Java Code Examples for burlap.mdp.singleagent.environment.SimulatedEnvironment#resetEnvironment()

The following examples show how to use burlap.mdp.singleagent.environment.SimulatedEnvironment#resetEnvironment() . You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may check out the related API usage on the sidebar.
Example 1
Source File: QLTutorial.java    From burlap_examples with MIT License 5 votes vote down vote up 
public static void main(String[] args) {

		GridWorldDomain gwd = new GridWorldDomain(11, 11);
		gwd.setMapToFourRooms();
		gwd.setProbSucceedTransitionDynamics(0.8);
		gwd.setTf(new GridWorldTerminalFunction(10, 10));

		SADomain domain = gwd.generateDomain();

		//get initial state with agent in 0,0
		State s = new GridWorldState(new GridAgent(0, 0));

		//create environment
		SimulatedEnvironment env = new SimulatedEnvironment(domain, s);

		//create Q-learning
		QLTutorial agent = new QLTutorial(domain, 0.99, new SimpleHashableStateFactory(),
				new ConstantValueFunction(), 0.1, 0.1);

		//run Q-learning and store results in a list
		List<Episode> episodes = new ArrayList<Episode>(1000);
		for(int i = 0; i < 1000; i++){
			episodes.add(agent.runLearningEpisode(env));
			env.resetEnvironment();
		}

		Visualizer v = GridWorldVisualizer.getVisualizer(gwd.getMap());
		new EpisodeSequenceVisualizer(v, domain, episodes);

	}
Example 2
Source File: ContinuousDomainTutorial.java    From burlap_examples with MIT License 4 votes vote down vote up 
public static void MCLSPIRBF(){

		MountainCar mcGen = new MountainCar();
		SADomain domain = mcGen.generateDomain();
		MCState s = new MCState(mcGen.physParams.valleyPos(), 0.);

		NormalizedVariableFeatures inputFeatures = new NormalizedVariableFeatures()
				.variableDomain("x", new VariableDomain(mcGen.physParams.xmin, mcGen.physParams.xmax))
				.variableDomain("v", new VariableDomain(mcGen.physParams.vmin, mcGen.physParams.vmax));

		StateGenerator rStateGen = new MCRandomStateGenerator(mcGen.physParams);
		SARSCollector collector = new SARSCollector.UniformRandomSARSCollector(domain);
		SARSData dataset = collector.collectNInstances(rStateGen, domain.getModel(), 5000, 20, null);

		RBFFeatures rbf = new RBFFeatures(inputFeatures, true);
		FlatStateGridder gridder = new FlatStateGridder()
				.gridDimension("x", mcGen.physParams.xmin, mcGen.physParams.xmax, 5)
				.gridDimension("v", mcGen.physParams.vmin, mcGen.physParams.vmax, 5);

		List<State> griddedStates = gridder.gridState(s);
		DistanceMetric metric = new EuclideanDistance();
		for(State g : griddedStates){
			rbf.addRBF(new GaussianRBF(inputFeatures.features(g), metric, 0.2));
		}

		LSPI lspi = new LSPI(domain, 0.99, new DenseCrossProductFeatures(rbf, 3), dataset);
		Policy p = lspi.runPolicyIteration(30, 1e-6);

		Visualizer v = MountainCarVisualizer.getVisualizer(mcGen);
		VisualActionObserver vob = new VisualActionObserver(v);
		vob.initGUI();


		SimulatedEnvironment env = new SimulatedEnvironment(domain, s);
		env.addObservers(vob);

		for(int i = 0; i < 5; i++){
			PolicyUtils.rollout(p, env);
			env.resetEnvironment();
		}

		System.out.println("Finished");


	}
Example 3
Source File: ContinuousDomainTutorial.java    From burlap_examples with MIT License 3 votes vote down vote up 
public static void MCLSPIFB(){

		MountainCar mcGen = new MountainCar();
		SADomain domain = mcGen.generateDomain();

		StateGenerator rStateGen = new MCRandomStateGenerator(mcGen.physParams);
		SARSCollector collector = new SARSCollector.UniformRandomSARSCollector(domain);
		SARSData dataset = collector.collectNInstances(rStateGen, domain.getModel(), 5000, 20, null);

		NormalizedVariableFeatures inputFeatures = new NormalizedVariableFeatures()
				.variableDomain("x", new VariableDomain(mcGen.physParams.xmin, mcGen.physParams.xmax))
				.variableDomain("v", new VariableDomain(mcGen.physParams.vmin, mcGen.physParams.vmax));

		FourierBasis fb = new FourierBasis(inputFeatures, 4);

		LSPI lspi = new LSPI(domain, 0.99, new DenseCrossProductFeatures(fb, 3), dataset);
		Policy p = lspi.runPolicyIteration(30, 1e-6);

		Visualizer v = MountainCarVisualizer.getVisualizer(mcGen);
		VisualActionObserver vob = new VisualActionObserver(v);
		vob.initGUI();

		SimulatedEnvironment env = new SimulatedEnvironment(domain, new MCState(mcGen.physParams.valleyPos(), 0.));
		env.addObservers(vob);

		for(int i = 0; i < 5; i++){
			PolicyUtils.rollout(p, env);
			env.resetEnvironment();
		}

		System.out.println("Finished");


	}
Example 4
Source File: ContinuousDomainTutorial.java    From burlap_examples with MIT License 2 votes vote down vote up 
public static void LLSARSA(){

		LunarLanderDomain lld = new LunarLanderDomain();
		OOSADomain domain = lld.generateDomain();

		LLState s = new LLState(new LLAgent(5, 0, 0), new LLBlock.LLPad(75, 95, 0, 10, "pad"));

		ConcatenatedObjectFeatures inputFeatures = new ConcatenatedObjectFeatures()
				.addObjectVectorizion(LunarLanderDomain.CLASS_AGENT, new NumericVariableFeatures());

		int nTilings = 5;
		double resolution = 10.;

		double xWidth = (lld.getXmax() - lld.getXmin()) / resolution;
		double yWidth = (lld.getYmax() - lld.getYmin()) / resolution;
		double velocityWidth = 2 * lld.getVmax() / resolution;
		double angleWidth = 2 * lld.getAngmax() / resolution;



		TileCodingFeatures tilecoding = new TileCodingFeatures(inputFeatures);
		tilecoding.addTilingsForAllDimensionsWithWidths(
				new double []{xWidth, yWidth, velocityWidth, velocityWidth, angleWidth},
				nTilings,
				TilingArrangement.RANDOM_JITTER);




		double defaultQ = 0.5;
		DifferentiableStateActionValue vfa = tilecoding.generateVFA(defaultQ/nTilings);
		GradientDescentSarsaLam agent = new GradientDescentSarsaLam(domain, 0.99, vfa, 0.02, 0.5);

		SimulatedEnvironment env = new SimulatedEnvironment(domain, s);
		List<Episode> episodes = new ArrayList<Episode>();
		for(int i = 0; i < 5000; i++){
			Episode ea = agent.runLearningEpisode(env);
			episodes.add(ea);
			System.out.println(i + ": " + ea.maxTimeStep());
			env.resetEnvironment();
		}

		Visualizer v = LLVisualizer.getVisualizer(lld.getPhysParams());
		new EpisodeSequenceVisualizer(v, domain, episodes);

	}