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

Example 1 with StateObservation

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

the class SimpleMaxNTest method runOnce.

public static double runOnce() {
    // make an agent to test
    StateObservation noiseFree = new SimpleMaxGame();
    // new NoisyMaxGame();
    StateObservation stateObs = new SimpleMaxGame();
    System.out.println(stateObs.getGameScore());
    System.out.println(stateObs.copy().getGameScore());
    // System.exit(0);
    ElapsedCpuTimer timer = new ElapsedCpuTimer();
    AbstractPlayer player;
    controllers.singlePlayer.sampleOLMCTS.Agent olmcts = new controllers.singlePlayer.sampleOLMCTS.Agent(stateObs, timer);
    controllers.singlePlayer.discountOLMCTS.Agent discountOlmcts = new controllers.singlePlayer.discountOLMCTS.Agent(stateObs, timer);
    controllers.singlePlayer.nestedMC.Agent nestedMC = new controllers.singlePlayer.nestedMC.Agent(stateObs, timer);
    player = olmcts;
    player = discountOlmcts;
    // for the following we can pass the Evolutionary algorithm to use
    int nResamples = 2;
    EvoAlg evoAlg = new SimpleRMHC(nResamples);
    int nEvals = 1000;
    double kExplore = 10;
    int nNeighbours = 100;
    evoAlg = new NTupleBanditEA(kExplore, nNeighbours);
    // DefaultMutator.totalRandomChaosMutation = true;
    Agent.useShiftBuffer = false;
    controllers.singlePlayer.ea.Agent.SEQUENCE_LENGTH = 100;
    player = new controllers.singlePlayer.ea.Agent(stateObs, timer, evoAlg, nEvals);
    nestedMC.maxRolloutLength = 5;
    nestedMC.nestDepth = 5;
    player = nestedMC;
    // in milliseconds
    int thinkingTime = 50;
    int delay = 30;
    // player = new controllers.singlePlayer.sampleRandom.Agent(stateObs, timer);
    // check that we can play the game
    Random random = new Random();
    // this is how many steps we'll take in the actual game ...
    int nSteps = 10;
    ElapsedTimer t = new ElapsedTimer();
    for (int i = 0; i < nSteps && !stateObs.isGameOver(); i++) {
        timer = new ElapsedCpuTimer();
        timer.setMaxTimeMillis(thinkingTime);
        Types.ACTIONS action = player.act(stateObs.copy(), timer);
        // System.out.println("Selected: " + action); //  + "\t " + action.ordinal());
        stateObs.advance(action);
        noiseFree.advance(action);
    // System.out.println(stateObs.getGameScore());
    }
    System.out.println(stateObs.getGameScore());
    System.out.println(noiseFree.getGameScore());
    System.out.println(stateObs.isGameOver());
    System.out.println(t);
    return noiseFree.getGameScore();
}

Example 2 with StateObservation

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

the class SpaceBattleLinkTest method runTrial.

public static double runTrial(boolean runVisible) {
    // make an agent to test
    StateObservation stateObs = new SimpleMaxGame();
    // BattleGameSearchSpace.inject(BattleGameSearchSpace.getRandomPoint());
    // SampleEvolvedParams.solutions[1][2] = 5;
    // BattleGameSearchSpace.inject(SampleEvolvedParams.solutions[1]);
    // BattleGameSearchSpace.inject(SampleEvolvedParams.solutions[2]);
    BattleGameSearchSpace.inject(SampleEvolvedParams.solutions[1]);
    System.out.println("Params are:");
    System.out.println(BattleGameParameters.params);
    // can also overide parameters by setting them directly as follows:
    // BattleGameParameters.loss = 1.1;
    SpaceBattleLinkState linkState = new SpaceBattleLinkState();
    // set some parameters for the experiment
    GameActionSpaceAdapter.useHeuristic = false;
    Agent.useShiftBuffer = true;
    // DefaultMutator.totalRandomChaosMutation = false;
    // // supercl
    // StateObservation stateObs = linkState;
    ElapsedCpuTimer timer = new ElapsedCpuTimer();
    AbstractPlayer player;
    // controllers.singlePlayer.sampleOLMCTS.Agent olmcts =
    // new controllers.singlePlayer.sampleOLMCTS.Agent(linkState, timer);
    player = new controllers.singlePlayer.discountOLMCTS.Agent(linkState, timer);
    // try the evolutionary players
    int nResamples = 2;
    EvoAlg evoAlg = new SimpleRMHC(nResamples);
    double kExplore = 10;
    int nNeighbours = 100;
    int nEvals = 200;
    evoAlg = new NTupleBanditEA(kExplore, nNeighbours);
    // player = new controllers.singlePlayer.ea.Agent(linkState, timer, evoAlg, nEvals);
    controllers.singlePlayer.nestedMC.Agent nestedMC = new controllers.singlePlayer.nestedMC.Agent(linkState, timer);
    nestedMC.maxRolloutLength = 10;
    nestedMC.nestDepth = 2;
    player = nestedMC;
    // in milliseconds
    int thinkingTime = 50;
    int delay = 10;
    // player = new controllers.singlePlayer.sampleRandom.Agent(stateObs, timer);
    // check that we can play the game
    Random random = new Random();
    int nSteps = 500;
    ElapsedTimer t = new ElapsedTimer();
    BattleView view = new BattleView(linkState.state);
    // set view to null to run fast with no visuals
    if (!runVisible)
        view = null;
    if (view != null) {
        new JEasyFrame(view, "Simple Battle Game");
    }
    boolean verbose = false;
    for (int i = 0; i < nSteps && !linkState.isGameOver(); i++) {
        ArrayList<Types.ACTIONS> actions = linkState.getAvailableActions();
        timer = new ElapsedCpuTimer();
        timer.setMaxTimeMillis(thinkingTime);
        Types.ACTIONS action = player.act(linkState.copy(), timer);
        // action = actions.get(random.nextInt(actions.size()));
        if (verbose)
            // + "\t " + action.ordinal());
            System.out.println(i + "\t Selected: " + action);
        linkState.advance(action);
        if (view != null) {
            view.repaint();
            try {
                Thread.sleep(delay);
            } catch (Exception e) {
            }
        }
        if (verbose)
            System.out.println(linkState.getGameScore());
    }
    System.out.println("Game score: " + linkState.getGameScore());
    return linkState.getGameScore();
}

Example 3 with StateObservation

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

the class Agent method nested.

void nested(StateObservation stateObservation, int nestingLevel, Types.ACTIONS[] moveSeq, int nActionsPlayed) {
    int nbMoves = 0;
    // Types.ACTIONS[] moves = new Types.ACTIONS[maxLegalMoves];
    lengthBestRollout[nestingLevel] = -1;
    scoreBestRollout[nestingLevel] = Double.NEGATIVE_INFINITY;
    float res;
    while (true) {
        if (stateObservation.isGameOver())
            return;
        if (nActionsPlayed >= maxRolloutLength)
            return;
        // return board.score ();
        for (int i = 0; i < num_actions; i++) {
            StateObservation state = stateObservation.copy();
            Types.ACTIONS[] moveSeqCopy = new Types.ACTIONS[maxRolloutLength];
            int nActionsCopy = nActionsPlayed;
            for (int j = 0; j < nActionsPlayed; j++) moveSeqCopy[j] = moveSeq[j];
            if (nestingLevel == 1) {
                state.advance(actions[i]);
                moveSeqCopy[nActionsCopy] = actions[i];
                nActionsCopy++;
                playout(state, moveSeqCopy, nActionsCopy);
            } else {
                state.advance(actions[i]);
                moveSeqCopy[nActionsCopy] = actions[i];
                nActionsCopy++;
                nested(state, nestingLevel - 1, moveSeqCopy, nActionsCopy);
            }
            double score = state.getGameScore();
            // System.out.println("Move Seq Copy: " + Arrays.toString(moveSeqCopy));
            if (score > scoreBestRollout[nestingLevel]) {
                // System.out.println ("level " + nestingLevel + "score " + score);
                scoreBestRollout[nestingLevel] = score;
                lengthBestRollout[nestingLevel] = maxRolloutLength;
                for (int j = 0; j < maxRolloutLength; j++) bestRollout[nestingLevel][j] = moveSeqCopy[j];
            } else {
            // this was to find a bug
            // System.out.println("score comparison: " + score + " : " + scoreBestRollout[nestingLevel] );
            }
        }
        // System.out.println(nestingLevel + " : " + nActionsPlayed);
        // for (Types.ACTIONS[] a : bestRollout) {
        // System.out.println(Arrays.toString(a));
        // }
        stateObservation.advance(bestRollout[nestingLevel][nActionsPlayed]);
        moveSeq[nActionsPlayed] = bestRollout[nestingLevel][nActionsPlayed];
        nActionsPlayed++;
    }
}

Example 4 with StateObservation

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

the class GameActionSpaceAdapter method evaluate.

@Override
public double evaluate(int[] actions) {
    // take a copy of the current game state and accumulate the score as we go along
    StateObservation obs = stateObservation.copy();
    // note the score now - for normalisation reasons
    // we wish to track the change in score, not the absolute score
    double initScore = obs.getGameScore();
    double discount = 1.0;
    double denom = 0;
    double discountedTot = 0;
    // need to do the visual stuff here ...
    LinePlot linePlot = null;
    if (visual) {
        float grey = (nEvals % 100) / 100;
        linePlot = new LinePlot().setColor(new Color(grey, grey, grey));
    }
    for (int i = 0; i < sequenceLength; i++) {
        obs.advance(gvgaiActions[actions[i]]);
        discountedTot += discount * (obs.getGameScore() - initScore);
        if (useHeuristic && obs instanceof SpaceBattleLinkState) {
            SpaceBattleLinkState state = (SpaceBattleLinkState) obs;
            discountedTot += state.getHeuristicScore();
        }
        denom += discount;
        discount *= discountFactor;
        if (linePlot != null) {
            linePlot.add(discountedTot + Math.random() * 0);
        }
    }
    if (visual) {
        linePlots.add(linePlot);
    }
    nEvals++;
    double delta;
    if (useDiscountFactor) {
        delta = discountedTot / denom;
    } else {
        delta = obs.getGameScore() - initScore;
    }
    delta += noiseLevel * random.nextGaussian();
    logger.log(delta, actions, false);
    return delta;
}

Example 5 with StateObservation

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

the class SimpleGridTest method runOnce.

public static double runOnce() {
    // make an agent to test
    StateObservation gridGame = new GridModel();
    System.out.println(gridGame.getGameScore());
    System.out.println(gridGame.copy().getGameScore());
    // System.exit(0);
    ElapsedCpuTimer timer = new ElapsedCpuTimer();
    AbstractPlayer player;
    controllers.singlePlayer.sampleOLMCTS.Agent olmcts = new controllers.singlePlayer.sampleOLMCTS.Agent(gridGame, timer);
    controllers.singlePlayer.discountOLMCTS.Agent discountOlmcts = new controllers.singlePlayer.discountOLMCTS.Agent(gridGame, timer);
    controllers.singlePlayer.nestedMC.Agent nestedMC = new controllers.singlePlayer.nestedMC.Agent(gridGame, timer);
    player = olmcts;
    // player = discountOlmcts;
    // for the following we can pass the Evolutionary algorithm to use
    int nResamples = 2;
    EvoAlg evoAlg = new SimpleRMHC(nResamples);
    int nEvals = 2000;
    double kExplore = 10;
    int nNeighbours = 100;
    evoAlg = new NTupleBanditEA(kExplore, nNeighbours);
    evoAlg = new SlidingMeanEDA();
    // DefaultMutator.totalRandomChaosMutation = false;
    Agent.useShiftBuffer = true;
    Agent.SEQUENCE_LENGTH = 30;
    player = new Agent(gridGame, timer, evoAlg, nEvals);
    nestedMC.maxRolloutLength = 30;
    nestedMC.nestDepth = 3;
    // player = nestedMC;
    // in milliseconds
    int thinkingTime = 50;
    int delay = 30;
    // player = new controllers.singlePlayer.sampleRandom.Agent(stateObs, timer);
    // check that we can play the game
    Random random = new Random();
    // this is how many steps we'll take in the actual game ...
    int nSteps = 30;
    ElapsedTimer t = new ElapsedTimer();
    for (int i = 0; i < nSteps && !gridGame.isGameOver(); i++) {
        timer = new ElapsedCpuTimer();
        timer.setMaxTimeMillis(thinkingTime);
        Types.ACTIONS action = player.act(gridGame.copy(), timer);
        System.out.println();
        // + "\t " + action.ordinal());
        System.out.println("Selected: " + action);
        gridGame.advance(action);
        System.out.println("Game state: " + gridGame);
        System.out.println();
    }
    System.out.println(gridGame.getGameScore());
    return gridGame.getGameScore();
}