Examples with StatisticalMoments de.lmu.ifi.dbs.elki.math.StatisticalMoments used on opensource projects

Example 1 with StatisticalMoments

use of de.lmu.ifi.dbs.elki.math.StatisticalMoments in project elki by elki-project.

the class BestFitEstimator method estimate.

@Override
public <A> Distribution estimate(A data, NumberArrayAdapter<?, A> adapter) {
    int numlmm = 0;
    for (LMMDistributionEstimator<?> est : lmmests) {
        numlmm = Math.max(numlmm, est.getNumMoments());
    }
    final int len = adapter.size(data);
    // Build various statistics:
    StatisticalMoments mom = new StatisticalMoments(), logmom = new StatisticalMoments();
    double[] x = new double[len], scratch = new double[len], logx = new double[len];
    if (LOG.isDebuggingFine()) {
        LOG.debugFine("Computing statistical moments and L-Moments.");
    }
    for (int i = 0; i < len; i++) {
        final double val = x[i] = adapter.getDouble(data, i);
        if (Double.NEGATIVE_INFINITY < val && val < Double.POSITIVE_INFINITY) {
            mom.put(val);
        }
    }
    if (mom.getMax() <= mom.getMin()) {
        LOG.warning("Constant distribution detected. Cannot fit.");
        return new UniformDistribution(mom.getMin() - 1., mom.getMax() + 1.);
    }
    // Sort: for L-Moments, but getting the median is now also cheap.
    Arrays.sort(x);
    double[] lmm;
    try {
        lmm = (numlmm > 0) ? ProbabilityWeightedMoments.samLMR(x, DoubleArrayAdapter.STATIC, numlmm) : null;
    } catch (ArithmeticException e) {
        lmm = null;
    }
    final double min = x[0], median = .5 * (x[len >> 1] + x[(len + 1) >> 1]), max = x[len - 1];
    if (LOG.isDebuggingFine()) {
        LOG.debugFine("Computing statistical moments in logspace.");
    }
    // Build logspace copy:
    double shift = Math.min(0., min - (max - min) * 1e-10);
    for (int i = 0; i < len; i++) {
        double val = x[i] - shift;
        val = val > 0. ? FastMath.log(val) : Double.NEGATIVE_INFINITY;
        logx[i] = val;
        if (Double.NEGATIVE_INFINITY < val && val < Double.POSITIVE_INFINITY) {
            logmom.put(val);
        }
    }
    double logmedian = .5 * (logx[len >> 1] + logx[(len + 1) >> 1]);
    if (LOG.isDebuggingFine()) {
        LOG.debugFine("Computing MADs.");
    }
    double mad = MADDistributionEstimator.computeMAD(x, len, median, scratch);
    double logmad = MADDistributionEstimator.computeMAD(logx, len, logmedian, scratch);
    BestFit best = new BestFit(x, scratch);
    for (MOMDistributionEstimator<?> est : momests) {
        try {
            best.test(est, est.estimateFromStatisticalMoments(mom));
        } catch (ArithmeticException e) {
            warnIfDebugging(e, est);
        }
    }
    for (MADDistributionEstimator<?> est : madests) {
        try {
            best.test(est, est.estimateFromMedianMAD(median, mad));
        } catch (ArithmeticException e) {
            warnIfDebugging(e, est);
        }
    }
    for (LMMDistributionEstimator<?> est : lmmests) {
        if (lmm != null) {
            try {
                best.test(est, est.estimateFromLMoments(lmm));
            } catch (ArithmeticException e) {
                warnIfDebugging(e, est);
            }
        }
    }
    for (LogMOMDistributionEstimator<?> est : logmomests) {
        try {
            best.test(est, est.estimateFromLogStatisticalMoments(logmom, shift));
        } catch (ArithmeticException e) {
            warnIfDebugging(e, est);
        }
    }
    for (LogMADDistributionEstimator<?> est : logmadests) {
        try {
            best.test(est, est.estimateFromLogMedianMAD(logmedian, logmad, shift));
        } catch (ArithmeticException e) {
            warnIfDebugging(e, est);
        }
    }
    {
        // Uniform estimators.
        final UniformMinMaxEstimator est = UniformMinMaxEstimator.STATIC;
        best.test(est, est.estimate(min, max));
    }
    {
        // Uniform estimators.
        final UniformEnhancedMinMaxEstimator est = UniformEnhancedMinMaxEstimator.STATIC;
        best.test(est, est.estimate(min, max, len));
    }
    if (LOG.isVeryVerbose()) {
        LOG.veryverbose("Best distribution fit: " + best.score + " " + best.toString() + " via " + best.est);
    }
    return best.dist;
}

Example 2 with StatisticalMoments

use of de.lmu.ifi.dbs.elki.math.StatisticalMoments in project elki by elki-project.

the class MOMDistributionEstimator method estimate.

@Override
default <A> D estimate(A data, NumberArrayAdapter<?, A> adapter) {
    final int size = adapter.size(data);
    StatisticalMoments mv = new StatisticalMoments();
    for (int i = 0; i < size; i++) {
        final double val = adapter.getDouble(data, i);
        if (Double.isInfinite(val) || Double.isNaN(val)) {
            continue;
        }
        mv.put(val);
    }
    return estimateFromStatisticalMoments(mv);
}

Example 3 with StatisticalMoments

use of de.lmu.ifi.dbs.elki.math.StatisticalMoments in project elki by elki-project.

the class LogMOMDistributionEstimator method estimate.

@Override
default <A> D estimate(A data, NumberArrayAdapter<?, A> adapter) {
    final int len = adapter.size(data);
    double min = min(data, adapter, 0., 1e-10);
    StatisticalMoments mv = new StatisticalMoments();
    for (int i = 0; i < len; i++) {
        final double val = adapter.getDouble(data, i) - min;
        if (Double.isInfinite(val) || Double.isNaN(val) || val <= 0.) {
            continue;
        }
        mv.put(FastMath.log(val));
    }
    return estimateFromLogStatisticalMoments(mv, min);
}