Examples with StateObservation core.game.StateObservation used on opensource projects

Example 6 with StateObservation

use of core.game.StateObservation in project SimpleAsteroids by ljialin.

the class SingleTreeNode method mctsSearch.

public void mctsSearch(ElapsedCpuTimer elapsedTimer) {
    double avgTimeTaken = 0;
    double acumTimeTaken = 0;
    long remaining = elapsedTimer.remainingTimeMillis();
    int numIters = 0;
    int remainingLimit = 10;
    // while(remaining > 2*avgTimeTaken && remaining > remainingLimit){
    while (numIters < Agent.MCTS_ITERATIONS) {
        StateObservation state = rootState.copy();
        ElapsedCpuTimer elapsedTimerIteration = new ElapsedCpuTimer();
        SingleTreeNode selected = treePolicy(state);
        double delta = selected.rollOut(state);
        backUp(selected, delta);
        numIters++;
        acumTimeTaken += (elapsedTimerIteration.elapsedMillis());
        // System.out.println(elapsedTimerIteration.elapsedMillis() + " --> " + acumTimeTaken + " (" + remaining + ")");
        avgTimeTaken = acumTimeTaken / numIters;
        remaining = elapsedTimer.remainingTimeMillis();
    }
// System.out.println("Iterations: " + numIters);
}

Example 7 with StateObservation

use of core.game.StateObservation in project SimpleAsteroids by ljialin.

the class Agent method act.

public Types.ACTIONS act(StateObservation stateObs, ElapsedCpuTimer elapsedTimer) {
    // Set the state observation object as the new root of the tree.
    // we'll set up a game adapter and run the algorithm independently each
    // time at least to being with
    StateObservation obs = stateObs.copy();
    // Types.ACTIONS[] moveSeq = new Types.ACTIONS[maxRolloutLength];
    bestRollout = new Types.ACTIONS[maxNestingDepth][maxRolloutLength];
    lengthBestRollout = new int[maxNestingDepth];
    scoreBestRollout = new double[maxNestingDepth];
    // nested(obs, nestDepth, moveSeq, 0);
    double bestScore = Double.NEGATIVE_INFINITY;
    Types.ACTIONS bestAction = actions[0];
    for (int i = 0; i < num_actions; i++) {
        StateObservation state = obs.copy();
        Types.ACTIONS[] moveSeqCopy = new Types.ACTIONS[maxRolloutLength];
        int nActionsPlayed = 0;
        state.advance(actions[i]);
        moveSeqCopy[nActionsPlayed] = actions[i];
        nActionsPlayed++;
        nested(state, nestDepth, moveSeqCopy, nActionsPlayed);
        double score = state.getGameScore();
        if (score > bestScore) {
            bestScore = score;
            bestAction = actions[i];
        }
    }
    return bestAction;
}