Examples with NTupleBanditEA ntuple.NTupleBanditEA used on opensource projects

Example 1 with NTupleBanditEA

use of ntuple.NTupleBanditEA 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 NTupleBanditEA

use of ntuple.NTupleBanditEA 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 NTupleBanditEA

use of ntuple.NTupleBanditEA in project SimpleAsteroids by ljialin.

the class HyperParamTuneRunner method runTrials.

public void runTrials(EvoAlg evoAlg, AnnotatedFitnessSpace annotatedFitnessSpace) {
    ElapsedTimer timer = new ElapsedTimer();
    StatSummary ss = new StatSummary("Overall results: " + evoAlg.getClass().getSimpleName());
    for (int i = 0; i < nTrials; i++) {
        System.out.println();
        System.out.println("Running trial: " + (i + 1));
        try {
            ss.add(runTrial(evoAlg, annotatedFitnessSpace));
            System.out.println("Stats so far");
            System.out.println(ss);
            if (verbose) {
                plotFitnessEvolution(annotatedFitnessSpace.logger(), annotatedFitnessSpace, plotChecks);
                // annotatedFitnessSpace.logger()
                // ((NTupleSystem) ((NTupleBanditEA) evoAlg).banditLandscapeModel).printDetailedReport(new EvoAgentSearchSpaceAsteroids().getParams());
                NTupleSystem nTupleSystem = ((NTupleSystem) ((NTupleBanditEA) evoAlg).banditLandscapeModel);
                nTupleSystem.printDetailedReport(annotatedFitnessSpace.getParams());
            // new Plotter().setModel(nTupleSystem).defaultPlot().plot1Tuples();
            }
        } catch (Exception e) {
            e.printStackTrace();
        }
    }
    if (verbose) {
        lineChart.addLineGroup(sampleEvolution);
        if (plotChecks > 0)
            lineChart.addLineGroup(bestGuess);
        new JEasyFrame(lineChart, "Sample Evolution");
    }
    System.out.println("nEvals per run: " + nEvals);
    System.out.println(ss);
    System.out.println("Total time for experiment: " + timer);
}

Example 4 with NTupleBanditEA

use of ntuple.NTupleBanditEA in project SimpleAsteroids by ljialin.

the class TestHyperParamAsteroids method main.

public static void main(String[] args) {
    AnnotatedFitnessSpace testAsteroids = new EvoAgentSearchSpaceAsteroids();
    EvoAlg[] evoAlgs = { new NTupleBanditEA().setKExplore(10000), // new CompactSlidingGA(),
    new SlidingMeanEDA() };
    int nChecks = 50;
    int nEvals = 50;
    int nTrials = 2;
    for (EvoAlg evoAlg : evoAlgs) {
        HyperParamTuneRunner runner = new HyperParamTuneRunner();
        runner.nChecks = nChecks;
        runner.nTrials = nTrials;
        runner.nEvals = nEvals;
        runner.runTrials(evoAlg, testAsteroids);
    }
}

Example 5 with NTupleBanditEA

use of ntuple.NTupleBanditEA in project SimpleAsteroids by ljialin.

the class TestHyperParamPlanetWars method main.

public static void main(String[] args) {
    int nEvals = 288;
    if (args.length == 1) {
        nEvals = Integer.parseInt(args[0]);
    }
    System.out.println("Optimization budget: " + nEvals);
    NTupleBanditEA ntbea = new NTupleBanditEA().setKExplore(1);
    GameState.includeBuffersInScore = true;
    EvoAgentSearchSpace.tickBudget = 2000;
    EvoAlg[] evoAlgs = { // new SimpleRMHC(5),
    ntbea };
    int nChecks = 100;
    int nTrials = 100;
    ElapsedTimer timer = new ElapsedTimer();
    for (EvoAlg evoAlg : evoAlgs) {
        // LineChart lineChart = new LineChart();
        // lineChart.yAxis = new LineChartAxis(new double[]{-2, -1, 0, 1, 2});
        // lineChart.setYLabel("Fitness");
        HyperParamTuneRunner runner = new HyperParamTuneRunner();
        // runner.verbose = true;
        // runner.setLineChart(lineChart);
        runner.nChecks = nChecks;
        runner.nTrials = nTrials;
        runner.nEvals = nEvals;
        runner.plotChecks = 0;
        AnnotatedFitnessSpace testPlanetWars = new EvoAgentSearchSpace();
        System.out.println("Testing: " + evoAlg);
        runner.runTrials(evoAlg, testPlanetWars);
        System.out.println("Finished testing: " + evoAlg);
    // note, this is a bit of a hack: it only reports the final solution
    // System.out.println(new EvoAgentSearchSpace().report(runner.solution));
    }
    // System.out.println(ntbea.getModel().s);
    System.out.println("Time for all experiments: " + timer);
}