Examples with DescriptiveStatistics org.apache.commons.math3.stat.descriptive.DescriptiveStatistics used on opensource projects

Example 41 with DescriptiveStatistics

use of org.apache.commons.math3.stat.descriptive.DescriptiveStatistics in project pinot by linkedin.

the class ForwardIndexReaderBenchmark method singleValuedReadBenchMarkV2.

public static void singleValuedReadBenchMarkV2(File file, int numDocs, int numBits) throws Exception {
    boolean signed = false;
    boolean isMmap = false;
    long start, end;
    boolean fullScan = true;
    boolean batchRead = true;
    boolean singleRead = true;
    PinotDataBuffer heapBuffer = PinotDataBuffer.fromFile(file, ReadMode.heap, FileChannel.MapMode.READ_ONLY, "benchmarking");
    com.linkedin.pinot.core.io.reader.impl.v2.FixedBitSingleValueReader reader = new com.linkedin.pinot.core.io.reader.impl.v2.FixedBitSingleValueReader(heapBuffer, numDocs, numBits, signed);
    if (fullScan) {
        DescriptiveStatistics stats = new DescriptiveStatistics();
        ByteBuffer buffer = ByteBuffer.allocateDirect((int) file.length());
        RandomAccessFile raf = new RandomAccessFile(file, "r");
        raf.getChannel().read(buffer);
        raf.close();
        int[] input = new int[numBits];
        int[] output = new int[32];
        int numBatches = (numDocs + 31) / 32;
        for (int run = 0; run < MAX_RUNS; run++) {
            start = System.currentTimeMillis();
            for (int i = 0; i < numBatches; i++) {
                for (int j = 0; j < numBits; j++) {
                    input[j] = buffer.getInt(i * numBits * 4 + j * 4);
                }
                BitPacking.fastunpack(input, 0, output, 0, numBits);
            }
            end = System.currentTimeMillis();
            stats.addValue((end - start));
        }
        System.out.println(" v2 full scan stats for " + file.getName());
        System.out.println(stats.toString().replaceAll("\n", ", ") + " raw:" + Arrays.toString(stats.getValues()));
    }
    if (singleRead) {
        DescriptiveStatistics stats = new DescriptiveStatistics();
        // sequential read
        for (int run = 0; run < MAX_RUNS; run++) {
            start = System.currentTimeMillis();
            for (int i = 0; i < numDocs; i++) {
                int value = reader.getInt(i);
            }
            end = System.currentTimeMillis();
            stats.addValue((end - start));
        }
        System.out.println(" v2 sequential single read for " + file.getName());
        System.out.println(stats.toString().replaceAll("\n", ", ") + " raw:" + Arrays.toString(stats.getValues()));
    }
    if (batchRead) {
        DescriptiveStatistics stats = new DescriptiveStatistics();
        int batchSize = Math.min(5000, numDocs);
        int[] output = new int[batchSize];
        int[] rowIds = new int[batchSize];
        // sequential read
        for (int run = 0; run < MAX_RUNS; run++) {
            start = System.currentTimeMillis();
            int rowId = 0;
            while (rowId < numDocs) {
                int length = Math.min(batchSize, numDocs - rowId);
                for (int i = 0; i < length; i++) {
                    rowIds[i] = rowId + i;
                }
                reader.getIntBatch(rowIds, output, length);
                rowId = rowId + length;
            }
            end = System.currentTimeMillis();
            stats.addValue((end - start));
        }
        System.out.println("v2 sequential batch read stats for " + file.getName());
        System.out.println(stats.toString().replaceAll("\n", ", ") + " raw:" + Arrays.toString(stats.getValues()));
    }
    reader.close();
}

Example 42 with DescriptiveStatistics

use of org.apache.commons.math3.stat.descriptive.DescriptiveStatistics in project GDSC-SMLM by aherbert.

the class PCPALMMolecules method calculateAveragePrecision.

/**
	 * Calculate the average precision by fitting a skewed Gaussian to the histogram of the precision distribution.
	 * <p>
	 * A simple mean and SD of the histogram is computed. If the mean of the Skewed Gaussian does not fit within 3 SDs
	 * of the simple mean then the simple mean is returned.
	 * 
	 * @param molecules
	 * @param title
	 *            the plot title (null if no plot should be displayed)
	 * @param histogramBins
	 * @param logFitParameters
	 *            Record the fit parameters to the ImageJ log
	 * @param removeOutliers
	 *            The distribution is created using all values within 1.5x the inter-quartile range (IQR) of the data
	 * @return The average precision
	 */
public double calculateAveragePrecision(ArrayList<Molecule> molecules, String title, int histogramBins, boolean logFitParameters, boolean removeOutliers) {
    // Plot histogram of the precision
    float[] data = new float[molecules.size()];
    DescriptiveStatistics stats = new DescriptiveStatistics();
    double yMin = Double.NEGATIVE_INFINITY, yMax = 0;
    for (int i = 0; i < data.length; i++) {
        data[i] = (float) molecules.get(i).precision;
        stats.addValue(data[i]);
    }
    // Set the min and max y-values using 1.5 x IQR 
    if (removeOutliers) {
        double lower = stats.getPercentile(25);
        double upper = stats.getPercentile(75);
        if (Double.isNaN(lower) || Double.isNaN(upper)) {
            if (logFitParameters)
                Utils.log("Error computing IQR: %f - %f", lower, upper);
        } else {
            double iqr = upper - lower;
            yMin = FastMath.max(lower - iqr, stats.getMin());
            yMax = FastMath.min(upper + iqr, stats.getMax());
            if (logFitParameters)
                Utils.log("  Data range: %f - %f. Plotting 1.5x IQR: %f - %f", stats.getMin(), stats.getMax(), yMin, yMax);
        }
    }
    if (yMin == Double.NEGATIVE_INFINITY) {
        yMin = stats.getMin();
        yMax = stats.getMax();
        if (logFitParameters)
            Utils.log("  Data range: %f - %f", yMin, yMax);
    }
    float[][] hist = Utils.calcHistogram(data, yMin, yMax, histogramBins);
    Plot2 plot = null;
    if (title != null) {
        plot = new Plot2(title, "Precision", "Frequency");
        float[] xValues = hist[0];
        float[] yValues = hist[1];
        if (xValues.length > 0) {
            double xPadding = 0.05 * (xValues[xValues.length - 1] - xValues[0]);
            plot.setLimits(xValues[0] - xPadding, xValues[xValues.length - 1] + xPadding, 0, Maths.max(yValues) * 1.05);
        }
        plot.addPoints(xValues, yValues, Plot2.BAR);
        Utils.display(title, plot);
    }
    // Extract non-zero data
    float[] x = Arrays.copyOf(hist[0], hist[0].length);
    float[] y = hist[1];
    int count = 0;
    float dx = (x[1] - x[0]) * 0.5f;
    for (int i = 0; i < y.length; i++) if (y[i] > 0) {
        x[count] = x[i] + dx;
        y[count] = y[i];
        count++;
    }
    x = Arrays.copyOf(x, count);
    y = Arrays.copyOf(y, count);
    // Sense check to fitted data. Get mean and SD of histogram
    double[] stats2 = Utils.getHistogramStatistics(x, y);
    double mean = stats2[0];
    if (logFitParameters)
        log("  Initial Statistics: %f +/- %f", stats2[0], stats2[1]);
    // Standard Gaussian fit
    double[] parameters = fitGaussian(x, y);
    if (parameters == null) {
        log("  Failed to fit initial Gaussian");
        return mean;
    }
    double newMean = parameters[1];
    double error = Math.abs(stats2[0] - newMean) / stats2[1];
    if (error > 3) {
        log("  Failed to fit Gaussian: %f standard deviations from histogram mean", error);
        return mean;
    }
    if (newMean < yMin || newMean > yMax) {
        log("  Failed to fit Gaussian: %f outside data range %f - %f", newMean, yMin, yMax);
        return mean;
    }
    mean = newMean;
    if (logFitParameters)
        log("  Initial Gaussian: %f @ %f +/- %f", parameters[0], parameters[1], parameters[2]);
    double[] initialSolution = new double[] { parameters[0], parameters[1], parameters[2], -1 };
    // Fit to a skewed Gaussian (or appropriate function)
    double[] skewParameters = fitSkewGaussian(x, y, initialSolution);
    if (skewParameters == null) {
        log("  Failed to fit Skewed Gaussian");
        return mean;
    }
    SkewNormalFunction sn = new SkewNormalFunction(skewParameters);
    if (logFitParameters)
        log("  Skewed Gaussian: %f @ %f +/- %f (a = %f) => %f +/- %f", skewParameters[0], skewParameters[1], skewParameters[2], skewParameters[3], sn.getMean(), Math.sqrt(sn.getVariance()));
    newMean = sn.getMean();
    error = Math.abs(stats2[0] - newMean) / stats2[1];
    if (error > 3) {
        log("  Failed to fit Skewed Gaussian: %f standard deviations from histogram mean", error);
        return mean;
    }
    if (newMean < yMin || newMean > yMax) {
        log("  Failed to fit Skewed Gaussian: %f outside data range %f - %f", newMean, yMin, yMax);
        return mean;
    }
    // Use original histogram x-axis to maintain all the bins
    if (plot != null) {
        x = hist[0];
        for (int i = 0; i < y.length; i++) x[i] += dx;
        plot.setColor(Color.red);
        addToPlot(plot, x, skewParameters, Plot2.LINE);
        plot.setColor(Color.black);
        Utils.display(title, plot);
    }
    // Return the average precision from the fitted curve
    return newMean;
}

Example 43 with DescriptiveStatistics

use of org.apache.commons.math3.stat.descriptive.DescriptiveStatistics in project GDSC-SMLM by aherbert.

the class SummariseResults method addSummary.

private void addSummary(StringBuilder sb, MemoryPeakResults result) {
    DescriptiveStatistics[] stats = new DescriptiveStatistics[2];
    char[] suffix = new char[2];
    for (int i = 0; i < stats.length; i++) {
        stats[i] = new DescriptiveStatistics();
        suffix[i] = 0;
    }
    if (result.hasNullResults()) {
        IJ.log("Null results in dataset: " + result.getName());
        if (removeNullResults == UNKNOWN) {
            GenericDialog gd = new GenericDialog(TITLE);
            gd.addMessage("There are invalid results in memory.\n \nClean these results?");
            gd.enableYesNoCancel();
            gd.hideCancelButton();
            gd.showDialog();
            removeNullResults = (gd.wasOKed()) ? YES : NO;
        }
        if (removeNullResults == NO)
            result = result.copy();
        result.removeNullResults();
    }
    final int size = result.size();
    if (size > 0) {
        // Check if we can use the stored precision
        if (result.hasStoredPrecision()) {
            suffix[0] = '*';
            for (PeakResult peakResult : result.getResults()) {
                stats[0].addValue(peakResult.getPrecision());
            }
        } else if (result.isCalibratedForPrecision()) {
            final double nmPerPixel = result.getNmPerPixel();
            final double gain = result.getGain();
            final boolean emCCD = result.isEMCCD();
            for (PeakResult peakResult : result.getResults()) {
                stats[0].addValue(peakResult.getPrecision(nmPerPixel, gain, emCCD));
            }
        }
        // SNR requires noise
        if (result.getHead().noise > 0) {
            for (PeakResult peakResult : result.getResults()) {
                stats[1].addValue(peakResult.getSignal() / peakResult.noise);
            }
        }
    }
    sb.append(result.getName());
    sb.append("\t").append(result.size());
    int maxT = getMaxT(result);
    sb.append("\t").append(maxT);
    final double exposureTime = (result.getCalibration() != null) ? result.getCalibration().getExposureTime() : 0;
    sb.append("\t").append(Utils.timeToString(maxT * exposureTime));
    if (size > 0) {
        boolean includeDeviations = result.getHead().paramsStdDev != null;
        long memorySize = MemoryPeakResults.estimateMemorySize(size, includeDeviations);
        String memory = MemoryPeakResults.memorySizeString(memorySize);
        sb.append("\t").append(memory);
    } else {
        sb.append("\t-");
    }
    Rectangle bounds = result.getBounds(true);
    sb.append(String.format("\t%d,%d,%d,%d\t%s\t%s\t%s", bounds.x, bounds.y, bounds.x + bounds.width, bounds.y + bounds.height, Utils.rounded(result.getNmPerPixel(), 4), Utils.rounded(result.getGain(), 4), Utils.rounded(exposureTime, 4)));
    for (int i = 0; i < stats.length; i++) {
        if (Double.isNaN(stats[i].getMean())) {
            sb.append("\t-\t-\t-\t-");
        } else {
            sb.append("\t").append(IJ.d2s(stats[i].getMean(), 3));
            if (suffix[i] != 0)
                sb.append(suffix[i]);
            sb.append("\t").append(IJ.d2s(stats[i].getPercentile(50), 3));
            sb.append("\t").append(IJ.d2s(stats[i].getMin(), 3));
            sb.append("\t").append(IJ.d2s(stats[i].getMax(), 3));
        }
    }
    sb.append("\n");
}

Example 44 with DescriptiveStatistics

use of org.apache.commons.math3.stat.descriptive.DescriptiveStatistics in project GDSC-SMLM by aherbert.

the class FIRE method calculatePrecisionHistogram.

/**
	 * Calculate a histogram of the precision. The precision can be either stored in the results or calculated using the
	 * Mortensen formula. If the precision method for Q estimation is not fixed then the histogram is fitted with a
	 * Gaussian to create an initial estimate.
	 *
	 * @param precisionMethod
	 *            the precision method
	 * @return The precision histogram
	 */
private PrecisionHistogram calculatePrecisionHistogram() {
    boolean logFitParameters = false;
    String title = results.getName() + " Precision Histogram";
    // Check if the results has the precision already or if it can be computed.
    boolean canUseStored = canUseStoredPrecision(results);
    boolean canCalculatePrecision = canCalculatePrecision(results);
    // Set the method to compute a histogram. Default to the user selected option.
    PrecisionMethod m = null;
    if (canUseStored && precisionMethod == PrecisionMethod.STORED)
        m = precisionMethod;
    else if (canCalculatePrecision && precisionMethod == PrecisionMethod.CALCULATE)
        m = precisionMethod;
    if (m == null) {
        // We only have two choices so if one is available then select it.
        if (canUseStored)
            m = PrecisionMethod.STORED;
        else if (canCalculatePrecision)
            m = PrecisionMethod.CALCULATE;
        // If the user selected a method not available then log a warning
        if (m != null && precisionMethod != PrecisionMethod.FIXED) {
            IJ.log(String.format("%s : Selected precision method '%s' not available, switching to '%s'", TITLE, precisionMethod, m.getName()));
        }
        if (m == null) {
            // This does not matter if the user has provide a fixed input.
            if (precisionMethod == PrecisionMethod.FIXED) {
                PrecisionHistogram histogram = new PrecisionHistogram(title);
                histogram.mean = mean;
                histogram.sigma = sigma;
                return histogram;
            }
            // No precision
            return null;
        }
    }
    // We get here if we can compute precision.
    // Build the histogram 
    StoredDataStatistics precision = new StoredDataStatistics(results.size());
    if (m == PrecisionMethod.STORED) {
        for (PeakResult r : results.getResults()) {
            precision.add(r.getPrecision());
        }
    } else {
        final double nmPerPixel = results.getNmPerPixel();
        final double gain = results.getGain();
        final boolean emCCD = results.isEMCCD();
        for (PeakResult r : results.getResults()) {
            precision.add(r.getPrecision(nmPerPixel, gain, emCCD));
        }
    }
    //System.out.printf("Raw p = %f\n", precision.getMean());
    double yMin = Double.NEGATIVE_INFINITY, yMax = 0;
    // Set the min and max y-values using 1.5 x IQR 
    DescriptiveStatistics stats = precision.getStatistics();
    double lower = stats.getPercentile(25);
    double upper = stats.getPercentile(75);
    if (Double.isNaN(lower) || Double.isNaN(upper)) {
        if (logFitParameters)
            Utils.log("Error computing IQR: %f - %f", lower, upper);
    } else {
        double iqr = upper - lower;
        yMin = FastMath.max(lower - iqr, stats.getMin());
        yMax = FastMath.min(upper + iqr, stats.getMax());
        if (logFitParameters)
            Utils.log("  Data range: %f - %f. Plotting 1.5x IQR: %f - %f", stats.getMin(), stats.getMax(), yMin, yMax);
    }
    if (yMin == Double.NEGATIVE_INFINITY) {
        int n = 5;
        yMin = Math.max(stats.getMin(), stats.getMean() - n * stats.getStandardDeviation());
        yMax = Math.min(stats.getMax(), stats.getMean() + n * stats.getStandardDeviation());
        if (logFitParameters)
            Utils.log("  Data range: %f - %f. Plotting mean +/- %dxSD: %f - %f", stats.getMin(), stats.getMax(), n, yMin, yMax);
    }
    // Get the data within the range
    double[] data = precision.getValues();
    precision = new StoredDataStatistics(data.length);
    for (double d : data) {
        if (d < yMin || d > yMax)
            continue;
        precision.add(d);
    }
    int histogramBins = Utils.getBins(precision, Utils.BinMethod.SCOTT);
    float[][] hist = Utils.calcHistogram(precision.getFloatValues(), yMin, yMax, histogramBins);
    PrecisionHistogram histogram = new PrecisionHistogram(hist, precision, title);
    if (precisionMethod == PrecisionMethod.FIXED) {
        histogram.mean = mean;
        histogram.sigma = sigma;
        return histogram;
    }
    // Fitting of the histogram to produce the initial estimate
    // Extract non-zero data
    float[] x = Arrays.copyOf(hist[0], hist[0].length);
    float[] y = hist[1];
    int count = 0;
    float dx = (x[1] - x[0]) * 0.5f;
    for (int i = 0; i < y.length; i++) if (y[i] > 0) {
        x[count] = x[i] + dx;
        y[count] = y[i];
        count++;
    }
    x = Arrays.copyOf(x, count);
    y = Arrays.copyOf(y, count);
    // Sense check to fitted data. Get mean and SD of histogram
    double[] stats2 = Utils.getHistogramStatistics(x, y);
    if (logFitParameters)
        Utils.log("  Initial Statistics: %f +/- %f", stats2[0], stats2[1]);
    histogram.mean = stats2[0];
    histogram.sigma = stats2[1];
    // Standard Gaussian fit
    double[] parameters = fitGaussian(x, y);
    if (parameters == null) {
        Utils.log("  Failed to fit initial Gaussian");
        return histogram;
    }
    double newMean = parameters[1];
    double error = Math.abs(stats2[0] - newMean) / stats2[1];
    if (error > 3) {
        Utils.log("  Failed to fit Gaussian: %f standard deviations from histogram mean", error);
        return histogram;
    }
    if (newMean < yMin || newMean > yMax) {
        Utils.log("  Failed to fit Gaussian: %f outside data range %f - %f", newMean, yMin, yMax);
        return histogram;
    }
    if (logFitParameters)
        Utils.log("  Initial Gaussian: %f @ %f +/- %f", parameters[0], parameters[1], parameters[2]);
    histogram.mean = parameters[1];
    histogram.sigma = parameters[2];
    return histogram;
}

Example 45 with DescriptiveStatistics

use of org.apache.commons.math3.stat.descriptive.DescriptiveStatistics in project GDSC-SMLM by aherbert.

the class BenchmarkSpotFit method showDoubleHistogram.

private double[] showDoubleHistogram(StoredDataStatistics[][] stats, final int i, WindowOrganiser wo, double[][] matchScores, double nPredicted) {
    String xLabel = filterCriteria[i].name;
    LowerLimit lower = filterCriteria[i].lower;
    UpperLimit upper = filterCriteria[i].upper;
    double[] j = null;
    double[] metric = null;
    double maxJ = 0;
    if (i <= FILTER_PRECISION && (showFilterScoreHistograms || upper.requiresJaccard || lower.requiresJaccard)) {
        // Jaccard score verses the range of the metric
        Arrays.sort(matchScores, new Comparator<double[]>() {

            public int compare(double[] o1, double[] o2) {
                if (o1[i] < o2[i])
                    return -1;
                if (o1[i] > o2[i])
                    return 1;
                return 0;
            }
        });
        final int scoreIndex = FILTER_PRECISION + 1;
        int n = results.size();
        double tp = 0;
        double fp = 0;
        j = new double[matchScores.length + 1];
        metric = new double[j.length];
        for (int k = 0; k < matchScores.length; k++) {
            final double score = matchScores[k][scoreIndex];
            tp += score;
            fp += (1 - score);
            j[k + 1] = tp / (fp + n);
            metric[k + 1] = matchScores[k][i];
        }
        metric[0] = metric[1];
        maxJ = Maths.max(j);
        if (showFilterScoreHistograms) {
            String title = TITLE + " Jaccard " + xLabel;
            Plot plot = new Plot(title, xLabel, "Jaccard", metric, j);
            // Remove outliers
            double[] limitsx = Maths.limits(metric);
            Percentile p = new Percentile();
            double l = p.evaluate(metric, 25);
            double u = p.evaluate(metric, 75);
            double iqr = 1.5 * (u - l);
            limitsx[1] = Math.min(limitsx[1], u + iqr);
            plot.setLimits(limitsx[0], limitsx[1], 0, Maths.max(j));
            PlotWindow pw = Utils.display(title, plot);
            if (Utils.isNewWindow())
                wo.add(pw);
        }
    }
    // [0] is all
    // [1] is matches
    // [2] is no match
    StoredDataStatistics s1 = stats[0][i];
    StoredDataStatistics s2 = stats[1][i];
    StoredDataStatistics s3 = stats[2][i];
    if (s1.getN() == 0)
        return new double[4];
    DescriptiveStatistics d = s1.getStatistics();
    double median = 0;
    Plot2 plot = null;
    String title = null;
    if (showFilterScoreHistograms) {
        median = d.getPercentile(50);
        String label = String.format("n = %d. Median = %s nm", s1.getN(), Utils.rounded(median));
        int id = Utils.showHistogram(TITLE, s1, xLabel, filterCriteria[i].minBinWidth, (filterCriteria[i].restrictRange) ? 1 : 0, 0, label);
        if (id == 0) {
            IJ.log("Failed to show the histogram: " + xLabel);
            return new double[4];
        }
        if (Utils.isNewWindow())
            wo.add(id);
        title = WindowManager.getImage(id).getTitle();
        // Reverse engineer the histogram settings
        plot = Utils.plot;
        double[] xValues = Utils.xValues;
        int bins = xValues.length;
        double yMin = xValues[0];
        double binSize = xValues[1] - xValues[0];
        double yMax = xValues[0] + (bins - 1) * binSize;
        if (s2.getN() > 0) {
            double[] values = s2.getValues();
            double[][] hist = Utils.calcHistogram(values, yMin, yMax, bins);
            if (hist[0].length > 0) {
                plot.setColor(Color.red);
                plot.addPoints(hist[0], hist[1], Plot2.BAR);
                Utils.display(title, plot);
            }
        }
        if (s3.getN() > 0) {
            double[] values = s3.getValues();
            double[][] hist = Utils.calcHistogram(values, yMin, yMax, bins);
            if (hist[0].length > 0) {
                plot.setColor(Color.blue);
                plot.addPoints(hist[0], hist[1], Plot2.BAR);
                Utils.display(title, plot);
            }
        }
    }
    // Do cumulative histogram
    double[][] h1 = Maths.cumulativeHistogram(s1.getValues(), true);
    double[][] h2 = Maths.cumulativeHistogram(s2.getValues(), true);
    double[][] h3 = Maths.cumulativeHistogram(s3.getValues(), true);
    if (showFilterScoreHistograms) {
        title = TITLE + " Cumul " + xLabel;
        plot = new Plot2(title, xLabel, "Frequency");
        // Find limits
        double[] xlimit = Maths.limits(h1[0]);
        xlimit = Maths.limits(xlimit, h2[0]);
        xlimit = Maths.limits(xlimit, h3[0]);
        // Restrict using the inter-quartile range 
        if (filterCriteria[i].restrictRange) {
            double q1 = d.getPercentile(25);
            double q2 = d.getPercentile(75);
            double iqr = (q2 - q1) * 2.5;
            xlimit[0] = Maths.max(xlimit[0], median - iqr);
            xlimit[1] = Maths.min(xlimit[1], median + iqr);
        }
        plot.setLimits(xlimit[0], xlimit[1], 0, 1.05);
        plot.addPoints(h1[0], h1[1], Plot.LINE);
        plot.setColor(Color.red);
        plot.addPoints(h2[0], h2[1], Plot.LINE);
        plot.setColor(Color.blue);
        plot.addPoints(h3[0], h3[1], Plot.LINE);
    }
    // Determine the maximum difference between the TP and FP
    double maxx1 = 0;
    double maxx2 = 0;
    double max1 = 0;
    double max2 = 0;
    // We cannot compute the delta histogram, or use percentiles
    if (s2.getN() == 0) {
        upper = UpperLimit.ZERO;
        lower = LowerLimit.ZERO;
    }
    final boolean requireLabel = (showFilterScoreHistograms && filterCriteria[i].requireLabel);
    if (requireLabel || upper.requiresDeltaHistogram() || lower.requiresDeltaHistogram()) {
        if (s2.getN() != 0 && s3.getN() != 0) {
            LinearInterpolator li = new LinearInterpolator();
            PolynomialSplineFunction f1 = li.interpolate(h2[0], h2[1]);
            PolynomialSplineFunction f2 = li.interpolate(h3[0], h3[1]);
            for (double x : h1[0]) {
                if (x < h2[0][0] || x < h3[0][0])
                    continue;
                try {
                    double v1 = f1.value(x);
                    double v2 = f2.value(x);
                    double diff = v2 - v1;
                    if (diff > 0) {
                        if (max1 < diff) {
                            max1 = diff;
                            maxx1 = x;
                        }
                    } else {
                        if (max2 > diff) {
                            max2 = diff;
                            maxx2 = x;
                        }
                    }
                } catch (OutOfRangeException e) {
                    // Because we reached the end
                    break;
                }
            }
        } else {
            // Switch to percentiles if we have no delta histogram
            if (upper.requiresDeltaHistogram())
                upper = UpperLimit.NINETY_NINE_PERCENT;
            if (lower.requiresDeltaHistogram())
                lower = LowerLimit.ONE_PERCENT;
        }
    //			System.out.printf("Bounds %s : %s, pos %s, neg %s, %s\n", xLabel, Utils.rounded(getPercentile(h2, 0.01)),
    //					Utils.rounded(maxx1), Utils.rounded(maxx2), Utils.rounded(getPercentile(h1, 0.99)));
    }
    if (showFilterScoreHistograms) {
        // We use bins=1 on charts where we do not need a label
        if (requireLabel) {
            String label = String.format("Max+ %s @ %s, Max- %s @ %s", Utils.rounded(max1), Utils.rounded(maxx1), Utils.rounded(max2), Utils.rounded(maxx2));
            plot.setColor(Color.black);
            plot.addLabel(0, 0, label);
        }
        PlotWindow pw = Utils.display(title, plot);
        if (Utils.isNewWindow())
            wo.add(pw.getImagePlus().getID());
    }
    // Now compute the bounds using the desired limit
    double l, u;
    switch(lower) {
        case ONE_PERCENT:
            l = getPercentile(h2, 0.01);
            break;
        case MAX_NEGATIVE_CUMUL_DELTA:
            l = maxx2;
            break;
        case ZERO:
            l = 0;
            break;
        case HALF_MAX_JACCARD_VALUE:
            l = getValue(metric, j, maxJ * 0.5);
            break;
        default:
            throw new RuntimeException("Missing lower limit method");
    }
    switch(upper) {
        case MAX_POSITIVE_CUMUL_DELTA:
            u = maxx1;
            break;
        case NINETY_NINE_PERCENT:
            u = getPercentile(h2, 0.99);
            break;
        case NINETY_NINE_NINE_PERCENT:
            u = getPercentile(h2, 0.999);
            break;
        case ZERO:
            u = 0;
            break;
        case MAX_JACCARD2:
            u = getValue(metric, j, maxJ) * 2;
            //System.out.printf("MaxJ = %.4f @ %.3f\n", maxJ, u / 2);
            break;
        default:
            throw new RuntimeException("Missing upper limit method");
    }
    double min = getPercentile(h1, 0);
    double max = getPercentile(h1, 1);
    return new double[] { l, u, min, max };
}