Java SparseSampling.toggleDebugPrinting方法代码示例

import burlap.behavior.singleagent.planning.stochastic.sparsesampling.SparseSampling; //导入方法依赖的package包/类
public static void IPSS(){

		InvertedPendulum ip = new InvertedPendulum();
		ip.physParams.actionNoise = 0.;
		Domain domain = ip.generateDomain();
		RewardFunction rf = new InvertedPendulum.InvertedPendulumRewardFunction(Math.PI/8.);
		TerminalFunction tf = new InvertedPendulum.InvertedPendulumTerminalFunction(Math.PI/8.);
		State initialState = InvertedPendulum.getInitialState(domain);

		SparseSampling ss = new SparseSampling(domain, rf, tf, 1, new SimpleHashableStateFactory(), 10 ,1);
		ss.setForgetPreviousPlanResults(true);
		ss.toggleDebugPrinting(false);
		Policy p = new GreedyQPolicy(ss);

		EpisodeAnalysis ea = p.evaluateBehavior(initialState, rf, tf, 500);
		System.out.println("Num steps: " + ea.maxTimeStep());
		Visualizer v = InvertedPendulumVisualizer.getInvertedPendulumVisualizer();
		new EpisodeSequenceVisualizer(v, domain, Arrays.asList(ea));

	}
 

import burlap.behavior.singleagent.planning.stochastic.sparsesampling.SparseSampling; //导入方法依赖的package包/类
public static void IPSS(){

		InvertedPendulum ip = new InvertedPendulum();
		ip.physParams.actionNoise = 0.;
		RewardFunction rf = new InvertedPendulum.InvertedPendulumRewardFunction(Math.PI/8.);
		TerminalFunction tf = new InvertedPendulum.InvertedPendulumTerminalFunction(Math.PI/8.);
		ip.setRf(rf);
		ip.setTf(tf);
		SADomain domain = ip.generateDomain();

		State initialState = new InvertedPendulumState();

		SparseSampling ss = new SparseSampling(domain, 1, new SimpleHashableStateFactory(), 10, 1);
		ss.setForgetPreviousPlanResults(true);
		ss.toggleDebugPrinting(false);
		Policy p = new GreedyQPolicy(ss);

		Episode e = PolicyUtils.rollout(p, initialState, domain.getModel(), 500);
		System.out.println("Num steps: " + e.maxTimeStep());
		Visualizer v = CartPoleVisualizer.getCartPoleVisualizer();
		new EpisodeSequenceVisualizer(v, domain, Arrays.asList(e));

	}
 

import burlap.behavior.singleagent.planning.stochastic.sparsesampling.SparseSampling; //导入方法依赖的package包/类
/**
 * Runs a single iteration of value iteration. Note that if the state samples have not been set, it will throw a runtime exception.
 * @return the maximum change in the value function.
 */
public double runIteration(){

	if(this.samples == null){
		throw new RuntimeException("FittedVI cannot run value iteration because the state samples have not been set. Use the setSamples method or the constructor to set them.");
	}

	SparseSampling ss = new SparseSampling(this.domain, this.rf, this.tf, this.gamma, this.hashingFactory, this.planningDepth, this.transitionSamples);
	ss.setValueForLeafNodes(this.leafNodeInit);
	ss.toggleDebugPrinting(false);

	List <SupervisedVFA.SupervisedVFAInstance> instances = new ArrayList<SupervisedVFA.SupervisedVFAInstance>(this.samples.size());
	List <Double> oldVs = new ArrayList<Double>(this.samples.size());
	for(State s : this.samples){
		oldVs.add(this.valueFunction.value(s));
		instances.add(new SupervisedVFA.SupervisedVFAInstance(s, QFunctionHelper.getOptimalValue(ss, s)));
	}

	this.valueFunction = this.valueFunctionTrainer.train(instances);

	double maxDiff = 0.;
	for(int i = 0; i < this.samples.size(); i++){
		double newV = this.valueFunction.value(this.samples.get(i));
		double diff = Math.abs(newV - oldVs.get(i));
		maxDiff = Math.max(maxDiff, diff);
	}

	return maxDiff;

}
 

import burlap.behavior.singleagent.planning.stochastic.sparsesampling.SparseSampling; //导入方法依赖的package包/类
@Override
public List<QValue> getQs(State s) {
	SparseSampling ss = new SparseSampling(this.domain, this.rf, this.tf, this.gamma, this.hashingFactory, this.controlDepth, this.transitionSamples);
	ss.setValueForLeafNodes(this.leafNodeInit);
	ss.toggleDebugPrinting(false);
	return ss.getQs(s);
}
 

import burlap.behavior.singleagent.planning.stochastic.sparsesampling.SparseSampling; //导入方法依赖的package包/类
@Override
public QValue getQ(State s, AbstractGroundedAction a) {
	SparseSampling ss = new SparseSampling(this.domain, this.rf, this.tf, this.gamma, this.hashingFactory, this.controlDepth, this.transitionSamples);
	ss.setValueForLeafNodes(this.leafNodeInit);
	ss.toggleDebugPrinting(false);
	return ss.getQ(s, a);
}
 

import burlap.behavior.singleagent.planning.stochastic.sparsesampling.SparseSampling; //导入方法依赖的package包/类
/**
 * Runs a single iteration of value iteration. Note that if the state samples have not been set, it will throw a runtime exception.
 * @return the maximum change in the value function.
 */
public double runIteration(){

	if(this.samples == null){
		throw new RuntimeException("FittedVI cannot run value iteration because the state samples have not been set. Use the setSamples method or the constructor to set them.");
	}

	SparseSampling ss = new SparseSampling(this.domain, this.gamma, this.hashingFactory, this.planningDepth, this.transitionSamples);
	ss.setModel(this.model);
	ss.setValueForLeafNodes(this.leafNodeInit);
	ss.toggleDebugPrinting(false);

	List <SupervisedVFA.SupervisedVFAInstance> instances = new ArrayList<SupervisedVFA.SupervisedVFAInstance>(this.samples.size());
	List <Double> oldVs = new ArrayList<Double>(this.samples.size());
	for(State s : this.samples){
		oldVs.add(this.valueFunction.value(s));
		instances.add(new SupervisedVFA.SupervisedVFAInstance(s, Helper.maxQ(ss, s)));
	}

	this.valueFunction = this.valueFunctionTrainer.train(instances);

	double maxDiff = 0.;
	for(int i = 0; i < this.samples.size(); i++){
		double newV = this.valueFunction.value(this.samples.get(i));
		double diff = Math.abs(newV - oldVs.get(i));
		maxDiff = Math.max(maxDiff, diff);
	}

	return maxDiff;

}
 

import burlap.behavior.singleagent.planning.stochastic.sparsesampling.SparseSampling; //导入方法依赖的package包/类
@Override
public List<QValue> qValues(State s) {
	SparseSampling ss = new SparseSampling(this.domain, this.gamma, this.hashingFactory, this.controlDepth, this.transitionSamples);
	ss.setModel(model);
	ss.setValueForLeafNodes(this.leafNodeInit);
	ss.toggleDebugPrinting(false);
	return ss.qValues(s);
}
 

import burlap.behavior.singleagent.planning.stochastic.sparsesampling.SparseSampling; //导入方法依赖的package包/类
@Override
public double qValue(State s, Action a) {
	SparseSampling ss = new SparseSampling(this.domain, this.gamma, this.hashingFactory, this.controlDepth, this.transitionSamples);
	ss.setModel(model);
	ss.setValueForLeafNodes(this.leafNodeInit);
	ss.toggleDebugPrinting(false);
	return ss.qValue(s, a);
}