burlap.mdp.singleagent.model.FactoredModel Java Examples

@Override
public SADomain generateDomain() {

	SADomain domain = new SADomain();


	domain.addActionTypes(
			new UniversalActionType(ACTION_NORTH),
			new UniversalActionType(ACTION_SOUTH),
			new UniversalActionType(ACTION_EAST),
			new UniversalActionType(ACTION_WEST));

	GridWorldStateModel smodel = new GridWorldStateModel();
	RewardFunction rf = new ExampleRF(this.goalx, this.goaly);
	TerminalFunction tf = new ExampleTF(this.goalx, this.goaly);

	domain.setModel(new FactoredModel(smodel, rf, tf));

	return domain;
}
 

@Override
public OOSADomain generateDomain() {

	OOSADomain domain = new OOSADomain();
	
	domain.addStateClass(CLASS_BLOCK, BlocksWorldBlock.class);

	domain.addActionType(new StackActionType(ACTION_STACK))
			.addActionType(new UnstackActionType(ACTION_UNSTACK));

	RewardFunction rf = this.rf;
	TerminalFunction tf = this.tf;

	if(rf == null){
		rf = new NullRewardFunction();
	}
	if(tf == null){
		tf = new NullTermination();
	}

	BWModel smodel = new BWModel();
	FactoredModel model = new FactoredModel(smodel, rf , tf);
	domain.setModel(model);

	OODomain.Helper.addPfsToDomain(domain, this.generatePfs());
	
	return domain;
}
 

@Override
public SADomain generateDomain() {
	
	SADomain domain = new SADomain();


	IPPhysicsParams cphys = this.physParams.copy();
	IPModel smodel = new IPModel(cphys);

	RewardFunction rf = this.rf;
	TerminalFunction tf = this.tf;

	if(rf == null){
		rf = new InvertedPendulumRewardFunction();
	}
	if(tf == null){
		tf = new InvertedPendulumTerminalFunction();
	}

	FactoredModel model = new FactoredModel(smodel, rf ,tf);
	domain.setModel(model);

	domain.addActionType(new UniversalActionType(ACTION_LEFT))
			.addActionType(new UniversalActionType(ACTION_RIGHT))
			.addActionType(new UniversalActionType(ACTION_NO_FORCE));

	
	return domain;
}
 

public void experimentAndPlotter(){

		//different reward function for more structured performance plots
		((FactoredModel)domain.getModel()).setRf(new GoalBasedRF(this.goalCondition, 5.0, -0.1));

		/**
		 * Create factories for Q-learning agent and SARSA agent to compare
		 */
		LearningAgentFactory qLearningFactory = new LearningAgentFactory() {

			public String getAgentName() {
				return "Q-Learning";
			}


			public LearningAgent generateAgent() {
				return new QLearning(domain, 0.99, hashingFactory, 0.3, 0.1);
			}
		};

		LearningAgentFactory sarsaLearningFactory = new LearningAgentFactory() {

			public String getAgentName() {
				return "SARSA";
			}


			public LearningAgent generateAgent() {
				return new SarsaLam(domain, 0.99, hashingFactory, 0.0, 0.1, 1.);
			}
		};

		LearningAlgorithmExperimenter exp = new LearningAlgorithmExperimenter(env, 10, 100, qLearningFactory, sarsaLearningFactory);
		exp.setUpPlottingConfiguration(500, 250, 2, 1000,
				TrialMode.MOST_RECENT_AND_AVERAGE,
				PerformanceMetric.CUMULATIVE_STEPS_PER_EPISODE,
				PerformanceMetric.AVERAGE_EPISODE_REWARD);

		exp.startExperiment();
		exp.writeStepAndEpisodeDataToCSV("expData");

	}
 

@Override
public OOSADomain generateDomain() {

	OOSADomain domain = new OOSADomain();

	domain.addStateClass(CLASS_AGENT, ExGridAgent.class)
			.addStateClass(CLASS_LOCATION, EXGridLocation.class);

	domain.addActionTypes(
			new UniversalActionType(ACTION_NORTH),
			new UniversalActionType(ACTION_SOUTH),
			new UniversalActionType(ACTION_EAST),
			new UniversalActionType(ACTION_WEST));


	OODomain.Helper.addPfsToDomain(domain, this.generatePfs());

	OOGridWorldStateModel smodel = new OOGridWorldStateModel();
	RewardFunction rf = new SingleGoalPFRF(domain.propFunction(PF_AT), 100, -1);
	TerminalFunction tf = new SinglePFTF(domain.propFunction(PF_AT));

	domain.setModel(new FactoredModel(smodel, rf, tf));


	return domain;
}
 

@Override
public SADomain generateDomain() {
	
	SADomain domain = new SADomain();


	MCModel smodel = new MCModel(this.physParams.copy());
	if(tf == null){
		tf = new ClassicMCTF(physParams.xmax);
	}
	if(rf == null){
		rf = new GoalBasedRF(tf, 100, 0);
	}

	FactoredModel model = new FactoredModel(smodel, rf, tf);

	domain.setModel(model);

	domain.addActionType(new UniversalActionType(ACTION_FORWARD))
			.addActionType(new UniversalActionType(ACTION_BACKWARDS))
			.addActionType(new UniversalActionType(ACTION_COAST));

	
	
	return domain;
}
 

@Override
public OOSADomain generateDomain() {
	
	OOSADomain domain = new OOSADomain();
	
	List <Double> thrustValuesTemp = this.thrustValues;
	if(thrustValuesTemp.isEmpty()){
		thrustValuesTemp.add(0.32);
		thrustValuesTemp.add(-physParams.gravity);
	}
	
	domain.addStateClass(CLASS_AGENT, LLAgent.class)
			.addStateClass(CLASS_PAD, LLBlock.LLPad.class)
			.addStateClass(CLASS_OBSTACLE, LLBlock.LLObstacle.class);

	//make copy of physics parameters
	LLPhysicsParams cphys = this.physParams.copy();
	
	//add actions
	domain.addActionType(new UniversalActionType(ACTION_TURN_LEFT))
			.addActionType(new UniversalActionType(ACTION_TURN_RIGHT))
			.addActionType(new UniversalActionType(ACTION_IDLE))
			.addActionType(new ThrustType(thrustValues));


	OODomain.Helper.addPfsToDomain(domain, this.generatePfs());

	LunarLanderModel smodel = new LunarLanderModel(cphys);
	RewardFunction rf = this.rf;
	TerminalFunction tf = this.tf;
	if(rf == null){
		rf = new LunarLanderRF(domain);
	}
	if(tf == null){
		tf = new LunarLanderTF(domain);
	}

	FactoredModel model = new FactoredModel(smodel, rf, tf);
	domain.setModel(model);
	
	return domain;
	
}
 

@Override
public SADomain generateDomain() {
	
	SADomain domain = new SADomain();

	Map<Integer, Map<Integer, Set<NodeTransitionProbability>>> ctd = this.copyTransitionDynamics();

	GraphStateModel stateModel = new GraphStateModel(ctd);
	FactoredModel model = new FactoredModel(stateModel, rf, tf);

	domain.setModel(model);

	for(int i = 0; i < this.maxActions; i++){
		domain.addActionType(new GraphActionType(i, ctd));
	}
	
	
	return domain;
}
 

/**
 * Creates a new frostbite domain.
 *
 * @return the generated domain object
 */
@Override
public OOSADomain generateDomain() {

	OOSADomain domain = new OOSADomain();

	domain.addStateClass(CLASS_AGENT, FrostbiteAgent.class)
			.addStateClass(CLASS_IGLOO, FrostbiteIgloo.class)
			.addStateClass(CLASS_PLATFORM, FrostbitePlatform.class);

	//add actions
	domain.addActionType(new UniversalActionType(ACTION_NORTH))
			.addActionType(new UniversalActionType(ACTION_SOUTH))
			.addActionType(new UniversalActionType(ACTION_EAST))
			.addActionType(new UniversalActionType(ACTION_WEST))
			.addActionType(new UniversalActionType(ACTION_IDLE));



	//add pfs
	List<PropositionalFunction> pfs = this.generatePFs();
	for(PropositionalFunction pf : pfs){
		domain.addPropFunction(pf);
	}


	FrostbiteModel smodel = new FrostbiteModel(scale);
	RewardFunction rf = this.rf;
	TerminalFunction tf = this.tf;
	if(rf == null){
		rf = new FrostbiteRF(domain);
	}
	if(tf == null){
		tf = new FrostbiteTF(domain);
	}


	FactoredModel model = new FactoredModel(smodel, rf, tf);
	domain.setModel(model);

	return domain;
}
 

@Override
public SADomain generateDomain() {
	
	SADomain domain = new SADomain();

	CPPhysicsParams cphys = this.physParams.copy();

	RewardFunction rf = this.rf;
	TerminalFunction tf = this.tf;

	if(rf == null){
		rf = new CartPoleRewardFunction();
	}
	if(tf == null){
		tf = new CartPoleTerminalFunction();
	}

	FullStateModel smodel = cphys.useCorrectModel ? new CPClassicModel(cphys) : new CPClassicModel(cphys);

	FactoredModel model = new FactoredModel(smodel, rf, tf);
	domain.setModel(model);


	domain.addActionType(new UniversalActionType(ACTION_LEFT))
			.addActionType(new UniversalActionType(ACTION_RIGHT));


	return domain;
}