cmontecarlo.cpp

强化学习算法（R-Learning）难得的珍贵资料

// Copyright (C) 2003
// Gerhard Neumann ([e-mail])

//                
// This file is part of RL Toolbox.
// http://www.igi.tugraz.at/ril_toolbox
//
// All rights reserved.
// 
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
// 1. Redistributions of source code must retain the above copyright
//    notice, this list of conditions and the following disclaimer.
// 2. Redistributions in binary form must reproduce the above copyright
//    notice, this list of conditions and the following disclaimer in the
//    documentation and/or other materials provided with the distribution.
// 3. The name of the author may not be used to endorse or promote products
//    derived from this software without specific prior written permission.
// 
// THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
// OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
// IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
// NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.


#include "cmontecarlo.h"
#include <math.h>

CPolicyEvaluator::CPolicyEvaluator(CAgent *agent, CRewardFunction *rewardFunction, int nEpisodes, int nStepsPerEpisode) : CSemiMDPRewardListener(rewardFunction)
{
	this->agent = agent;
	this->nEpisodes = nEpisodes;
	this->nStepsPerEpisode = nStepsPerEpisode;
}

rlt_real CPolicyEvaluator::evaluatePolicy()
{
	rlt_real value = 0;
	agent->addSemiMDPListener(this);
	for (int i = 0; i < nEpisodes; i ++)
	{
		agent->startNewEpisode();
		agent->doControllerEpisode(1, nStepsPerEpisode);
		value += this->getEpisodeValue();
	}
	value /= nEpisodes;
	agent->removeSemiMDPListener(this);
	return value;
}

CAverageRewardCalculator::CAverageRewardCalculator(CAgent *agent, CRewardFunction *rewardFunction, int nEpisodes, int nStepsPerEpisode, rlt_real minReward) : CPolicyEvaluator(agent, rewardFunction, nEpisodes, nStepsPerEpisode)
{
	nSteps = 0;
	averageReward = 0;
	this->minReward = minReward;
}

rlt_real CAverageRewardCalculator::getEpisodeValue()
{
	if (agent->getEnvironmentModel()->isFailed())
	{
		averageReward += minReward * (nStepsPerEpisode - nSteps);
		nSteps = nStepsPerEpisode;
	}
	return averageReward / nSteps;
}


void CAverageRewardCalculator::nextStep(CStateCollection *oldState, CAction *action, rlt_real reward, CStateCollection *nextState)
{
	averageReward += reward;
	nSteps ++;
}

void CAverageRewardCalculator::newEpisode()
{
	averageReward = 0;
	nSteps = 0;
}

CValueCalculator::CValueCalculator(CAgent *agent, CRewardFunction *rewardFunction, int nEpisodes, int nStepsPerEpisode, rlt_real gamma) : CPolicyEvaluator(agent, rewardFunction, nEpisodes, nStepsPerEpisode)
{
	nSteps = 0;
	value = 0;
	addParameter("DiscountFactor", gamma);
}

rlt_real CValueCalculator::getEpisodeValue()
{
	return value;
}


void CValueCalculator::nextStep(CStateCollection *oldState, CAction *action, rlt_real reward, CStateCollection *nextState)
{
	value += pow(getParameter("DiscountFactor"), nSteps) * reward;
	nSteps += action->getDuration();
}

void CValueCalculator::newEpisode()
{
	value = 0;
	nSteps = 0;
}

CPolicySameStateEvaluator::CPolicySameStateEvaluator(CAgent *agent, CRewardFunction *rewardFunction, CTransitionFunctionEnvironment *environment, CStateList *startStates, int nStepsPerEpisode) : CPolicyEvaluator(agent, rewardFunction, startStates->getNumStates(), nStepsPerEpisode)
{
	this->startStates = startStates;
	this->environment = environment;
}

rlt_real CPolicySameStateEvaluator::evaluatePolicy()
{
	rlt_real value = 0;
	agent->addSemiMDPListener(this);
	CState *startState = new CState(environment->getStateProperties());
	for (int i = 0; i < nEpisodes; i ++)
	{
		agent->startNewEpisode();
		startStates->getState(i, startState);
		environment->setState(startState);
		agent->doControllerEpisode(1, nStepsPerEpisode);
		value += this->getEpisodeValue();
	}
	value /= nEpisodes;
	agent->removeSemiMDPListener(this);
	delete startState;
	return value;
}

void CPolicySameStateEvaluator::getNewStartStates()
{
	int numStartStates = startStates->getNumStates();

	CState *startState = new CState(environment->getStateProperties());

	startStates->clear();

	for (int i = 0; i < numStartStates; i++)
	{
		environment->resetModel();
		environment->getState(startState);
		startStates->addState(startState);
	}
	delete startState;

}

CAverageRewardSameStateCalculator::CAverageRewardSameStateCalculator(CAgent *agent, CRewardFunction *rewardFunction, CTransitionFunctionEnvironment *environment, CStateList *startStates, int nStepsPerEpisode, rlt_real minReward) : CPolicySameStateEvaluator(agent, rewardFunction, environment, startStates, nStepsPerEpisode)
{
	nSteps = 0;
	averageReward = 0;

	this->minReward = minReward;

}

rlt_real CAverageRewardSameStateCalculator::getEpisodeValue()
{
	if (agent->getEnvironmentModel()->isFailed())
	{
		averageReward += minReward * (nStepsPerEpisode - nSteps);
		nSteps = nStepsPerEpisode;
	}
	return averageReward / nSteps;
}


void CAverageRewardSameStateCalculator::nextStep(CStateCollection *oldState, CAction *action, rlt_real reward, CStateCollection *nextState)
{
	averageReward += reward;
	nSteps ++;
}

void CAverageRewardSameStateCalculator::newEpisode()
{
	averageReward = 0;
	nSteps = 0;
}

CValueSameStateCalculator::CValueSameStateCalculator(CAgent *agent, CRewardFunction *rewardFunction, CTransitionFunctionEnvironment *environment, CStateList *startStates, int nStepsPerEpisode, rlt_real gamma) : CPolicySameStateEvaluator(agent, rewardFunction, environment, startStates, nStepsPerEpisode)
{
	nSteps = 0;
	value = 0;
	addParameter("DiscountFactor", gamma);
}

rlt_real CValueSameStateCalculator::getEpisodeValue()
{
	return value;
}


void CValueSameStateCalculator::nextStep(CStateCollection *oldState, CAction *action, rlt_real reward, CStateCollection *nextState)
{
	value += pow(getParameter("DiscountFactor"), nSteps) * reward;
	nSteps += action->getDuration();
}

void CValueSameStateCalculator::newEpisode()
{
	value = 0;
	nSteps = 0;
}

CPolicyGreedynessEvaluator::CPolicyGreedynessEvaluator(CAgent *agent, CRewardFunction *reward, int nEpisodes, int nStepsPerEpsiode, CAgentController *l_greedyPolicy) : CPolicyEvaluator(agent, reward, nEpisodes, nStepsPerEpsiode)
{
	nGreedyActions = 0;
	greedyPolicy = l_greedyPolicy;

	actionDataSet = new CActionDataSet(greedyPolicy->getActions());
}

CPolicyGreedynessEvaluator::~CPolicyGreedynessEvaluator()
{
	delete actionDataSet;
}

rlt_real CPolicyGreedynessEvaluator::getEpisodeValue()
{
	return (rlt_real) nGreedyActions / (rlt_real) nStepsPerEpisode;
}

void CPolicyGreedynessEvaluator::nextStep(CStateCollection *oldState, CAction *action, rlt_real reward, CStateCollection *nextState)
{
	CAction *greedyAction = greedyPolicy->getNextAction(oldState);
	if (action->isSameAction(greedyAction, actionDataSet->getActionData(greedyAction)))
	{
		nGreedyActions ++;
	}
}

void CPolicyGreedynessEvaluator::newEpisode()
{
	nGreedyActions = 0;
}