const

Matrix< std::string >	to_string_matrix (void) const

tinyxml2::XMLDocument *	to_XML (void) const

void	from_XML (const tinyxml2::XMLDocument &)

Vector< double >	perform_Householder_QR_decomposition (const Matrix< double > &, const Vector< double > &) const

Public Member Functions inherited from OpenNN::TrainingAlgorithm
	TrainingAlgorithm (void)

	TrainingAlgorithm (PerformanceFunctional *)

	TrainingAlgorithm (const tinyxml2::XMLDocument &)

virtual	~TrainingAlgorithm (void)

virtual TrainingAlgorithm &	operator= (const TrainingAlgorithm &)

virtual bool	operator== (const TrainingAlgorithm &) const

PerformanceFunctional *	get_performance_functional_pointer (void) const

bool	has_performance_functional (void) const

const bool &	get_display (void) const

const size_t &	get_display_period (void) const

const size_t &	get_save_period (void) const

const std::string &	get_neural_network_file_name (void) const

void	set (void)

void	set (PerformanceFunctional *)

virtual void	set_performance_functional_pointer (PerformanceFunctional *)

void	set_display (const bool &)

void	set_display_period (const size_t &)

void	set_save_period (const size_t &)

void	set_neural_network_file_name (const std::string &)

virtual std::string	to_string (void) const

void	print (void) const

void	save (const std::string &) const

void	load (const std::string &)

virtual void	initialize_random (void)

Private Attributes
double	damping_parameter

double	minimum_damping_parameter

double	maximum_damping_parameter

double	damping_parameter_factor

bool	reserve_damping_parameter_history

Vector< double >	damping_parameter_history

double	warning_parameters_norm

double	warning_gradient_norm

double	error_parameters_norm

double	error_gradient_norm

double	minimum_parameters_increment_norm

double	minimum_performance_increase

double	performance_goal

double	gradient_norm_goal

size_t	maximum_generalization_performance_decreases

size_t	maximum_iterations_number

double	maximum_time

bool	reserve_parameters_history

bool	reserve_parameters_norm_history

bool	reserve_performance_history

bool	reserve_gradient_history

bool	reserve_gradient_norm_history

bool	reserve_Hessian_approximation_history

bool	reserve_elapsed_time_history

bool	reserve_generalization_performance_history

Additional Inherited Members
Protected Attributes inherited from OpenNN::TrainingAlgorithm
PerformanceFunctional *	performance_functional_pointer

size_t	display_period

size_t	save_period

std::string	neural_network_file_name

bool	display

Detailed Description

This concrete class represents a Levenberg-Marquardt Algorithm training algorithm for the sum squared error performance functional for a multilayer perceptron.

Definition at line 43 of file levenberg_marquardt_algorithm.h.

Constructor & Destructor Documentation

OpenNN::LevenbergMarquardtAlgorithm::LevenbergMarquardtAlgorithm ( void )

explicit

Default constructor. It creates a Levenberg-Marquardt training algorithm object not associated to any performance functional object. It also initializes the class members to their default values.

Definition at line 27 of file levenberg_marquardt_algorithm.cpp.

OpenNN::LevenbergMarquardtAlgorithm::LevenbergMarquardtAlgorithm ( PerformanceFunctional * new_performance_functional_pointer )

explicit

Performance functional constructor. It creates a Levenberg-Marquardt training algorithm object associated associated with a given performance functional object. It also initializes the class members to their default values.

Parameters

new_performance_functional_pointer Pointer to an external performance functional object.

Definition at line 41 of file levenberg_marquardt_algorithm.cpp.

OpenNN::LevenbergMarquardtAlgorithm::LevenbergMarquardtAlgorithm ( const tinyxml2::XMLDocument & document )

explicit

XML Constructor. Creates a Levenberg-Marquardt algorithm object, and loads its members from a XML document.

Parameters

document TinyXML document containing the Levenberg-Marquardt algorithm data.

Definition at line 66 of file levenberg_marquardt_algorithm.cpp.

OpenNN::LevenbergMarquardtAlgorithm::~LevenbergMarquardtAlgorithm ( void )

virtual

Destructor. This destructor does not delete any object.

Definition at line 80 of file levenberg_marquardt_algorithm.cpp.

Member Function Documentation

Vector< double > OpenNN::LevenbergMarquardtAlgorithm::calculate_gradient	(	const Vector< double > &	terms,
		const Matrix< double > &	terms_Jacobian
	)		const

Returns the exact gradient vector of the performance function as a function of the terms vector and the terms Jacobian matrix.

Parameters

terms	Vector with the performance terms values.
terms_Jacobian	Jacobian matrix of the performance terms function.

Definition at line 993 of file levenberg_marquardt_algorithm.cpp.

Matrix< double > OpenNN::LevenbergMarquardtAlgorithm::calculate_Hessian_approximation ( const Matrix< double > & terms_Jacobian ) const

Returns an approximation of the Hessian matrix of the performance function as a function of the performance terms Jacobian.

Parameters

terms_Jacobian Jacobian matrix of the terms function.

Definition at line 1028 of file levenberg_marquardt_algorithm.cpp.

double OpenNN::LevenbergMarquardtAlgorithm::calculate_performance ( const Vector< double > & terms ) const

Evaluates the performance function from the evaluation of the terms function.

Parameters

terms Vector of performance terms.

Definition at line 980 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::check ( void ) const

virtual

Checks that the Levenberg-Marquard object is ok for training. In particular, it checks that:

The performance functional pointer associated to the training algorithm is not NULL,
The neural network associated to that performance functional is neither NULL.
The data set associated to that performance functional is neither NULL.

If that checkings are not hold, an exception is thrown.

Reimplemented from OpenNN::TrainingAlgorithm.

Definition at line 937 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::from_XML ( const tinyxml2::XMLDocument & document )

virtual

Loads a Levenberg-Marquardt method object from a XML document. Please mind about the format, wich is specified in the OpenNN manual.

Parameters

document TinyXML document containint the object data.

Reimplemented from OpenNN::TrainingAlgorithm.

Definition at line 2164 of file levenberg_marquardt_algorithm.cpp.

const double & OpenNN::LevenbergMarquardtAlgorithm::get_error_gradient_norm ( void ) const

Returns the value for the norm of the gradient vector at wich an error message is written to the screen and the program exits.

Definition at line 123 of file levenberg_marquardt_algorithm.cpp.

const double & OpenNN::LevenbergMarquardtAlgorithm::get_error_parameters_norm ( void ) const

Returns the value for the norm of the parameters vector at wich an error message is written to the screen and the program exits.

Definition at line 112 of file levenberg_marquardt_algorithm.cpp.

const double & OpenNN::LevenbergMarquardtAlgorithm::get_gradient_norm_goal ( void ) const

Returns the goal value for the norm of the performance function gradient. This is used as a stopping criterion when training a neural network.

Definition at line 165 of file levenberg_marquardt_algorithm.cpp.

const double & OpenNN::LevenbergMarquardtAlgorithm::get_performance_goal ( void ) const

Returns the goal value for the performance. This is used as a stopping criterion when training a neural network.

Definition at line 154 of file levenberg_marquardt_algorithm.cpp.

const bool & OpenNN::LevenbergMarquardtAlgorithm::get_reserve_Hessian_approximation_history ( void ) const

Returns true if the history of the Hessian approximations is to be reserved, and false otherwise.

Definition at line 256 of file levenberg_marquardt_algorithm.cpp.

const double & OpenNN::LevenbergMarquardtAlgorithm::get_warning_gradient_norm ( void ) const

Returns the minimum value for the norm of the gradient vector at wich a warning message is written to the screen.

Definition at line 101 of file levenberg_marquardt_algorithm.cpp.

const double & OpenNN::LevenbergMarquardtAlgorithm::get_warning_parameters_norm ( void ) const

Returns the minimum value for the norm of the parameters vector at wich a warning message is written to the screen.

Definition at line 90 of file levenberg_marquardt_algorithm.cpp.

LevenbergMarquardtAlgorithm::LevenbergMarquardtAlgorithmResults * OpenNN::LevenbergMarquardtAlgorithm::perform_training ( void )

virtual

Trains a neural network with an associated performance functional according to the Levenberg-Marquardt algorithm. Training occurs according to the training parameters.

Implements OpenNN::TrainingAlgorithm.

Definition at line 1306 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_damping_parameter ( const double & new_damping_parameter )

Sets a new damping parameter (lambda in the User's Guide) for the Hessian approximation.

Parameters

new_damping_parameter Damping parameter value.

Definition at line 420 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_damping_parameter_factor ( const double & new_damping_parameter_factor )

Sets a new damping parameter factor (beta in the User's Guide) for the Hessian approximation.

Parameters

new_damping_parameter_factor Damping parameter factor value.

Definition at line 442 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_default ( void )

virtual

Sets the following default values for the Levenberg-Marquardt algorithm: Training parameters:

Levenberg-Marquardt parameter: 0.001.

Stopping criteria:

Performance goal: 1.0e-6.
Gradient norm goal: 1.0e-6.
Maximum training time: 1000 seconds.
Maximum number of iterations: 1000.

User stuff:

Iterations between showing progress: 10.

Reimplemented from OpenNN::TrainingAlgorithm.

Definition at line 361 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_display_period ( const size_t & new_display_period )

Sets a new number of iterations between the training showing progress.

Parameters

new_display_period Number of iterations between the training showing progress.

Definition at line 903 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_error_gradient_norm ( const double & new_error_gradient_norm )

Sets a new value for the gradient vector norm at which an error message is written to the screen and the program exits.

Parameters

new_error_gradient_norm Error norm of gradient vector value.

Definition at line 625 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_error_parameters_norm ( const double & new_error_parameters_norm )

Sets a new value for the parameters vector norm at which an error message is written to the screen and the program exits.

Parameters

new_error_parameters_norm Error norm of parameters vector value.

Definition at line 594 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_gradient_norm_goal ( const double & new_gradient_norm_goal )

Sets a new the goal value for the norm of the performance function gradient. This is used as a stopping criterion when training a neural network.

Parameters

new_gradient_norm_goal Goal value for the norm of the performance function gradient.

Definition at line 728 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_maximum_damping_parameter ( const double & new_maximum_damping_parameter )

Sets a new maximum damping parameter allowed in the algorithm.

Parameters

new_maximum_damping_parameter Maximum damping parameter value.

Definition at line 494 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_maximum_generalization_performance_decreases ( const size_t & new_maximum_generalization_performance_decreases )

Sets a new maximum number of generalization failures.

Parameters

new_maximum_generalization_performance_decreases Maximum number of iterations in which the generalization evalutation decreases.

Definition at line 758 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_maximum_iterations_number ( const size_t & new_maximum_iterations_number )

Sets a maximum number of iterations for training.

Parameters

new_maximum_iterations_number Maximum number of iterations for training.

Definition at line 769 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_maximum_time ( const double & new_maximum_time )

Sets a new maximum training time.

Parameters

new_maximum_time Maximum training time.

Definition at line 780 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_minimum_damping_parameter ( const double & new_minimum_damping_parameter )

Sets a new minimum damping parameter allowed in the algorithm.

Parameters

new_minimum_damping_parameter Minimum damping parameter value.

Definition at line 468 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_minimum_parameters_increment_norm ( const double & new_minimum_parameters_increment_norm )

Sets a new value for the minimum parameters increment norm stopping criterion.

Parameters

new_minimum_parameters_increment_norm Value of norm of parameters increment norm used to stop training.

Definition at line 655 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_minimum_performance_increase ( const double & new_minimum_performance_increase )

Sets a new minimum performance improvement during training.

Parameters

new_minimum_performance_increase Minimum improvement in the performance between two iterations.

Definition at line 685 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_performance_goal ( const double & new_performance_goal )

Sets a new goal value for the performance. This is used as a stopping criterion when training a neural network.

Parameters

new_performance_goal Goal value for the performance.

Definition at line 716 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_damping_parameter_history ( const bool & new_reserve_damping_parameter_history )

Makes the damping parameter history vector to be reseved or not in memory.

Parameters

new_reserve_damping_parameter_history True if the damping parameter history vector is to be reserved, false otherwise.

Definition at line 520 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_elapsed_time_history ( const bool & new_reserve_elapsed_time_history )

Makes the elapsed time over the iterations to be reseved or not in memory. This is a vector.

Parameters

new_reserve_elapsed_time_history True if the elapsed time history vector is to be reserved, false otherwise.

Definition at line 879 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_generalization_performance_history ( const bool & new_reserve_generalization_performance_history )

Makes the Generalization performance history to be reserved or not in memory. This is a vector.

Parameters

new_reserve_generalization_performance_history True if the Generalization performance history is to be reserved, false otherwise.

Definition at line 891 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_gradient_history ( const bool & new_reserve_gradient_history )

Makes the gradient history vector of vectors to be reseved or not in memory.

Parameters

new_reserve_gradient_history True if the gradient history matrix is to be reserved, false otherwise.

Definition at line 843 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_gradient_norm_history ( const bool & new_reserve_gradient_norm_history )

Makes the gradient norm history vector to be reseved or not in memory.

Parameters

new_reserve_gradient_norm_history True if the gradient norm history matrix is to be reserved, false otherwise.

Definition at line 855 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_Hessian_approximation_history ( const bool & new_reserve_Hessian_approximation_history )

Sets the history of the Hessian approximation to be reserved or not in memory. This is a vector of matrices.

Parameters

new_reserve_Hessian_approximation_history True if the Hessian approximation history is to be reserved, false otherwise.

Definition at line 867 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_parameters_history ( const bool & new_reserve_parameters_history )

Makes the parameters history vector of vectors to be reseved or not in memory.

Parameters

new_reserve_parameters_history True if the parameters history vector of vectors is to be reserved, false otherwise.

Definition at line 810 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_parameters_norm_history ( const bool & new_reserve_parameters_norm_history )

Makes the parameters norm history vector to be reseved or not in memory.

Parameters

new_reserve_parameters_norm_history True if the parameters norm history vector is to be reserved, false otherwise.

Definition at line 821 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_reserve_performance_history ( const bool & new_reserve_performance_history )

Makes the performance history vector to be reseved or not in memory.

Parameters

new_reserve_performance_history True if the performance history vector is to be reserved, false otherwise.

Definition at line 832 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_warning_gradient_norm ( const double & new_warning_gradient_norm )

Sets a new value for the gradient vector norm at which a warning message is written to the screen.

Parameters

new_warning_gradient_norm Warning norm of gradient vector value.

Definition at line 563 of file levenberg_marquardt_algorithm.cpp.

void OpenNN::LevenbergMarquardtAlgorithm::set_warning_parameters_norm ( const double & new_warning_parameters_norm )

Sets a new value for the parameters vector norm at which a warning message is written to the screen.

Parameters

new_warning_parameters_norm Warning norm of parameters vector value.

Definition at line 532 of file levenberg_marquardt_algorithm.cpp.

Matrix< std::string > OpenNN::LevenbergMarquardtAlgorithm::to_string_matrix ( void ) const

virtual

Returns a default (empty) string matrix containing the members of the training algorithm object.

Reimplemented from OpenNN::TrainingAlgorithm.

Definition at line 1688 of file levenberg_marquardt_algorithm.cpp.

tinyxml2::XMLDocument * OpenNN::LevenbergMarquardtAlgorithm::to_XML ( void ) const

virtual

Serializes a default training algorithm object into a XML document of the TinyXML library. See the OpenNN manual for more information about the format of this document.

Reimplemented from OpenNN::TrainingAlgorithm.

Definition at line 1833 of file levenberg_marquardt_algorithm.cpp.

Member Data Documentation

size_t OpenNN::LevenbergMarquardtAlgorithm::maximum_generalization_performance_decreases

private

Maximum number of iterations at which the generalization performance decreases. This is an early stopping method for improving generalization.

Definition at line 369 of file levenberg_marquardt_algorithm.h.

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

Classes

Public Member Functions

Private Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation