performance_functional.h

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

#ifndef __PERFORMANCEFUNCTIONAL_H__
#define __PERFORMANCEFUNCTIONAL_H__

// System includes

#include <string>
#include <sstream>
#include <iostream>
#include <cmath>

// OpenNN includes

#include "vector.h"
#include "matrix.h"
#include "numerical_differentiation.h"

#include "data_set.h"
#include "mathematical_model.h"

#include "neural_network.h"
#include "performance_term.h"

#include "sum_squared_error.h"
#include "mean_squared_error.h"
#include "root_mean_squared_error.h"
#include "normalized_squared_error.h"
#include "minkowski_error.h"
#include "cross_entropy_error.h"
#include "outputs_integrals.h"
#include "solutions_error.h"
#include "final_solutions_error.h"
#include "independent_parameters_error.h"
#include "inverse_sum_squared_error.h"

#include "neural_parameters_norm.h"

// TinyXml includes

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

namespace OpenNN
{


class PerformanceFunctional
{

public:

   // DEFAULT CONSTRUCTOR

   explicit PerformanceFunctional(void);

   // OBJECTIVE FUNCTIONAL CONSTRUCTOR

   explicit PerformanceFunctional(PerformanceTerm*);

   // NEURAL NETWORK CONSTRUCTOR

   explicit PerformanceFunctional(NeuralNetwork*);

   // NEURAL NETWORK AND DATA SET CONSTRUCTOR

   explicit PerformanceFunctional(NeuralNetwork*, DataSet*);

   // NEURAL NETWORK AND MATHEMATICAL MODEL CONSTRUCTOR

   explicit PerformanceFunctional(NeuralNetwork*, MathematicalModel*);

   // NEURAL NETWORK, MATHEMATICAL MODEL AND DATA SET CONSTRUCTOR

   explicit PerformanceFunctional(NeuralNetwork*, MathematicalModel*, DataSet*);

   // FILE CONSTRUCTOR

   explicit PerformanceFunctional(const std::string&);

   // XML CONSTRUCTOR

   explicit PerformanceFunctional(const tinyxml2::XMLDocument&);


   // COPY CONSTRUCTOR

   PerformanceFunctional(const PerformanceFunctional&);

   // DESTRUCTOR

   virtual ~PerformanceFunctional(void);

   // STRUCTURES 


   struct ZeroOrderperformance
   {

      double performance;
   };


   struct FirstOrderperformance
   {

      double performance;


      Vector<double> gradient;
   };


   struct SecondOrderperformance
   {

      double performance;


      Vector<double> gradient;


      Matrix<double> Hessian;
   };


   // ENUMERATIONS


   enum ObjectiveType
   {
      NO_OBJECTIVE,
      SUM_SQUARED_ERROR_OBJECTIVE,
      MEAN_SQUARED_ERROR_OBJECTIVE,
      ROOT_MEAN_SQUARED_ERROR_OBJECTIVE,
      NORMALIZED_SQUARED_ERROR_OBJECTIVE,
      MINKOWSKI_ERROR_OBJECTIVE,
      CROSS_ENTROPY_ERROR_OBJECTIVE,
      OUTPUTS_INTEGRALS_OBJECTIVE,
      SOLUTIONS_ERROR_OBJECTIVE,
      FINAL_SOLUTIONS_ERROR_OBJECTIVE,
      INDEPENDENT_PARAMETERS_ERROR_OBJECTIVE,
      INVERSE_SUM_SQUARED_ERROR_OBJECTIVE,
      USER_OBJECTIVE
   };


   enum RegularizationType
   {
      NO_REGULARIZATION,
      NEURAL_PARAMETERS_NORM_REGULARIZATION,
      OUTPUTS_INTEGRALS_REGULARIZATION,
      USER_REGULARIZATION
   };


   enum ConstraintsType
   {
      NO_CONSTRAINTS,
      OUTPUTS_INTEGRALS_CONSTRAINTS,
      SOLUTIONS_ERROR_CONSTRAINTS,
      FINAL_SOLUTIONS_ERROR_CONSTRAINTS,
      INDEPENDENT_PARAMETERS_ERROR_CONSTRAINTS,
      USER_CONSTRAINTS
   };

   
   // METHODS

   // Check methods

   void check_neural_network(void) const;

   void check_performance_terms(void) const;

   // Get methods


   inline NeuralNetwork* get_neural_network_pointer(void) const 
   {
      #ifndef NDEBUG

      if(!neural_network_pointer)
      {
           std::ostringstream buffer;

           buffer << "OpenNN Exception: PerformanceFunctional class.\n"
                  << "NeuralNetwork* get_neural_network_pointer(void) const method.\n"
                  << "Neural network pointer is NULL.\n";

           throw std::logic_error(buffer.str());
      }

      #endif

      return(neural_network_pointer);
   }


   inline MathematicalModel* get_mathematical_model_pointer(void) const
   {
        #ifndef NDEBUG

        if(!mathematical_model_pointer)
        {
             std::ostringstream buffer;

             buffer << "OpenNN Exception: PerformanceFunctional class.\n"
                    << "MathematicalModel* get_mathematical_model_pointer(void) const method.\n"
                    << "MathematicalModel pointer is NULL.\n";

             throw std::logic_error(buffer.str());
        }

        #endif

      return(mathematical_model_pointer);
   }


   inline DataSet* get_data_set_pointer(void) const
   {
        #ifndef NDEBUG

        if(!data_set_pointer)
        {
             std::ostringstream buffer;

             buffer << "OpenNN Exception: PerformanceFunctional class.\n"
                    << "DataSet* get_data_set_pointer(void) const method.\n"
                    << "DataSet pointer is NULL.\n";

             throw std::logic_error(buffer.str());
        }

        #endif

      return(data_set_pointer);
   }


   bool has_neural_network(void) const;
   bool has_mathematical_model(void) const;
   bool has_data_set(void) const;

   bool has_generalization(void) const;

   bool is_sum_squared_terms(void) const;


   // Objective terms

   SumSquaredError* get_sum_squared_error_objective_pointer(void) const;
   MeanSquaredError* get_mean_squared_error_objective_pointer(void) const;
   RootMeanSquaredError* get_root_mean_squared_error_objective_pointer(void) const;
   NormalizedSquaredError* get_normalized_squared_error_objective_pointer(void) const;
   MinkowskiError* get_Minkowski_error_objective_pointer(void) const;
   CrossEntropyError* get_cross_entropy_error_objective_pointer(void) const;
   OutputsIntegrals* get_outputs_integrals_objective_pointer(void) const;
   SolutionsError* get_solutions_error_objective_pointer(void) const;
   FinalSolutionsError* get_final_solutions_error_objective_pointer(void) const;
   IndependentParametersError* get_independent_parameters_error_objective_pointer(void) const;
   InverseSumSquaredError* get_inverse_sum_squared_error_objective_pointer(void) const;
   PerformanceTerm* get_user_objective_pointer(void) const;

   // Regularization terms

   NeuralParametersNorm* get_neural_parameters_norm_regularization_pointer(void) const;
   OutputsIntegrals* get_outputs_integrals_regularization_pointer(void) const;
   PerformanceTerm* get_user_regularization_pointer(void) const;

   // Constraints terms

   OutputsIntegrals* get_outputs_integrals_constraints_pointer(void) const;
   SolutionsError* get_solutions_error_constraints_pointer(void) const;
   FinalSolutionsError* get_final_solutions_error_constraints_pointer(void) const;
   IndependentParametersError* get_independent_parameters_error_constraints_pointer(void) const;
   PerformanceTerm* get_user_constraints_pointer(void) const;

   // Functional type methods

   const ObjectiveType& get_objective_type(void) const;
   const RegularizationType& get_regularization_type(void) const;
   const ConstraintsType& get_constraints_type(void) const;

   std::string write_objective_type(void) const;
   std::string write_regularization_type(void) const;
   std::string write_constraints_type(void) const;

   std::string write_objective_type_text(void) const;
   std::string write_regularization_type_text(void) const;
   std::string write_constraints_type_text(void) const;

   // Serialization methods

   const bool& get_display(void) const;

   // Set methods

   void set_neural_network_pointer(NeuralNetwork*);

   void set_mathematical_model_pointer(MathematicalModel*);
   void set_data_set_pointer(DataSet*);

   void set_user_objective_pointer(PerformanceTerm*);
   void set_user_regularization_pointer(PerformanceTerm*);
   void set_user_constraints_pointer(PerformanceTerm*);

   void set_default(void);

   // Functionals methods 

   void set_objective_type(const ObjectiveType&);
   void set_regularization_type(const RegularizationType&);
   void set_constraints_type(const ConstraintsType&);

   void set_objective_type(const std::string&);
   void set_regularization_type(const std::string&);
   void set_constraints_type(const std::string&);
   
   void destruct_objective(void);
   void destruct_regularization(void);
   void destruct_constraints(void);

   void destruct_all_terms(void);

   void set_display(const bool&);

   // Performance functional methods

   double calculate_objective(void) const;
   double calculate_regularization(void) const;
   double calculate_constraints(void) const;

   double calculate_objective(const Vector<double>&) const;
   double calculate_regularization(const Vector<double>&) const;
   double calculate_constraints(const Vector<double>&) const;

   Vector<double> calculate_objective_terms(void) const;
   Vector<double> calculate_regularization_terms(void) const;
   Vector<double> calculate_constraints_terms(void) const;

   Matrix<double> calculate_objective_terms_Jacobian(void) const;
   Matrix<double> calculate_regularization_terms_Jacobian(void) const;
   Matrix<double> calculate_constraints_terms_Jacobian(void) const;

   Vector<double> calculate_objective_gradient(void) const;
   Vector<double> calculate_regularization_gradient(void) const;
   Vector<double> calculate_constraints_gradient(void) const;

   Vector<double> calculate_objective_gradient(const Vector<double>&) const;
   Vector<double> calculate_regularization_gradient(const Vector<double>&) const;
   Vector<double> calculate_constraints_gradient(const Vector<double>&) const;

   Matrix<double> calculate_objective_Hessian(void) const;
   Matrix<double> calculate_regularization_Hessian(void) const;
   Matrix<double> calculate_constraints_Hessian(void) const;

   Matrix<double> calculate_objective_Hessian(const Vector<double>&) const;
   Matrix<double> calculate_regularization_Hessian(const Vector<double>&) const;
   Matrix<double> calculate_constraints_Hessian(const Vector<double>&) const;

   double calculate_performance(void) const;
   Vector<double> calculate_gradient(void) const;
   Matrix<double> calculate_Hessian(void) const;

   double calculate_performance(const Vector<double>&) const;
   Vector<double> calculate_gradient(const Vector<double>&) const;
   Matrix<double> calculate_Hessian(const Vector<double>&) const;

   virtual Matrix<double> calculate_inverse_Hessian(void) const;

   virtual Vector<double> calculate_vector_dot_Hessian(const Vector<double>&) const;

   Vector<double> calculate_terms(void) const;
   Matrix<double> calculate_terms_Jacobian(void) const;

   virtual ZeroOrderperformance calculate_zero_order_performance(void) const;
   virtual FirstOrderperformance calculate_first_order_performance(void) const;
   virtual SecondOrderperformance calculate_second_order_performance(void) const;

   double calculate_generalization_objective(void) const;
   double calculate_generalization_regularization(void) const;
   double calculate_generalization_constraints(void) const;

   virtual double calculate_generalization_performance(void) const;

   // Taylor approximation methods

   double calculate_zero_order_Taylor_approximation(const Vector<double>&) const;
   double calculate_first_order_Taylor_approximation(const Vector<double>&) const;
   double calculate_second_order_Taylor_approximation(const Vector<double>&) const;

   // Directional performance

   double calculate_performance(const Vector<double>&, const double&) const;
   double calculate_performance_derivative(const Vector<double>&, const double&) const;
   double calculate_performance_second_derivative(const Vector<double>&, const double&) const;

   // Serialization methods

   virtual tinyxml2::XMLDocument* to_XML(void) const;   
   virtual void from_XML(const tinyxml2::XMLDocument&);
   
   virtual std::string to_string(void) const;
   virtual void save(const std::string&) const;   
   virtual void load(const std::string&);   

   virtual std::string write_information(void);   

   void print(void) const;


private:


   NeuralNetwork* neural_network_pointer;


   DataSet* data_set_pointer;


   MathematicalModel* mathematical_model_pointer;


   ObjectiveType objective_type;


   RegularizationType regularization_type;


   ConstraintsType constraints_type;

   // Objective terms


   SumSquaredError* sum_squared_error_objective_pointer;


   MeanSquaredError* mean_squared_error_objective_pointer;


   RootMeanSquaredError* root_mean_squared_error_objective_pointer;


   NormalizedSquaredError* normalized_squared_error_objective_pointer;


   MinkowskiError* Minkowski_error_objective_pointer;


   CrossEntropyError* cross_entropy_error_objective_pointer;


   OutputsIntegrals* outputs_integrals_objective_pointer;


   SolutionsError* solutions_error_objective_pointer;


   FinalSolutionsError* final_solutions_error_objective_pointer;


   IndependentParametersError* independent_parameters_error_objective_pointer;


   InverseSumSquaredError* inverse_sum_squared_error_objective_pointer;


   PerformanceTerm* user_objective_pointer;

   // Regularization terms


   NeuralParametersNorm* neural_parameters_norm_regularization_pointer;


   OutputsIntegrals* outputs_integrals_regularization_pointer;


   PerformanceTerm* user_regularization_pointer;

   // Constraints terms


   OutputsIntegrals* outputs_integrals_constraints_pointer;


   SolutionsError* solutions_error_constraints_pointer;


   FinalSolutionsError* final_solutions_error_constraints_pointer;


   IndependentParametersError* independent_parameters_error_constraints_pointer;


   PerformanceTerm* user_constraints_pointer;


   bool display;  
};

}

#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