stateactioninfo.java

来自「Reinforcement Learning」· Java 代码 · 共 61 行

/*
 * Created on Dec 21, 2004
 *
 * TODO To change the template for this generated file go to
 * Window - Preferences - Java - Code Style - Code Templates
 */

/**
 * @author ryk
 *
 * TODO To change the template for this generated type comment go to
 * Window - Preferences - Java - Code Style - Code Templates
 */
import java.util.Vector;

public class StateActionInfo {

	int actionCounter;	//maintains the number of times this action has been taken
	Vector succs;		//maintains successors of this state and numofvisits to each successor	
	double totalReward;
	double qsa;			//q-value associated with this this action 
	
	//constructor for predecessor and successor
	public StateActionInfo()
	{
		actionCounter = 0;
		succs = new Vector();
		qsa = 0;
		totalReward = 0;
	}
	
	public void updateData(State succSt, double reward)
	{
		totalReward += reward;
		actionCounter+=1;
		addSuccessor(succSt);
	}
	
	private void addSuccessor(State succSt)
	{
		StateAndVisits snv = new StateAndVisits(succSt);
		int positionOfNext = succs.indexOf(snv);

		if(-1 == positionOfNext)
			succs.add(snv);
		else
		{
			snv = (StateAndVisits)succs.get(positionOfNext);//snv points to that particular element in the vector
			snv.incrementVisits();
		}
	}

	public double getReward()
	{
		if(totalReward==0)
			return 0;
		
		return totalReward/actionCounter;
	}
}