Examples with DBIDArrayIter de.lmu.ifi.dbs.elki.database.ids.DBIDArrayIter used on opensource projects

Example 56 with DBIDArrayIter

use of de.lmu.ifi.dbs.elki.database.ids.DBIDArrayIter in project elki by elki-project.

the class EvaluateDBCV method evaluateClustering.

/**
 * Evaluate a single clustering.
 *
 * @param db Database
 * @param rel Data relation
 * @param cl Clustering
 *
 * @return dbcv DBCV-index
 */
public double evaluateClustering(Database db, Relation<O> rel, Clustering<?> cl) {
    final DistanceQuery<O> dq = rel.getDistanceQuery(distanceFunction);
    List<? extends Cluster<?>> clusters = cl.getAllClusters();
    final int numc = clusters.size();
    // DBCV needs a "dimensionality".
    @SuppressWarnings("unchecked") final Relation<? extends SpatialComparable> vrel = (Relation<? extends SpatialComparable>) rel;
    final int dim = RelationUtil.dimensionality(vrel);
    // precompute all core distances
    ArrayDBIDs[] cids = new ArrayDBIDs[numc];
    double[][] coreDists = new double[numc][];
    for (int c = 0; c < numc; c++) {
        Cluster<?> cluster = clusters.get(c);
        // Singletons are considered as Noise, because they have no sparseness
        if (cluster.isNoise() || cluster.size() < 2) {
            coreDists[c] = null;
            continue;
        }
        // Store for use below:
        ArrayDBIDs ids = cids[c] = DBIDUtil.ensureArray(cluster.getIDs());
        double[] clusterCoreDists = coreDists[c] = new double[ids.size()];
        for (DBIDArrayIter it = ids.iter(), it2 = ids.iter(); it.valid(); it.advance()) {
            double currentCoreDist = 0;
            int neighbors = 0;
            for (it2.seek(0); it2.valid(); it2.advance()) {
                if (DBIDUtil.equal(it, it2)) {
                    continue;
                }
                double dist = dq.distance(it, it2);
                // We ignore such objects.
                if (dist > 0) {
                    currentCoreDist += MathUtil.powi(1. / dist, dim);
                    ++neighbors;
                }
            }
            // Average, and undo power.
            clusterCoreDists[it.getOffset()] = FastMath.pow(currentCoreDist / neighbors, -1. / dim);
        }
    }
    // compute density sparseness of all clusters
    int[][] clusterDegrees = new int[numc][];
    double[] clusterDscMax = new double[numc];
    // describes if a cluster contains any internal edges
    boolean[] internalEdges = new boolean[numc];
    for (int c = 0; c < numc; c++) {
        Cluster<?> cluster = clusters.get(c);
        if (cluster.isNoise() || cluster.size() < 2) {
            clusterDegrees[c] = null;
            clusterDscMax[c] = Double.NaN;
            continue;
        }
        double[] clusterCoreDists = coreDists[c];
        ArrayDBIDs ids = cids[c];
        // Density Sparseness of the Cluster
        double dscMax = 0;
        double[][] distances = new double[cluster.size()][cluster.size()];
        // create mutability distance matrix for Minimum Spanning Tree
        for (DBIDArrayIter it = ids.iter(), it2 = ids.iter(); it.valid(); it.advance()) {
            double currentCoreDist = clusterCoreDists[it.getOffset()];
            for (it2.seek(it.getOffset() + 1); it2.valid(); it2.advance()) {
                double mutualReachDist = MathUtil.max(currentCoreDist, clusterCoreDists[it2.getOffset()], dq.distance(it, it2));
                distances[it.getOffset()][it2.getOffset()] = mutualReachDist;
                distances[it2.getOffset()][it.getOffset()] = mutualReachDist;
            }
        }
        // generate Minimum Spanning Tree
        int[] nodes = PrimsMinimumSpanningTree.processDense(distances);
        // get degree of all nodes in the spanning tree
        int[] degree = new int[cluster.size()];
        for (int i = 0; i < nodes.length; i++) {
            degree[nodes[i]]++;
        }
        // check if cluster contains any internal edges
        for (int i = 0; i < nodes.length; i += 2) {
            if (degree[nodes[i]] > 1 && degree[nodes[i + 1]] > 1) {
                internalEdges[c] = true;
            }
        }
        clusterDegrees[c] = degree;
        // find maximum sparseness in the Minimum Spanning Tree
        for (int i = 0; i < nodes.length; i = i + 2) {
            final int n1 = nodes[i], n2 = nodes[i + 1];
            // If a cluster has no internal nodes we consider all edges.
            if (distances[n1][n2] > dscMax && (!internalEdges[c] || (degree[n1] > 1 && degree[n2] > 1))) {
                dscMax = distances[n1][n2];
            }
        }
        clusterDscMax[c] = dscMax;
    }
    // compute density separation of all clusters
    double dbcv = 0;
    for (int c = 0; c < numc; c++) {
        Cluster<?> cluster = clusters.get(c);
        if (cluster.isNoise() || cluster.size() < 2) {
            continue;
        }
        double currentDscMax = clusterDscMax[c];
        double[] clusterCoreDists = coreDists[c];
        int[] currentDegree = clusterDegrees[c];
        // minimal Density Separation of the Cluster
        double dspcMin = Double.POSITIVE_INFINITY;
        for (DBIDArrayIter it = cids[c].iter(); it.valid(); it.advance()) {
            // nodes.
            if (currentDegree[it.getOffset()] < 2 && internalEdges[c]) {
                continue;
            }
            double currentCoreDist = clusterCoreDists[it.getOffset()];
            for (int oc = 0; oc < numc; oc++) {
                Cluster<?> ocluster = clusters.get(oc);
                if (ocluster.isNoise() || ocluster.size() < 2 || cluster == ocluster) {
                    continue;
                }
                int[] oDegree = clusterDegrees[oc];
                double[] oclusterCoreDists = coreDists[oc];
                for (DBIDArrayIter it2 = cids[oc].iter(); it2.valid(); it2.advance()) {
                    if (oDegree[it2.getOffset()] < 2 && internalEdges[oc]) {
                        continue;
                    }
                    double mutualReachDist = MathUtil.max(currentCoreDist, oclusterCoreDists[it2.getOffset()], dq.distance(it, it2));
                    dspcMin = mutualReachDist < dspcMin ? mutualReachDist : dspcMin;
                }
            }
        }
        // compute DBCV
        double vc = (dspcMin - currentDscMax) / MathUtil.max(dspcMin, currentDscMax);
        double weight = cluster.size() / (double) rel.size();
        dbcv += weight * vc;
    }
    EvaluationResult ev = EvaluationResult.findOrCreate(db.getHierarchy(), cl, "Internal Clustering Evaluation", "internal evaluation");
    MeasurementGroup g = ev.findOrCreateGroup("Distance-based Evaluation");
    g.addMeasure("Density Based Clustering Validation", dbcv, 0., Double.POSITIVE_INFINITY, 0., true);
    db.getHierarchy().resultChanged(ev);
    return dbcv;
}

Example 57 with DBIDArrayIter

use of de.lmu.ifi.dbs.elki.database.ids.DBIDArrayIter in project elki by elki-project.

the class EvaluateSilhouette method evaluateClustering.

/**
 * Evaluate a single clustering.
 *
 * @param db Database
 * @param rel Data relation
 * @param dq Distance query
 * @param c Clustering
 * @return Average silhouette
 */
public double evaluateClustering(Database db, Relation<O> rel, DistanceQuery<O> dq, Clustering<?> c) {
    List<? extends Cluster<?>> clusters = c.getAllClusters();
    MeanVariance msil = new MeanVariance();
    int ignorednoise = 0;
    for (Cluster<?> cluster : clusters) {
        // Note: we treat 1-element clusters the same as noise.
        if (cluster.size() <= 1 || cluster.isNoise()) {
            switch(noiseOption) {
                case IGNORE_NOISE:
                    ignorednoise += cluster.size();
                    // Ignore noise elements
                    continue;
                case TREAT_NOISE_AS_SINGLETONS:
                    // As suggested in Rousseeuw, we use 0 for singletons.
                    msil.put(0., cluster.size());
                    continue;
                case MERGE_NOISE:
                    // Treat as cluster below
                    break;
            }
        }
        ArrayDBIDs ids = DBIDUtil.ensureArray(cluster.getIDs());
        // temporary storage.
        double[] as = new double[ids.size()];
        DBIDArrayIter it1 = ids.iter(), it2 = ids.iter();
        for (it1.seek(0); it1.valid(); it1.advance()) {
            // a: In-cluster distances
            // Already computed distances
            double a = as[it1.getOffset()];
            for (it2.seek(it1.getOffset() + 1); it2.valid(); it2.advance()) {
                final double dist = dq.distance(it1, it2);
                a += dist;
                as[it2.getOffset()] += dist;
            }
            a /= (ids.size() - 1);
            // b: minimum average distance to other clusters:
            double b = Double.POSITIVE_INFINITY;
            for (Cluster<?> ocluster : clusters) {
                if (ocluster == /* yes, reference identity */
                cluster) {
                    // Same cluster
                    continue;
                }
                if (ocluster.size() <= 1 || ocluster.isNoise()) {
                    switch(noiseOption) {
                        case IGNORE_NOISE:
                            // Ignore noise elements
                            continue;
                        case TREAT_NOISE_AS_SINGLETONS:
                            // Treat noise cluster as singletons:
                            for (DBIDIter it3 = ocluster.getIDs().iter(); it3.valid(); it3.advance()) {
                                final double dist = dq.distance(it1, it3);
                                // Minimum average
                                b = dist < b ? dist : b;
                            }
                            continue;
                        case MERGE_NOISE:
                            // Treat as cluster below
                            break;
                    }
                }
                final DBIDs oids = ocluster.getIDs();
                double btmp = 0.;
                for (DBIDIter it3 = oids.iter(); it3.valid(); it3.advance()) {
                    btmp += dq.distance(it1, it3);
                }
                // Average
                btmp /= oids.size();
                // Minimum average
                b = btmp < b ? btmp : b;
            }
            // One cluster only?
            b = b < Double.POSITIVE_INFINITY ? b : a;
            msil.put((b - a) / (b > a ? b : a));
        }
    }
    double penalty = 1.;
    // Only if {@link NoiseHandling#IGNORE_NOISE}:
    if (penalize && ignorednoise > 0) {
        penalty = (rel.size() - ignorednoise) / (double) rel.size();
    }
    final double meansil = penalty * msil.getMean();
    final double stdsil = penalty * msil.getSampleStddev();
    if (LOG.isStatistics()) {
        LOG.statistics(new StringStatistic(key + ".silhouette.noise-handling", noiseOption.toString()));
        if (ignorednoise > 0) {
            LOG.statistics(new LongStatistic(key + ".silhouette.noise", ignorednoise));
        }
        LOG.statistics(new DoubleStatistic(key + ".silhouette.mean", meansil));
        LOG.statistics(new DoubleStatistic(key + ".silhouette.stddev", stdsil));
    }
    EvaluationResult ev = EvaluationResult.findOrCreate(db.getHierarchy(), c, "Internal Clustering Evaluation", "internal evaluation");
    MeasurementGroup g = ev.findOrCreateGroup("Distance-based Evaluation");
    g.addMeasure("Silhouette +-" + FormatUtil.NF2.format(stdsil), meansil, -1., 1., 0., false);
    db.getHierarchy().resultChanged(ev);
    return meansil;
}

Example 58 with DBIDArrayIter

use of de.lmu.ifi.dbs.elki.database.ids.DBIDArrayIter in project elki by elki-project.

the class PrecomputedSimilarityMatrix method initialize.

@Override
public void initialize() {
    DBIDs rids = relation.getDBIDs();
    if (!(rids instanceof DBIDRange)) {
        throw new AbortException("Similarity matrixes are currently only supported for DBID ranges (as used by static databases) for performance reasons (Patches welcome).");
    }
    ids = (DBIDRange) rids;
    size = ids.size();
    if (size > 65536) {
        throw new AbortException("Similarity matrixes currently have a limit of 65536 objects (~16 GB). After this, the array size exceeds the Java integer range, and a different data structure needs to be used.");
    }
    similarityQuery = similarityFunction.instantiate(relation);
    int msize = triangleSize(size);
    matrix = new double[msize];
    DBIDArrayIter ix = ids.iter(), iy = ids.iter();
    FiniteProgress prog = LOG.isVerbose() ? new FiniteProgress("Precomputing similarity matrix", msize, LOG) : null;
    int pos = 0;
    for (ix.seek(0); ix.valid(); ix.advance()) {
        // y < x -- must match {@link #getOffset}!
        for (iy.seek(0); iy.getOffset() < ix.getOffset(); iy.advance()) {
            matrix[pos] = similarityQuery.similarity(ix, iy);
            pos++;
        }
        if (prog != null) {
            prog.setProcessed(prog.getProcessed() + ix.getOffset(), LOG);
        }
    }
    LOG.ensureCompleted(prog);
}

Example 59 with DBIDArrayIter

use of de.lmu.ifi.dbs.elki.database.ids.DBIDArrayIter in project elki by elki-project.

the class PrecomputeDistancesAsciiApplication method run.

@Override
public void run() {
    database.initialize();
    Relation<O> relation = database.getRelation(distance.getInputTypeRestriction());
    DistanceQuery<O> distanceQuery = database.getDistanceQuery(relation, distance);
    DBIDRange ids = DBIDUtil.assertRange(relation.getDBIDs());
    final int size = ids.size();
    FiniteProgress prog = LOG.isVerbose() ? new FiniteProgress("Precomputing distances", (int) (((size - 1) * (long) size) >>> 1), LOG) : null;
    try (PrintStream fout = openStream(out)) {
        DBIDArrayIter id1 = ids.iter(), id2 = ids.iter();
        for (; id1.valid(); id1.advance()) {
            String idstr1 = Integer.toString(id1.getOffset());
            if (debugExtraCheckSymmetry && distanceQuery.distance(id1, id1) != 0.) {
                LOG.warning("Distance function doesn't satisfy d(0,0) = 0.");
            }
            for (id2.seek(id1.getOffset() + 1); id2.valid(); id2.advance()) {
                double d = distanceQuery.distance(id1, id2);
                if (debugExtraCheckSymmetry) {
                    double d2 = distanceQuery.distance(id2, id1);
                    if (Math.abs(d - d2) > 0.0000001) {
                        LOG.warning("Distance function doesn't appear to be symmetric!");
                    }
                }
                // 
                fout.append(idstr1).append('\t').append(Integer.toString(id2.getOffset())).append(// 
                '\t').append(Double.toString(d)).append('\n');
            }
            if (prog != null) {
                prog.setProcessed(prog.getProcessed() + (size - id1.getOffset() - 1), LOG);
            }
        }
    } catch (IOException e) {
        throw new AbortException("Could not write to output file.", e);
    }
    LOG.ensureCompleted(prog);
}

Example 60 with DBIDArrayIter

use of de.lmu.ifi.dbs.elki.database.ids.DBIDArrayIter in project elki by elki-project.

the class CacheFloatDistanceInOnDiskMatrix method run.

@Override
public void run() {
    database.initialize();
    Relation<O> relation = database.getRelation(distance.getInputTypeRestriction());
    DistanceQuery<O> distanceQuery = database.getDistanceQuery(relation, distance);
    DBIDRange ids = DBIDUtil.assertRange(relation.getDBIDs());
    int size = ids.size();
    FiniteProgress prog = LOG.isVerbose() ? new FiniteProgress("Precomputing distances", (int) (((size + 1) * (long) size) >>> 1), LOG) : null;
    try (// 
    OnDiskUpperTriangleMatrix matrix = new OnDiskUpperTriangleMatrix(out, DiskCacheBasedFloatDistanceFunction.FLOAT_CACHE_MAGIC, 0, ByteArrayUtil.SIZE_FLOAT, size)) {
        DBIDArrayIter id1 = ids.iter(), id2 = ids.iter();
        for (; id1.valid(); id1.advance()) {
            for (id2.seek(id1.getOffset()); id2.valid(); id2.advance()) {
                float d = (float) distanceQuery.distance(id1, id2);
                if (debugExtraCheckSymmetry) {
                    float d2 = (float) distanceQuery.distance(id2, id1);
                    if (Math.abs(d - d2) > 0.0000001) {
                        LOG.warning("Distance function doesn't appear to be symmetric!");
                    }
                }
                try {
                    matrix.getRecordBuffer(id1.getOffset(), id2.getOffset()).putFloat(d);
                } catch (IOException e) {
                    throw new AbortException("Error writing distance record " + DBIDUtil.toString(id1) + "," + DBIDUtil.toString(id2) + " to matrix.", e);
                }
            }
            if (prog != null) {
                prog.setProcessed(prog.getProcessed() + (size - id1.getOffset()), LOG);
            }
        }
    } catch (IOException e) {
        throw new AbortException("Error precomputing distance matrix.", e);
    }
    prog.ensureCompleted(LOG);
}