org.ggp.base.player.gamer.statemachine.sample.SampleMonteCarloGamer Maven / Gradle / Ivy

Go to download
Show more of this group Show more artifacts with this name
Show all versions of alloy-ggp-base Show documentation
A modified version of the GGP-Base library for Alloy.
The newest version!
package org.ggp.base.player.gamer.statemachine.sample;

import java.util.List;

import org.ggp.base.player.gamer.event.GamerSelectedMoveEvent;
import org.ggp.base.util.statemachine.MachineState;
import org.ggp.base.util.statemachine.Move;
import org.ggp.base.util.statemachine.StateMachine;
import org.ggp.base.util.statemachine.exceptions.GoalDefinitionException;
import org.ggp.base.util.statemachine.exceptions.MoveDefinitionException;
import org.ggp.base.util.statemachine.exceptions.TransitionDefinitionException;

/**
 * SampleMonteCarloGamer is a simple state-machine-based Gamer. It will use a
 * pure Monte Carlo approach towards picking moves, doing simulations and then
 * choosing the move that has the highest expected score. It should be slightly
 * more challenging than the RandomGamer, while still playing reasonably fast.
 *
 * However, right now it isn't challenging at all. It's extremely mediocre, and
 * doesn't even block obvious one-move wins. This is partially due to the speed
 * of the default state machine (which is slow) and mostly due to the algorithm
 * assuming that the opponent plays completely randomly, which is inaccurate.
 *
 * @author Sam Schreiber
 */
public final class SampleMonteCarloGamer extends SampleGamer
{
    /**
     * Employs a simple sample "Monte Carlo" algorithm.
     */
    @Override
    public Move stateMachineSelectMove(long timeout) throws TransitionDefinitionException, MoveDefinitionException, GoalDefinitionException
    {
        StateMachine theMachine = getStateMachine();
        long start = System.currentTimeMillis();
        long finishBy = timeout - 1000;

        List moves = theMachine.getLegalMoves(getCurrentState(), getRole());
        Move selection = moves.get(0);
        if (moves.size() > 1) {
            int[] moveTotalPoints = new int[moves.size()];
            int[] moveTotalAttempts = new int[moves.size()];

            // Perform depth charges for each candidate move, and keep track
            // of the total score and total attempts accumulated for each move.
            for (int i = 0; true; i = (i+1) % moves.size()) {
                if (System.currentTimeMillis() > finishBy)
                    break;

                int theScore = performDepthChargeFromMove(getCurrentState(), moves.get(i));
                moveTotalPoints[i] += theScore;
                moveTotalAttempts[i] += 1;
            }

            // Compute the expected score for each move.
            double[] moveExpectedPoints = new double[moves.size()];
            for (int i = 0; i < moves.size(); i++) {
                moveExpectedPoints[i] = (double)moveTotalPoints[i] / moveTotalAttempts[i];
            }

            // Find the move with the best expected score.
            int bestMove = 0;
            double bestMoveScore = moveExpectedPoints[0];
            for (int i = 1; i < moves.size(); i++) {
                if (moveExpectedPoints[i] > bestMoveScore) {
                    bestMoveScore = moveExpectedPoints[i];
                    bestMove = i;
                }
            }
            selection = moves.get(bestMove);
        }

        long stop = System.currentTimeMillis();

        notifyObservers(new GamerSelectedMoveEvent(moves, selection, stop - start));
        return selection;
    }

    private int[] depth = new int[1];
    int performDepthChargeFromMove(MachineState theState, Move myMove) {
        StateMachine theMachine = getStateMachine();
        try {
            List goalValues = theMachine.performDepthCharge(theMachine.getRandomNextState(theState, getRole(), myMove), depth);
            return goalValues.get(getStateMachine().getRoles().indexOf(getRole()));
        } catch (Exception e) {
            e.printStackTrace();
            return 0;
        }
    }
}