conjugate_gradient.h

/****************************************************************************************************************/
/*                                                                                                              */
/*   OpenNN: Open Neural Networks Library                                                                       */
/*   www.artelnics.com/opennn                                                                                   */
/*                                                                                                              */
/*   C O N J U G A T E   G R A D I E N T   C L A S S   H E A D E R                                              */
/*                                                                                                              */
/*   Roberto Lopez                                                                                              */
/*   Artelnics - Making intelligent use of data                                                                 */
/*   [email protected]                                                                                 */
/*                                                                                                              */
/****************************************************************************************************************/

#ifndef __CONJUGATEGRADIENT_H__
#define __CONJUGATEGRADIENT_H__

// System inlcludes

#include <string>
#include <sstream>
#include <iostream>
#include <fstream>
#include <algorithm>
#include <functional>
#include <climits>
#include <cmath>
#include <ctime>

// OpenNN includes

#include "performance_functional.h"

#include "training_algorithm.h"
#include "training_rate_algorithm.h"

// TinyXml includes

#include "../tinyxml2/tinyxml2.h"

namespace OpenNN
{


class ConjugateGradient : public TrainingAlgorithm
{

public:

   // ENUMERATIONS


   enum TrainingDirectionMethod{PR, FR};

   // DEFAULT CONSTRUCTOR

   explicit ConjugateGradient(void); 


   // GENERAL CONSTRUCTOR

   explicit ConjugateGradient(PerformanceFunctional*);


   // XML CONSTRUCTOR

   explicit ConjugateGradient(const tinyxml2::XMLDocument&); 


   // DESTRUCTOR

   virtual ~ConjugateGradient(void);


   // STRUCTURES


   struct ConjugateGradientResults : public TrainingAlgorithm::TrainingAlgorithmResults
   {

       ConjugateGradientResults(void)
       {
           conjugate_gradient_pointer = NULL;
       }


       ConjugateGradientResults(ConjugateGradient* new_conjugate_gradient_pointer)
       {
           conjugate_gradient_pointer = new_conjugate_gradient_pointer;
       }


       virtual ~ConjugateGradientResults(void)
       {
       }


      ConjugateGradient* conjugate_gradient_pointer;

      // TRAINING HISTORY


      Vector< Vector<double> > parameters_history;


      Vector<double> parameters_norm_history;


      Vector<double> performance_history;


      Vector<double> generalization_performance_history;


      Vector< Vector<double> > gradient_history;


      Vector<double> gradient_norm_history;


      Vector< Vector<double> > training_direction_history;


      Vector<double> training_rate_history;


      Vector<double> elapsed_time_history;

      // FINAL VALUES


      Vector<double> final_parameters;


      double final_parameters_norm;


      double final_performance;


      double final_generalization_performance;


      Vector<double> final_gradient;


      double final_gradient_norm;


      Vector<double> final_training_direction;


      double final_training_rate;


      double elapsed_time;


      size_t iterations_number;

      void resize_training_history(const size_t&);
      std::string to_string(void) const;

      Matrix<std::string> write_final_results(const size_t& precision = 3) const;
   };


   // METHODS

   // Get methods

   const TrainingRateAlgorithm& get_training_rate_algorithm(void) const;
   TrainingRateAlgorithm* get_training_rate_algorithm_pointer(void);

   // Training operators

   const TrainingDirectionMethod& get_training_direction_method(void) const;
   std::string write_training_direction_method(void) const;

   // Training parameters

   const double& get_warning_parameters_norm(void) const;
   const double& get_warning_gradient_norm(void) const;
   const double& get_warning_training_rate(void) const;

   const double& get_error_parameters_norm(void) const;
   const double& get_error_gradient_norm(void) const;
   const double& get_error_training_rate(void) const;

   // Stopping criteria

   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 size_t& get_maximum_generalization_performance_decreases(void) const;
   const double& get_gradient_norm_goal(void) const;

   const size_t& get_maximum_iterations_number(void) const;
   const double& get_maximum_time(void) const;

   // Reserve training history

   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_generalization_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_training_direction_history(void) const;
   const bool& get_reserve_training_rate_history(void) const;
   const bool& get_reserve_elapsed_time_history(void) const;

   // Set methods

   void set_default(void);

   void set_performance_functional_pointer(PerformanceFunctional*);

   // Training operators

   void set_training_direction_method(const TrainingDirectionMethod&);
   void set_training_direction_method(const std::string&);

   // Training parameters

   void set_warning_parameters_norm(const double&);
   void set_warning_gradient_norm(const double&);
   void set_warning_training_rate(const double&);

   void set_error_parameters_norm(const double&);
   void set_error_gradient_norm(const double&);
   void set_error_training_rate(const double&);

   // Stopping criteria

   void set_minimum_parameters_increment_norm(const double&);

   void set_performance_goal(const double&);
   void set_minimum_performance_increase(const double&);
   void set_maximum_generalization_performance_decreases(const size_t&);
   void set_gradient_norm_goal(const double&);

   void set_maximum_iterations_number(const size_t&);
   void set_maximum_time(const double&);

   // Reserve training history

   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_generalization_performance_history(const bool&);
   void set_reserve_gradient_history(const bool&);
   void set_reserve_gradient_norm_history(const bool&);

   void set_reserve_training_direction_history(const bool&);
   void set_reserve_training_rate_history(const bool&);
   void set_reserve_elapsed_time_history(const bool&);

   void set_reserve_all_training_history(const bool&);

   // Utilities

   void set_display_period(const size_t&);
   void set_save_period(const size_t&);

   // Training direction methods

   double calculate_PR_parameter(const Vector<double>&, const Vector<double>&) const;
   double calculate_FR_parameter(const Vector<double>&, const Vector<double>&) const;

   Vector<double> calculate_PR_training_direction(const Vector<double>&, const Vector<double>&, const Vector<double>&) const;
   Vector<double> calculate_FR_training_direction(const Vector<double>&, const Vector<double>&, const Vector<double>&) const;

   Vector<double> calculate_gradient_descent_training_direction(const Vector<double>&) const;

   Vector<double> calculate_training_direction(const Vector<double>&, const Vector<double>&, const Vector<double>&) const;

   // Training methods

   ConjugateGradientResults* perform_training(void);

   std::string write_training_algorithm_type(void) const;

   // Serialization methods

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

   tinyxml2::XMLDocument* to_XML(void) const;
   void from_XML(const tinyxml2::XMLDocument&);

private:


   TrainingDirectionMethod training_direction_method;


   TrainingRateAlgorithm training_rate_algorithm;


   double warning_parameters_norm;


   double warning_gradient_norm;   


   double warning_training_rate;

   
   double error_parameters_norm;


   double error_gradient_norm;


   double error_training_rate;


   // STOPPING CRITERIA


   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;

   // TRAINING HISTORY


   bool reserve_parameters_history;


   bool reserve_parameters_norm_history;


   bool reserve_performance_history;


   bool reserve_gradient_history;


   bool reserve_gradient_norm_history;

   
   bool reserve_training_direction_history;


   bool reserve_training_rate_history;


   bool reserve_elapsed_time_history;


   bool reserve_generalization_performance_history;


};

}

#endif


// OpenNN: Open Neural Networks Library.
// Copyright (c) 2005-2015 Roberto Lopez.
//
// This library is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 2.1 of the License, or any later version.
//
// This library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
// Lesser General Public License for more details.

// You should have received a copy of the GNU Lesser General Public
// License along with this library; if not, write to the Free Software
// Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA