sdyna.h

Go to the documentation of this file.

/**
 *
 *   Copyright 2005-2020 Pierre-Henri WUILLEMIN(@LIP6) & Christophe GONZALES(@AMU)
 *   info_at_agrum_dot_org
 *
 *  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 3 of the License, or
 *  (at your option) 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, see <http://www.gnu.org/licenses/>.
 *
 */


/**
 * @file
 * @brief Headers of the SDyna abstract class.
 *
 * @author Pierre-Henri WUILLEMIN(@LIP6) and Jean-Christophe MAGNAN and Christophe
 * GONZALES(@AMU)
 */

// =========================================================================
#ifndef GUM_SDYNA_H
#define GUM_SDYNA_H
// =========================================================================
#include <agrum/agrum.h>
#include <agrum/tools/multidim/instantiation.h>
// =========================================================================
#include <agrum/tools/variables/discreteVariable.h>
// =========================================================================
#include <agrum/FMDP/SDyna/Strategies/IDecisionStrategy.h>
#include <agrum/FMDP/SDyna/Strategies/ILearningStrategy.h>
#include <agrum/FMDP/SDyna/Strategies/IPlanningStrategy.h>
#include <agrum/FMDP/decision/E_GreedyDecider.h>
#include <agrum/FMDP/decision/lazyDecider.h>
#include <agrum/FMDP/decision/randomDecider.h>
#include <agrum/FMDP/decision/statisticalLazyDecider.h>
#include <agrum/FMDP/fmdp.h>
#include <agrum/FMDP/learning/fmdpLearner.h>
#include <agrum/FMDP/learning/observation.h>
#include <agrum/FMDP/planning/actionSet.h>
#include <agrum/FMDP/planning/adaptiveRMaxPlaner.h>
#include <agrum/FMDP/planning/structuredPlaner.h>
// =========================================================================

namespace gum {

  /**
   * @class SDYNA
   * @headerfile sdyna.h <agrum/FMDP/SDyna/sdyna.h>
   * @brief
   * @ingroup fmdp_group
   *
   * The general SDyna architecture abstract class.
   * Instance of SDyna architecture should inherit
   *
   */
  class SDYNA {
    // ###################################################################
    /// @name
    // ###################################################################
    /// @
    public:
    // ==========================================================================
    ///
    // ==========================================================================
    static SDYNA* spitiInstance(double attributeSelectionThreshold = 0.99,
                                double discountFactor              = 0.9,
                                double epsilon                     = 1,
                                Idx    observationPhaseLenght      = 100,
                                Idx    nbValueIterationStep        = 10) {
      bool               actionReward = false;
      ILearningStrategy* ls = new FMDPLearner< CHI2TEST, CHI2TEST, ITILEARNER >(
         attributeSelectionThreshold,
         actionReward);
      IPlanningStrategy< double >* ps
         = StructuredPlaner< double >::sviInstance(discountFactor, epsilon);
      IDecisionStrategy* ds = new E_GreedyDecider();
      return new SDYNA(ls,
                       ps,
                       ds,
                       observationPhaseLenght,
                       nbValueIterationStep,
                       actionReward);
    }

    // ==========================================================================
    ///
    // ==========================================================================
    static SDYNA* spimddiInstance(double attributeSelectionThreshold = 0.99,
                                  double similarityThreshold         = 0.3,
                                  double discountFactor              = 0.9,
                                  double epsilon                     = 1,
                                  Idx    observationPhaseLenght      = 100,
                                  Idx    nbValueIterationStep        = 10) {
      bool               actionReward = false;
      ILearningStrategy* ls = new FMDPLearner< GTEST, GTEST, IMDDILEARNER >(
         attributeSelectionThreshold,
         actionReward,
         similarityThreshold);
      IPlanningStrategy< double >* ps
         = StructuredPlaner< double >::spumddInstance(discountFactor,
                                                      epsilon,
                                                      false);
      IDecisionStrategy* ds = new E_GreedyDecider();
      return new SDYNA(ls,
                       ps,
                       ds,
                       observationPhaseLenght,
                       nbValueIterationStep,
                       actionReward,
                       false);
    }

    // ==========================================================================
    ///
    // ==========================================================================
    static SDYNA* RMaxMDDInstance(double attributeSelectionThreshold = 0.99,
                                  double similarityThreshold         = 0.3,
                                  double discountFactor              = 0.9,
                                  double epsilon                     = 1,
                                  Idx    observationPhaseLenght      = 100,
                                  Idx    nbValueIterationStep        = 10) {
      bool               actionReward = true;
      ILearningStrategy* ls = new FMDPLearner< GTEST, GTEST, IMDDILEARNER >(
         attributeSelectionThreshold,
         actionReward,
         similarityThreshold);
      AdaptiveRMaxPlaner* rm
         = AdaptiveRMaxPlaner::ReducedAndOrderedInstance(ls,
                                                         discountFactor,
                                                         epsilon);
      IPlanningStrategy< double >* ps = rm;
      IDecisionStrategy*           ds = rm;
      return new SDYNA(ls,
                       ps,
                       ds,
                       observationPhaseLenght,
                       nbValueIterationStep,
                       actionReward);
    }

    // ==========================================================================
    ///
    // ==========================================================================
    static SDYNA* RMaxTreeInstance(double attributeSelectionThreshold = 0.99,
                                   double discountFactor              = 0.9,
                                   double epsilon                     = 1,
                                   Idx    observationPhaseLenght      = 100,
                                   Idx    nbValueIterationStep        = 10) {
      bool               actionReward = true;
      ILearningStrategy* ls
         = new FMDPLearner< GTEST, GTEST, ITILEARNER >(attributeSelectionThreshold,
                                                       actionReward);
      AdaptiveRMaxPlaner* rm
         = AdaptiveRMaxPlaner::TreeInstance(ls, discountFactor, epsilon);
      IPlanningStrategy< double >* ps = rm;
      IDecisionStrategy*           ds = rm;
      return new SDYNA(ls,
                       ps,
                       ds,
                       observationPhaseLenght,
                       nbValueIterationStep,
                       actionReward);
    }

    // ==========================================================================
    ///
    // ==========================================================================
    static SDYNA* RandomMDDInstance(double attributeSelectionThreshold = 0.99,
                                    double similarityThreshold         = 0.3,
                                    double discountFactor              = 0.9,
                                    double epsilon                     = 1,
                                    Idx    observationPhaseLenght      = 100,
                                    Idx    nbValueIterationStep        = 10) {
      bool               actionReward = true;
      ILearningStrategy* ls = new FMDPLearner< GTEST, GTEST, IMDDILEARNER >(
         attributeSelectionThreshold,
         actionReward,
         similarityThreshold);
      IPlanningStrategy< double >* ps
         = StructuredPlaner< double >::spumddInstance(discountFactor, epsilon);
      IDecisionStrategy* ds = new RandomDecider();
      return new SDYNA(ls,
                       ps,
                       ds,
                       observationPhaseLenght,
                       nbValueIterationStep,
                       actionReward);
    }

    // ==========================================================================
    ///
    // ==========================================================================
    static SDYNA* RandomTreeInstance(double attributeSelectionThreshold = 0.99,
                                     double discountFactor              = 0.9,
                                     double epsilon                     = 1,
                                     Idx    observationPhaseLenght      = 100,
                                     Idx    nbValueIterationStep        = 10) {
      bool               actionReward = true;
      ILearningStrategy* ls = new FMDPLearner< CHI2TEST, CHI2TEST, ITILEARNER >(
         attributeSelectionThreshold,
         actionReward);
      IPlanningStrategy< double >* ps
         = StructuredPlaner< double >::sviInstance(discountFactor, epsilon);
      IDecisionStrategy* ds = new RandomDecider();
      return new SDYNA(ls,
                       ps,
                       ds,
                       observationPhaseLenght,
                       nbValueIterationStep,
                       actionReward);
    }


    /// @}

    // ###################################################################
    /// @name Constructor & destructor.
    // ###################################################################
    /// @{

    // ==========================================================================
    /**
     * Constructor
     *
     * @return an instance of SDyna architecture
     */
    // ==========================================================================
    private:
    SDYNA(ILearningStrategy*           learner,
          IPlanningStrategy< double >* planer,
          IDecisionStrategy*           decider,
          Idx                          observationPhaseLenght,
          Idx                          nbValueIterationStep,
          bool                         actionReward,
          bool                         verbose = true);

    // ==========================================================================
    /// Destructor
    // ==========================================================================
    public:
    ~SDYNA();

    /// @}


    // ###################################################################
    /// @name Problem specification methods
    // ###################################################################
    /// @{
    public:
    // ==========================================================================
    /**
     * Inserts a new action in the SDyna instance.
     * @warning Without effect until method initialize is called
     * @param actionId : an id to identify the action
     * @param actionName : its human name
     */
    // ==========================================================================
    void addAction(const Idx actionId, const std::string& actionName) {
      fmdp_->addAction(actionId, actionName);
    }

    // ==========================================================================
    /**
     * Inserts a new variable in the SDyna instance.
     * @warning Without effect until method initialize is called
     * @param var : the var to be added.
     * Note that variable may or may not have all its modalities given.
     * If not they will be discovered by the SDyna architecture during the
     * process
     */
    // ==========================================================================
    void addVariable(const DiscreteVariable* var) { fmdp_->addVariable(var); }

    /// @}


    // ###################################################################
    /// @name Initialization
    // ###################################################################
    /// @{
    public:
    // ==========================================================================
    /**
     * Initializes the Sdyna instance.
     */
    // ==========================================================================
    void initialize();

    // ==========================================================================
    /**
     * Initializes the Sdyna instance at given state.
     * @param initialState : the state of the studied system from which we will
     * begin the explore, learn and exploit process
     */
    // ==========================================================================
    void initialize(const Instantiation& initialState);

    /// @}


    // ###################################################################
    /// @name Incremental methods
    // ###################################################################
    /// @{
    public:
    // ==========================================================================
    /**
     * Sets last state visited to the given state.
     * During the learning process, we will consider that were in this state
     * before the transition.
     * @param currentState : the state
     */
    // ==========================================================================
    void setCurrentState(const Instantiation& currentState) {
      lastState_ = currentState;
    }

    // ==========================================================================
    /**
     * @return actionId the id of the action the SDyna instance wish to be
     * performed
     * @param curState the state in which we currently are
     */
    // ==========================================================================
    Idx takeAction(const Instantiation& curState);

    // ==========================================================================
    /**
     * @return the id of the action the SDyna instance wish to be performed
     */
    // ==========================================================================
    Idx takeAction();

    // ==========================================================================
    /**
     * Performs a feedback on the last transition.
     * In extenso, learn from the transition.
     * @param originalState : the state we were in before the transition
     * @param reachedState : the state we reached after
     * @param performedAction : the action we performed
     * @param obtainedReward : the reward we obtained
     */
    // ==========================================================================
    void feedback(const Instantiation& originalState,
                  const Instantiation& reachedState,
                  Idx                  performedAction,
                  double               obtainedReward);

    // ==========================================================================
    /**
     * Performs a feedback on the last transition.
     * In extenso, learn from the transition.
     * @param reachedState : the state reached after the transition
     * @param obtainedReward : the reward obtained during the transition
     * @warning Uses the originalState__ and performedAction__ stored in cache
     * If you want to specify the original state and the performed action, see
     * below
     */
    // ==========================================================================
    void feedback(const Instantiation& reachedState, double obtainedReward);

    // ==========================================================================
    /**
     * Starts a new planning
     * @param nbStep : the maximal number of value iteration performed in this
     * planning
     */
    // ==========================================================================
    void makePlanning(Idx nbStep);

    /// @}


    public:
    // ==========================================================================
    /**
     * Returns
     * @return a string describing the learned FMDP, and the associated
     * optimal policy.
     * Both in DOT language.
     */
    // ==========================================================================
    std::string toString();

    std::string optimalPolicy2String() { return planer__->optimalPolicy2String(); }


    // ###################################################################
    /// @name Size methods
    /// @brief just to get the size of the different data structure for
    /// performance evaluation purposes only
    // ###################################################################
    /// @{
    public:
    // ==========================================================================
    /**
     * @brief learnerSize
     * @return
     */
    // ==========================================================================
    Size learnerSize() { return learner__->size(); }

    // ==========================================================================
    /**
     * @brief modelSize
     * @return
     */
    // ==========================================================================
    Size modelSize() { return fmdp_->size(); }

    // ==========================================================================
    /**
     * @brief valueFunctionSize
     * @return
     */
    // ==========================================================================
    Size valueFunctionSize() { return planer__->vFunctionSize(); }

    // ==========================================================================
    /**
     * @brief optimalPolicySize
     * @return
     */
    // ==========================================================================
    Size optimalPolicySize() { return planer__->optimalPolicySize(); }

    /// @}


    protected:
    /// The learnt Markovian Decision Process
    FMDP< double >* fmdp_;

    /// The state in which the system is before we perform a new action
    Instantiation lastState_;

    private:
    /// The learner used to learn the FMDP
    ILearningStrategy* learner__;

    /// The planer used to plan an optimal strategy
    IPlanningStrategy< double >* planer__;

    /// The decider
    IDecisionStrategy* decider__;


    /// The number of observation we make before using again the planer
    Idx observationPhaseLenght__;

    /// The total number of observation made so far
    Idx nbObservation__;

    /// The number of Value Iteration step we perform
    Idx nbValueIterationStep__;

    /// The last performed action
    Idx lastAction__;

    /// Since SDYNA made these observation, it has to delete them on quitting
    Set< Observation* > bin__;

    bool actionReward__;

    bool verbose_;
  };


} /* namespace gum */


#endif   // GUM_SDYNA_H