elman.hpp

//| This file is a part of the sferes2 framework.
//| Copyright 2009, ISIR / Universite Pierre et Marie Curie (UPMC)
//| Main contributor(s): Jean-Baptiste Mouret, mouret@isir.fr
//|
//| This software is a computer program whose purpose is to facilitate
//| experiments in evolutionary computation and evolutionary robotics.
//|
//| This software is governed by the CeCILL license under French law
//| and abiding by the rules of distribution of free software.  You
//| can use, modify and/ or redistribute the software under the terms
//| of the CeCILL license as circulated by CEA, CNRS and INRIA at the
//| following URL "http://www.cecill.info".
//|
//| As a counterpart to the access to the source code and rights to
//| copy, modify and redistribute granted by the license, users are
//| provided only with a limited warranty and the software's author,
//| the holder of the economic rights, and the successive licensors
//| have only limited liability.
//|
//| In this respect, the user's attention is drawn to the risks
//| associated with loading, using, modifying and/or developing or
//| reproducing the software by the user in light of its specific
//| status of free software, that may mean that it is complicated to
//| manipulate, and that also therefore means that it is reserved for
//| developers and experienced professionals having in-depth computer
//| knowledge. Users are therefore encouraged to load and test the
//| software's suitability as regards their requirements in conditions
//| enabling the security of their systems and/or data to be ensured
//| and, more generally, to use and operate it in the same conditions
//| as regards security.
//|
//| The fact that you are presently reading this means that you have
//| had knowledge of the CeCILL license and that you accept its terms.




#ifndef _NN_ELMAN_HPP_
#define _NN_ELMAN_HPP_

#include "nn.hpp"

namespace nn {
  // a "modified" Elman network with self-recurrent context units
  // E.g. : Training Elman and Jordan networks for system
  // identification using genetic algorithms
  // Artificial Intelligence in Engineering
  // Volume 13, Issue 2, April 1999, Pages 107-117
  // first input is a BIAS input (it should be set to 1)
  template<typename N, typename C>
  class Elman : public NN<N, C> {
   public:
    typedef nn::NN<N, C> nn_t;
    typedef typename nn_t::io_t io_t;
    typedef typename nn_t::vertex_desc_t vertex_desc_t;
    typedef typename nn_t::edge_desc_t edge_desc_t;
    typedef typename nn_t::adj_it_t adj_it_t;
    typedef typename nn_t::graph_t graph_t;
    typedef N neuron_t;
    typedef C conn_t;

    Elman(size_t nb_inputs,
          size_t nb_hidden,
          size_t nb_outputs) {
      // neurons
      this->set_nb_inputs(nb_inputs + 1);
      this->set_nb_outputs(nb_outputs);
      for (size_t i = 0; i < nb_hidden; ++i)
        _hidden_neurons.
        push_back(this->add_neuron(std::string("h")
                                   + boost::lexical_cast<std::string>(i)));
      for (size_t i = 0; i < nb_hidden; ++i)
        _context_neurons.
        push_back(this->add_neuron(std::string("c")
                                   + boost::lexical_cast<std::string>(i)));
      // connections
      this->full_connect(this->_inputs, this->_hidden_neurons,
                         trait<typename N::weight_t>::zero());
      this->full_connect(this->_hidden_neurons, this->_outputs,
                         trait<typename N::weight_t>::zero());
      this->connect(this->_hidden_neurons, this->_context_neurons,
                    trait<typename N::weight_t>::zero());
      this->connect(this->_context_neurons, this->_context_neurons,
                    trait<typename N::weight_t>::zero());
      this->full_connect(this->_context_neurons, this->_hidden_neurons,
                         trait<typename N::weight_t>::zero());
      // bias
      // (hidden layer is already connect to input(0))
      size_t last = this->get_nb_inputs();
      for (size_t i = 0; i < _context_neurons.size(); ++i)
        this->add_connection(this->get_input(last), _context_neurons[i],
                             trait<typename N::weight_t>::zero());
      for (size_t i = 0; i < this->get_nb_outputs(); ++i)
        this->add_connection(this->get_input(last), this->get_output(i),
                             trait<typename N::weight_t>::zero());
    }
    unsigned get_nb_inputs() const {
      return this->_inputs.size() - 1;
    }
    void step(const std::vector<io_t>& in) {
      assert(in.size() == this->get_nb_inputs());
      std::vector<io_t> inf = in;
      inf.push_back(1.0f);
      nn_t::_step(inf);
    }
   protected:

    std::vector<vertex_desc_t> _hidden_neurons;
    std::vector<vertex_desc_t> _context_neurons;

  };
  namespace elman {
    template<int NbInputs, int NbHidden, int NbOutputs>
    struct Count {
      SFERES_CONST int nb_inputs = NbInputs + 1; // bias is an input
      SFERES_CONST int nb_outputs = NbOutputs;
      SFERES_CONST int nb_hidden = NbHidden;
      SFERES_CONST int nb_params =
        nb_inputs * nb_hidden // input to hidden (full)
        + nb_hidden * nb_outputs // hidden to output (full)
        + nb_hidden // hidden to context (1-1)
        + nb_hidden // context to itself (1-1)
        + nb_hidden * nb_hidden // context to hidden (full)
        + nb_hidden // bias context
        + nb_outputs; // bias outputs
    };

  }
}

#endif