Flood::ConjugateGradient Class Reference

#include <ConjugateGradient.h>

Inheritance diagram for Flood::ConjugateGradient:

Flood::TrainingAlgorithm

List of all members.

Public Types

enum  TrainingDirectionMethod { PolakRibiere, FletcherReeves }

Public Member Functions

 ConjugateGradient (ObjectiveFunctional *)
 ConjugateGradient (void)
virtual ~ConjugateGradient (void)
TrainingDirectionMethod get_training_direction_method (void)
std::string get_training_direction_method_name (void)
void set_training_direction_method (const TrainingDirectionMethod &)
void set_training_direction_method (const std::string &)
double calculate_Polak_Ribiere_parameter (const Vector< double > &, const Vector< double > &)
double calculate_Fletcher_Reeves_parameter (const Vector< double > &, const Vector< double > &)
Vector< double > calculate_Polak_Ribiere_training_direction (const Vector< double > &, const Vector< double > &, const Vector< double > &)
Vector< double > calculate_Fletcher_Reeves_training_direction (const Vector< double > &, const Vector< double > &, const Vector< double > &)
Vector< double > calculate_training_direction (const Vector< double > &, const Vector< double > &, const Vector< double > &)
Vector< double > calculate_gradient_descent_training_direction (const Vector< double > &)
void train (void)
void set_reserve_all_training_history (bool)
void resize_training_history (int)
std::string get_training_history_XML (bool)
std::string to_XML (bool)
void load (const char *)

Detailed Description

This concrete class represents a conjugate gradient training algorithm for an objective functional of a multilayer perceptron.

Definition at line 29 of file ConjugateGradient.h.

Member Enumeration Documentation

enum Flood::ConjugateGradient::TrainingDirectionMethod

Enumeration of the available training operators for obtaining the training direction.

Definition at line 38 of file ConjugateGradient.h.

Constructor & Destructor Documentation

Flood::ConjugateGradient::ConjugateGradient ( ObjectiveFunctional *  new_objective_functional_pointer  )  [explicit]

General constructor. It creates a conjugate gradient training algorithm object associated to an objective functional object. It also initializes the class members to their default values:

Training operators:

  • Training direction method = Polak-Ribiere;
  • Training rate method = Brent;

Training parameters:

  • First training rate: 1.0.
  • Bracketing factor: 2.0.
  • Training rate tolerance: 1.0e-3.

Stopping criteria:

  • Evaluation goal: -1.0e99.
  • Gradient norm goal: 0.0.
  • Maximum training time: 1.0e6.
  • Maximum number of epochs: 100.

User stuff:

  • Warning training rate: 1.0e6.
  • Error training rate: 1.0e12.
  • Display: true.
  • Display period: 25.

Reserve:

  • Reserve training direction history: false.
  • Reserve training direction norm history: false.
  • Reserve training rate history: false.

Parameters:
new_objective_functional_pointer Pointer to an objective functional object.

Definition at line 75 of file ConjugateGradient.cpp.

Flood::ConjugateGradient::ConjugateGradient ( void   )  [explicit]

Default constructor. It creates a conjugate gradient training algorithm object not associated to any objective functional object. It also initializes the class members to their default values:

Training operators:

  • Training direction method = Polak-Ribiere;
  • Training rate method = Brent;

Training parameters:

  • First training rate: 1.0.
  • Bracketing factor: 2.0.
  • Training rate tolerance: 1.0e-3.

Stopping criteria:

  • Evaluation goal: -1.0e99.
  • Gradient norm goal: 0.0.
  • Maximum training time: 1.0e6.
  • Maximum number of epochs: 100.

User stuff:

  • Warning training rate: 1.0e6.
  • Error training rate: 1.0e12.
  • Display: true.
  • Display period: 25.

Reserve:

  • Reserve training direction history: false.
  • Reserve training direction norm history: false.
  • Reserve training rate history: false.

Definition at line 126 of file ConjugateGradient.cpp.

Flood::ConjugateGradient::~ConjugateGradient ( void   )  [virtual]

Destructor.

Definition at line 138 of file ConjugateGradient.cpp.

Member Function Documentation

double Flood::ConjugateGradient::calculate_Fletcher_Reeves_parameter ( const Vector< double > &  old_gradient,
const Vector< double > &  gradient 
)

This method returns the Fletcher-Reeves parameter used to calculate the training direction.

Parameters:
old_gradient Previous objective function gradient.
gradient,: Current objective function gradient.

Definition at line 281 of file ConjugateGradient.cpp.

Vector< double > Flood::ConjugateGradient::calculate_Fletcher_Reeves_training_direction ( const Vector< double > &  old_gradient,
const Vector< double > &  gradient,
const Vector< double > &  old_training_direction 
)

This method returns the training direction using the Fletcher-Reeves update.

Parameters:
old_gradient Previous objective function gradient.
gradient Current objective function gradient.
old_training_direction Previous training direction vector.

Definition at line 382 of file ConjugateGradient.cpp.

double Flood::ConjugateGradient::calculate_Polak_Ribiere_parameter ( const Vector< double > &  old_gradient,
const Vector< double > &  gradient 
)

This method returns the Polak-Ribiere parameter used to calculate the training direction.

Parameters:
old_gradient Previous objective function gradient.
gradient Current objective function gradient.

Definition at line 319 of file ConjugateGradient.cpp.

Vector< double > Flood::ConjugateGradient::calculate_Polak_Ribiere_training_direction ( const Vector< double > &  old_gradient,
const Vector< double > &  gradient,
const Vector< double > &  old_training_direction 
)

This method returns the training direction using the Polak-Ribiere update.

Parameters:
old_gradient Previous objective function gradient.
gradient Current objective function gradient.
old_training_direction Previous training direction vector.

Definition at line 358 of file ConjugateGradient.cpp.

ConjugateGradient::TrainingDirectionMethod Flood::ConjugateGradient::get_training_direction_method ( void   ) 

This method returns the conjugate gradient training direction method used for training.

Definition at line 149 of file ConjugateGradient.cpp.

std::string Flood::ConjugateGradient::get_training_direction_method_name ( void   ) 

This method returns a string with the name of the training direction.

Definition at line 159 of file ConjugateGradient.cpp.

std::string Flood::ConjugateGradient::get_training_history_XML ( bool  show_declaration  )  [virtual]

This method returns a string representation of the training history in XML-type format.

Parameters:
show_declaration True if an XML-type declaration is to be included at the beginning of the string.

Reimplemented from Flood::TrainingAlgorithm.

Definition at line 1574 of file ConjugateGradient.cpp.

void Flood::ConjugateGradient::load ( const char *  filename  )  [virtual]

This method loads a conjugate gradient object from an XML-type file. Please mind about the file format, wich is specified in the User's Guide.

Parameters:
filename Name of conjugate gradient XML-type.

Reimplemented from Flood::TrainingAlgorithm.

Definition at line 966 of file ConjugateGradient.cpp.

void Flood::ConjugateGradient::resize_training_history ( int  new_size  )  [virtual]

This method resizes the vectors of training history variables to a new size. Only those reserved variablers will be resized.

Parameters:
new_size Size of training history.

Reimplemented from Flood::TrainingAlgorithm.

Definition at line 1512 of file ConjugateGradient.cpp.

void Flood::ConjugateGradient::set_reserve_all_training_history ( bool  new_reserve_all_training_history  )  [virtual]

This method makes the training history of all variables to reseved or not in memory when training.

  • Parameters.
  • Parameters norm.
  • Evaluation.
  • Gradient.
  • Gradient norm.
  • Validation error.
  • Training direction.
  • Training direction norm.
  • Training rate.

Parameters:
new_reserve_all_training_history True if all training history variables are to be reserved, false otherwise.

Reimplemented from Flood::TrainingAlgorithm.

Definition at line 250 of file ConjugateGradient.cpp.

void Flood::ConjugateGradient::set_training_direction_method ( const std::string &  new_training_direction_method_name  ) 

This method sets a new conjugate gradient training direction from a string representation. Possible values are:

  • "PolakRibiere"
  • "FletcherReeves"
Parameters:
new_training_direction_method_name String with the name of the training direction method.

Definition at line 211 of file ConjugateGradient.cpp.

void Flood::ConjugateGradient::set_training_direction_method ( const TrainingDirectionMethod &  new_training_direction_method  ) 

This method sets a new training direction method to be used for training.

Parameters:
new_training_direction_method Conjugate gradient training direction method.

Definition at line 195 of file ConjugateGradient.cpp.

std::string Flood::ConjugateGradient::to_XML ( bool  show_declaration  )  [virtual]

This method returns a string representation of the conjuage gradient object in a XML formnat. See the user's guide.

Parameters:
show_declaration True if an XML-type declaration is to be included at the beginning.

Reimplemented from Flood::TrainingAlgorithm.

Definition at line 840 of file ConjugateGradient.cpp.

void Flood::ConjugateGradient::train ( void   )  [virtual]

This method trains a multilayer perceptron with an associated evaluation function according to the conjugate gradient algorithm. Training occurs according to the training operators, training parameters and stopping criteria.

Implements Flood::TrainingAlgorithm.

Definition at line 444 of file ConjugateGradient.cpp.

The documentation for this class was generated from the following files:
Generated on Fri Jul 30 09:51:58 2010 for Flood by  doxygen 1.5.9