@Override
public Clustering<KMeansModel> run(Database database, Relation<V> relation) {
if (relation.size() <= 0) {
return new Clustering<>("k-Means Clustering", "kmeans-clustering");
}
// Choose initial means
if (LOG.isStatistics()) {
LOG.statistics(new StringStatistic(KEY + ".initialization", initializer.toString()));
}
double[][] means = initializer.chooseInitialMeans(database, relation, k, getDistanceFunction());
// Setup cluster assignment store
List<ModifiableDBIDs> clusters = new ArrayList<>();
for (int i = 0; i < k; i++) {
clusters.add(DBIDUtil.newHashSet((int) (relation.size() * 2. / k)));
}
WritableIntegerDataStore assignment =
DataStoreUtil.makeIntegerStorage(
relation.getDBIDs(), DataStoreFactory.HINT_TEMP | DataStoreFactory.HINT_HOT, -1);
double[] varsum = new double[k];
IndefiniteProgress prog =
LOG.isVerbose() ? new IndefiniteProgress("K-Means iteration", LOG) : null;
DoubleStatistic varstat =
LOG.isStatistics()
? new DoubleStatistic(this.getClass().getName() + ".variance-sum")
: null;
int iteration = 0;
for (; maxiter <= 0 || iteration < maxiter; iteration++) {
LOG.incrementProcessed(prog);
boolean changed = assignToNearestCluster(relation, means, clusters, assignment, varsum);
logVarstat(varstat, varsum);
// Stop if no cluster assignment changed.
if (!changed) {
break;
}
// Recompute means.
means = means(clusters, means, relation);
}
LOG.setCompleted(prog);
if (LOG.isStatistics()) {
LOG.statistics(new LongStatistic(KEY + ".iterations", iteration));
}
// Wrap result
Clustering<KMeansModel> result = new Clustering<>("k-Means Clustering", "kmeans-clustering");
for (int i = 0; i < clusters.size(); i++) {
DBIDs ids = clusters.get(i);
if (ids.size() == 0) {
continue;
}
KMeansModel model = new KMeansModel(means[i], varsum[i]);
result.addToplevelCluster(new Cluster<>(ids, model));
}
return result;
}
/**
* Run the algorithm
*
* @param db Database
* @param relation Relation
* @return Clustering hierarchy
*/
public PointerHierarchyRepresentationResult run(Database db, Relation<O> relation) {
DistanceQuery<O> dq = db.getDistanceQuery(relation, getDistanceFunction());
ArrayDBIDs ids = DBIDUtil.ensureArray(relation.getDBIDs());
final int size = ids.size();
if (size > 0x10000) {
throw new AbortException(
"This implementation does not scale to data sets larger than "
+ 0x10000
+ " instances (~17 GB RAM), which results in an integer overflow.");
}
if (Linkage.SINGLE.equals(linkage)) {
LOG.verbose("Notice: SLINK is a much faster algorithm for single-linkage clustering!");
}
// Compute the initial (lower triangular) distance matrix.
double[] scratch = new double[triangleSize(size)];
DBIDArrayIter ix = ids.iter(), iy = ids.iter(), ij = ids.iter();
// Position counter - must agree with computeOffset!
int pos = 0;
boolean square =
Linkage.WARD.equals(linkage)
&& !(SquaredEuclideanDistanceFunction.class.isInstance(getDistanceFunction()));
for (int x = 0; ix.valid(); x++, ix.advance()) {
iy.seek(0);
for (int y = 0; y < x; y++, iy.advance()) {
scratch[pos] = dq.distance(ix, iy);
// Ward uses variances -- i.e. squared values
if (square) {
scratch[pos] *= scratch[pos];
}
pos++;
}
}
// Initialize space for result:
WritableDBIDDataStore parent =
DataStoreUtil.makeDBIDStorage(
ids, DataStoreFactory.HINT_HOT | DataStoreFactory.HINT_STATIC);
WritableDoubleDataStore height =
DataStoreUtil.makeDoubleStorage(
ids, DataStoreFactory.HINT_HOT | DataStoreFactory.HINT_STATIC);
WritableIntegerDataStore csize =
DataStoreUtil.makeIntegerStorage(
ids, DataStoreFactory.HINT_HOT | DataStoreFactory.HINT_TEMP);
for (DBIDIter it = ids.iter(); it.valid(); it.advance()) {
parent.put(it, it);
height.put(it, Double.POSITIVE_INFINITY);
csize.put(it, 1);
}
// Repeat until everything merged, except the desired number of clusters:
FiniteProgress prog =
LOG.isVerbose() ? new FiniteProgress("Agglomerative clustering", size - 1, LOG) : null;
for (int i = 1; i < size; i++) {
double min = Double.POSITIVE_INFINITY;
int minx = -1, miny = -1;
for (ix.seek(0); ix.valid(); ix.advance()) {
if (height.doubleValue(ix) < Double.POSITIVE_INFINITY) {
continue;
}
final int xbase = triangleSize(ix.getOffset());
for (iy.seek(0); iy.getOffset() < ix.getOffset(); iy.advance()) {
if (height.doubleValue(iy) < Double.POSITIVE_INFINITY) {
continue;
}
final int idx = xbase + iy.getOffset();
if (scratch[idx] <= min) {
min = scratch[idx];
minx = ix.getOffset();
miny = iy.getOffset();
}
}
}
assert (minx >= 0 && miny >= 0);
// Avoid allocating memory, by reusing existing iterators:
ix.seek(minx);
iy.seek(miny);
// Perform merge in data structure: x -> y
// Since y < x, prefer keeping y, dropping x.
int sizex = csize.intValue(ix), sizey = csize.intValue(iy);
height.put(ix, min);
parent.put(ix, iy);
csize.put(iy, sizex + sizey);
// Update distance matrix. Note: miny < minx
final int xbase = triangleSize(minx), ybase = triangleSize(miny);
// Write to (y, j), with j < y
for (ij.seek(0); ij.getOffset() < miny; ij.advance()) {
if (height.doubleValue(ij) < Double.POSITIVE_INFINITY) {
continue;
}
final int sizej = csize.intValue(ij);
scratch[ybase + ij.getOffset()] =
linkage.combine(
sizex,
scratch[xbase + ij.getOffset()],
sizey,
scratch[ybase + ij.getOffset()],
sizej,
min);
}
// Write to (j, y), with y < j < x
for (ij.seek(miny + 1); ij.getOffset() < minx; ij.advance()) {
if (height.doubleValue(ij) < Double.POSITIVE_INFINITY) {
continue;
}
final int jbase = triangleSize(ij.getOffset());
final int sizej = csize.intValue(ij);
scratch[jbase + miny] =
linkage.combine(
sizex, scratch[xbase + ij.getOffset()], sizey, scratch[jbase + miny], sizej, min);
}
// Write to (j, y), with y < x < j
for (ij.seek(minx + 1); ij.valid(); ij.advance()) {
if (height.doubleValue(ij) < Double.POSITIVE_INFINITY) {
continue;
}
final int jbase = triangleSize(ij.getOffset());
final int sizej = csize.intValue(ij);
scratch[jbase + miny] =
linkage.combine(sizex, scratch[jbase + minx], sizey, scratch[jbase + miny], sizej, min);
}
LOG.incrementProcessed(prog);
}
LOG.ensureCompleted(prog);
return new PointerHierarchyRepresentationResult(ids, parent, height);
}