PredictiveModel.h

Go to the documentation of this file.

/************************************************************************
* Verve                                                                 *
* Copyright (C) 2004-2006                                               *
* Tyler Streeter  tylerstreeter@gmail.com                               *
* All rights reserved.                                                  *
* Web: http://verve-agents.sourceforge.net                              *
*                                                                       *
* This library is free software; you can redistribute it and/or         *
* modify it under the terms of EITHER:                                  *
*   (1) The GNU Lesser General Public License as published by the Free  *
*       Software Foundation; either version 2.1 of the License, or (at  *
*       your option) any later version. The text of the GNU Lesser      *
*       General Public License is included with this library in the     *
*       file license-LGPL.txt.                                          *
*   (2) The BSD-style license that is included with this library in     *
*       the file license-BSD.txt.                                       *
*                                                                       *
* 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 files    *
* license-LGPL.txt and license-BSD.txt for more details.                *
************************************************************************/

#ifndef VERVE_PREDICTIVE_MODEL_H
#define VERVE_PREDICTIVE_MODEL_H

#include "Globals.h"
#include "Observation.h"
#include "RBFInputData.h"

// PREDICTIVE MODEL STATE REPRESENTATION (currently, METHOD 1 is used): 
// METHOD 1. predictive model uses a state-action representation 
//              (takes obs and action inputs).  This is more standard.
// METHOD 2. predictive model just uses a state representation 
//              (takes only an obs input); since the policy eventually 
//              becomes predictable, presumably the next state will 
//              become predictable.

namespace verve
{
        class Population;
        class RBFPopulation;

        class PredictiveModel
        {
        public:
                VERVE_DECL PredictiveModel(const Observation& obs, 
                        bool isDynamicRBFEnabled, unsigned int numActions);

                VERVE_DECL virtual ~PredictiveModel();

                VERVE_DECL virtual void VERVE_CALL resetShortTermMemory();

                VERVE_DECL virtual void VERVE_CALL predictAndTrain(
                        const Observation& actualPrevObs, 
                        unsigned int prevAction, const Observation& actualCurrentObs, 
                        const real actualCurrentReward, Observation& predCurrentObs, 
                        real& predCurrentReward, real& predUncertainty);

                VERVE_DECL virtual void VERVE_CALL predict(
                        const Observation& actualCurrentObs, 
                        unsigned int currentAction, Observation& predNextObs, 
                        real& predNextReward, real& predUncertainty, 
                        bool allowDynamicRBFCreation);

                VERVE_DECL virtual void VERVE_CALL changeStepSize(real newValue);

                VERVE_DECL virtual void VERVE_CALL setDeltaLearningRate(
                        real timeConstant, real stepSize);

                VERVE_DECL virtual real VERVE_CALL getPredictionMSE();

        protected:
                RBFInputData mStateActionInputData;

                real* mDiscObsTrainingData;

                RBFPopulation* mStateActionRepresentation;

                Population* mDiscObsPredPopulation;

                Population* mContObsPredPopulation;

                Population* mRewardPredPopulation;

                Population* mUncertaintyPredPopulation;

                std::vector<Population*> mAllPopulations;

                real mLatestPredMSE;

                real mDeltaLearningTimeConstant;

                real mDeltaLearningFactor;
        };
}

#endif

---