burlap.mdp.auxiliary.common.SinglePFTF Java Examples

public GridWorldDQN(String solverFile, double gamma) {

        //create the domain
        gwdg = new GridWorldDomain(11, 11);
        gwdg.setMapToFourRooms();
        rf = new UniformCostRF();
        tf = new SinglePFTF(PropositionalFunction.findPF(gwdg.generatePfs(), GridWorldDomain.PF_AT_LOCATION));
        gwdg.setRf(rf);
        gwdg.setTf(tf);
        domain = gwdg.generateDomain();

        goalCondition = new TFGoalCondition(tf);

        //set up the initial state of the task
        initialState = new GridWorldState(new GridAgent(0, 0), new GridLocation(10, 10, "loc0"));

        //set up the state hashing system for tabular algorithms
        hashingFactory = new SimpleHashableStateFactory();

        //set up the environment for learners algorithms
        env = new SimulatedEnvironment(domain, initialState);

        dqn = new DQN(solverFile, actionSet, new NNGridStateConverter(), gamma);
    }
 

@Before
public void setup() {
	this.gw = new GridWorldDomain(11, 11);
	this.gw.setMapToFourRooms();
	this.gw.setRf(new UniformCostRF());
	TerminalFunction tf = new SinglePFTF(PropositionalFunction.findPF(gw.generatePfs(), PF_AT_LOCATION));
	this.gw.setTf(tf);
	this.domain = this.gw.generateDomain();
	this.goalCondition = new TFGoalCondition(tf);
	this.hashingFactory = new SimpleHashableStateFactory();
}
 

@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;
}
 

public static void main(String [] args){

		GridWorldDomain gw = new GridWorldDomain(11,11); //11x11 grid world
		gw.setMapToFourRooms(); //four rooms layout
		gw.setProbSucceedTransitionDynamics(0.8); //stochastic transitions with 0.8 success rate

		//ends when the agent reaches a location
		final TerminalFunction tf = new SinglePFTF(
				PropositionalFunction.findPF(gw.generatePfs(), GridWorldDomain.PF_AT_LOCATION));

		//reward function definition
		final RewardFunction rf = new GoalBasedRF(new TFGoalCondition(tf), 5., -0.1);

		gw.setTf(tf);
		gw.setRf(rf);


		final OOSADomain domain = gw.generateDomain(); //generate the grid world domain

		//setup initial state
		GridWorldState s = new GridWorldState(new GridAgent(0, 0), new GridLocation(10, 10, "loc0"));



		//initial state generator
		final ConstantStateGenerator sg = new ConstantStateGenerator(s);


		//set up the state hashing system for looking up states
		final SimpleHashableStateFactory hashingFactory = new SimpleHashableStateFactory();


		/**
		 * Create factory for Q-learning agent
		 */
		LearningAgentFactory qLearningFactory = new LearningAgentFactory() {

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

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

		//define learning environment
		SimulatedEnvironment env = new SimulatedEnvironment(domain, sg);

		//define experiment
		LearningAlgorithmExperimenter exp = new LearningAlgorithmExperimenter(env,
				10, 100, qLearningFactory);

		exp.setUpPlottingConfiguration(500, 250, 2, 1000, TrialMode.MOST_RECENT_AND_AVERAGE,
				PerformanceMetric.CUMULATIVE_STEPS_PER_EPISODE, PerformanceMetric.AVERAGE_EPISODE_REWARD);


		//start experiment
		exp.startExperiment();


	}