Examples with InAxDoubleArray org.twak.utils.streams.InAxDoubleArray used on opensource projects

Example 1 with InAxDoubleArray

use of org.twak.utils.streams.InAxDoubleArray in project chordatlas by twak.

the class FindLines method regress.

private Line regress(Set<Line> things, LinearForm lfDir) {
    SimpleRegression fit = new SimpleRegression();
    // regression isn't happy on lines with infinite slope: so swap params!
    boolean flip = Math.abs(lfDir.x) > Math.abs(lfDir.y);
    for (Line l : things) if (flip) {
        fit.addData(l.start.y, l.start.x);
        fit.addData(l.end.y, l.end.x);
    } else {
        fit.addData(l.start.x, l.start.y);
        fit.addData(l.end.x, l.end.y);
    }
    double intercept = fit.getIntercept(), slope = fit.getSlope();
    if (Double.isNaN(intercept))
        return null;
    LinearForm lf;
    if (flip)
        lf = new LinearForm(1, -slope);
    else
        lf = new LinearForm(-slope, 1);
    if (lf.unitVector().angle(lfDir.unitVector()) > Math.PI / 2) {
        // if regression is pointing wrong way, flip
        lf.x = -lf.x;
        lf.y = -lf.y;
    }
    if (flip)
        lf.findC(intercept, 0);
    else
        lf.findC(0, intercept);
    double[] minMax = things.stream().map(x -> new double[] { lf.findPParam(x.start), lf.findPParam(x.end) }).collect(new InAxDoubleArray());
    // do regression
    return new Line(lf, minMax[0], minMax[1]);
}

Example 2 with InAxDoubleArray

use of org.twak.utils.streams.InAxDoubleArray in project chordatlas by twak.

the class Prof method findProfileLines.

/**
 * We find an initial base offset. Then we cluster the start point of all
 * (clean) profiles. If any are a good distance from the initial base, we
 * add those as their own profile lines.
 *
 * The original line is offset by the remaiing data.
 */
public static List<SuperLine> findProfileLines(Collection<Prof> profiles, Line3d line) {
    List<SuperLine> out = new ArrayList();
    // PaintThing.debug.clear();
    SuperLine superLine = new SuperLine(line.start.x, line.start.z, line.end.x, line.end.z);
    double outLen = superLine.length();
    double min = Double.MAX_VALUE, max = -Double.MAX_VALUE;
    Cache<Prof, Double> vLength = new Cache<Prof, Double>() {

        @Override
        public Double create(Prof i) {
            return i.verticalLength(0.5);
        }
    };
    double vLen = profiles.stream().mapToDouble(p -> vLength.get(p)).sum();
    boolean useVertical = vLen / profiles.size() > 1;
    class Wrapper implements Clusterable {

        double[] pt;

        public Wrapper(Point2d pt) {
            this.pt = new double[] { pt.x, pt.y };
        }

        @Override
        public double[] getPoint() {
            return pt;
        }
    }
    List<Wrapper> toCluster = new ArrayList();
    List<Double> baseLineOffset = new ArrayList();
    for (Prof p : profiles) {
        if (// vLen / (5*profiles.size()))
        useVertical && vLength.get(p) < 1)
            continue;
        Prof clean = p.parameterize();
        Point2d pt = clean.get(0);
        Point3d pt3 = clean.to3d(pt);
        double ppram = superLine.findPPram(new Point2d(pt3.x, pt3.z));
        baseLineOffset.add(pt.x);
        toCluster.add(new Wrapper(new Point2d(pt.x, ppram * outLen)));
        min = Math.min(min, ppram);
        max = Math.max(max, ppram);
    }
    if (min == max || toCluster.isEmpty())
        return out;
    if (true) {
        baseLineOffset.sort(Double::compareTo);
        double modeBaselineOffset = baseLineOffset.get(baseLineOffset.size() / 2);
        DBSCANClusterer<Wrapper> cr = new DBSCANClusterer<>(1.5, 0);
        List<Cluster<Wrapper>> results = cr.cluster(toCluster);
        Iterator<Cluster<Wrapper>> cit = results.iterator();
        while (cit.hasNext()) {
            Cluster<Wrapper> cw = cit.next();
            if (cw.getPoints().size() < 2 / TweedSettings.settings.profileHSampleDist) {
                cit.remove();
                double cMeanY = cw.getPoints().stream().mapToDouble(x -> x.pt[1]).average().getAsDouble();
                double bestDist = Double.MAX_VALUE;
                Cluster<Wrapper> bestWrapper = null;
                for (Cluster<Wrapper> near : results) {
                    double meanY = near.getPoints().stream().mapToDouble(x -> x.pt[1]).average().getAsDouble();
                    double dist = Math.abs(meanY - cMeanY);
                    if (dist < bestDist) {
                        bestDist = dist;
                        bestWrapper = near;
                    }
                }
                if (bestWrapper != null)
                    bestWrapper.getPoints().addAll(cw.getPoints());
            }
        }
        {
            baseLineOffset.clear();
            int c = 0;
            for (Cluster<Wrapper> cw : results) {
                double[] minMax = cw.getPoints().stream().map(p -> new double[] { p.pt[1] }).collect(new InAxDoubleArray());
                double[] offsetA = cw.getPoints().stream().mapToDouble(p -> p.pt[0]).sorted().toArray();
                double offset = offsetA[offsetA.length / 2];
                if (offset - modeBaselineOffset < 1) {
                    for (Wrapper w : cw.getPoints()) baseLineOffset.add(w.pt[0]);
                    continue;
                }
                SuperLine sl = new SuperLine(superLine.fromPPram(minMax[0] / outLen), superLine.fromPPram(minMax[1] / outLen));
                sl.moveLeft(offset);
                out.add(sl);
                List<Point2d> pts = cw.getPoints().stream().map(w -> new Point2d(w.pt[0], w.pt[1])).collect(Collectors.toList());
                PaintThing.debug(Rainbow.getColour(c++), 1, pts);
            }
        }
    }
    Point2d nStart = superLine.fromPPram(min), nEnd = superLine.fromPPram(max);
    superLine.start = nStart;
    superLine.end = nEnd;
    baseLineOffset.sort(Double::compare);
    if (!baseLineOffset.isEmpty())
        superLine.moveLeft(baseLineOffset.get(baseLineOffset.size() / 2));
    out.add(0, superLine);
    return out;
}