Examples with CameraSensorView nars.video.CameraSensorView used on opensource projects

Example 1 with CameraSensorView

use of nars.video.CameraSensorView in project narchy by automenta.

the class TrackXY method main.

public static void main(String[] args) {
    boolean nars = true;
    boolean rl = false;
    int dur = 1;
    NARS nb = new NARS().exe(new UniExec(64)).time(new CycleTime().dur(dur)).index(// new HijackConceptIndex(4 * 1024, 4)
    new CaffeineIndex(32 * 1024));
    NAR n = nb.get();
    n.termVolumeMax.set(20);
    // n.priDefault(BELIEF, 0.2f);
    // n.priDefault(GOAL, 0.5f);
    n.activationRate.set(0.2f);
    // n.forgetRate.set(0.9f);
    TrackXY t = new TrackXY(4, 4);
    n.on(t);
    int experimentTime = 8048;
    n.synch();
    if (rl) {
        new RLBooster(t, // HaiQ::new,
        HaiQae::new, // RandomAgent::new,
        1);
        t.curiosity.set(0);
    }
    if (nars) {
        // Param.DEBUG = true;
        // n.log();
        // for (String action : new String[]{"up", "down", "left", "right"}) {
        // //n.goal($.the(action), Tense.Present, 0f, 0.1f);
        // n.goal($.the(action), Tense.Present, 1f, 0.1f);
        // }
        Deriver d = new Deriver(Derivers.rules(// 1,
        1, 8, n, // "list.nal",
        "motivation.nal"), n);
        d.conceptsPerIteration.set(32);
        n.timeFocus.set(2);
        ConjClustering cjB = new ConjClustering(n, BELIEF, // (tt)->true,
        (tt) -> tt.isInput(), 4, 16);
        // ConjClustering cjG = new ConjClustering(n, GOAL,
        // (tt)->true,
        // //(tt) -> tt.isInput(),
        // 5, 16);
        // Implier ii = new Implier(t , 0, 1);
        // ArithmeticIntroduction ai = new ArithmeticIntroduction(4, n);
        window(new Gridding(new AutoSurface(d), new AutoSurface(cjB)), 400, 300);
        n.onTask(tt -> {
            if (tt instanceof DerivedTask && tt.isGoal()) {
                System.out.println(tt.proof());
            }
        });
    }
    // n.log();
    // n.startFPS(fps);
    // t.runFPS(fps);
    n.onCycle(t);
    final double[] rewardSum = { 0 };
    n.onCycle(() -> {
        rewardSum[0] += t.reward;
    });
    n.runLater(() -> {
        window(Vis.top(n), 800, 250);
        NAgentX.chart(t);
        window(new CameraSensorView(t.cam, n) {

            @Override
            protected void paint(GL2 gl, int dtMS) {
                super.paint(gl, dtMS);
                RectFloat2D at = cellRect(t.sx, t.sy, 0.5f, 0.5f);
                gl.glColor4f(1, 0, 0, 0.9f);
                Draw.rect(gl, at.move(x(), y(), 0.01f));
            }
        }.withControls(), 800, 800);
    });
    n.run(experimentTime);
// n.startFPS(10f);
// t.runFPS(10f);
// System.out.println(
// 
// n4(rewardSum[0] / n.time()) + " avg reward");
// System.exit(0);
}

Example 2 with CameraSensorView

use of nars.video.CameraSensorView in project narchy by automenta.

the class Recog2D method conceptTraining.

Surface conceptTraining(BeliefVector tv, NAR nar) {
    // LinkedHashMap<TaskConcept, BeliefVector.Neuron> out = tv.out;
    Plot2D p;
    int history = 256;
    Gridding g = new Gridding(p = new Plot2D(history, Plot2D.Line).add("Reward", () -> reward), new AspectAlign(new CameraSensorView(sp, this), AspectAlign.Align.Center, sp.width, sp.height), new Gridding(beliefTableCharts(nar, List.of(tv.concepts), 16)), new Gridding(IntStream.range(0, tv.concepts.length).mapToObj(i -> new spacegraph.space2d.widget.text.Label(String.valueOf(i)) {

        @Override
        protected void paintBelow(GL2 gl) {
            Concept c = tv.concepts[i];
            BeliefVector.Neuron nn = tv.neurons[i];
            float freq, conf;
            Truth t = nar.beliefTruth(c, nar.time());
            if (t != null) {
                conf = t.conf();
                freq = t.freq();
            } else {
                conf = nar.confMin.floatValue();
                float defaultFreq = // interpret no-belief as maybe
                0.5f;
                // Float.NaN  //use NaN to force learning of negation as separate from no-belief
                freq = defaultFreq;
            }
            Draw.colorBipolar(gl, 2f * (freq - 0.5f));
            float m = 0.5f * conf;
            Draw.rect(gl, bounds);
            if (tv.verify) {
                float error = nn.error;
                if (error != error) {
                // training phase
                // Draw.rect(gl, m / 2, m / 2, 1 - m, 1 - m);
                } else {
                // verification
                // draw backgroudn/border
                // gl.glColor3f(error, 1f - error, 0f);
                // 
                // float fontSize = 0.08f;
                // gl.glColor3f(1f, 1f, 1f);
                // Draw.text(gl, c.term().toString(), fontSize, m / 2, 1f - m / 2, 0);
                // Draw.text(gl, "err=" + n2(error), fontSize, m / 2, m / 2, 0);
                }
            }
        }
    }).toArray(Surface[]::new)));
    final int[] frames = { 0 };
    onFrame(() -> {
        if (frames[0]++ % imagePeriod == 0) {
            nextImage();
        }
        redraw();
        // if (neural.get()) {
        // if (nar.time() < trainFrames) {
        outs.expect(image);
        if (neural.get()) {
            train.update(mlpLearn, mlpSupport);
        }
        p.update();
    // s.update();
    });
    return g;
}