Class SpectralClusterer

java.lang.Object

org.ojalgo.data.cluster.FeatureBasedClusterer

org.ojalgo.data.cluster.SpectralClusterer

All Implemented Interfaces:

ClusteringAlgorithm<Point>

final class SpectralClusterer
extends FeatureBasedClusterer

Spectral clustering for Points using a fully connected RBF similarity graph and the symmetric normalised Laplacian. Pairwise distances are computed with the configured DistanceMeasure; the kernel scale is set from the median distance. The algorithm computes the k eigenvectors corresponding to the smallest eigenvalues of L = I - D^{-1/2} W D^{-1/2}, row-normalises the embedding, runs k-means, and maps clusters back to the original points. For n == 0 it returns empty; for n invalid input: '<'= k it returns singletons. Requires point ids contiguous in [0,n). Complexity dominated by O(n^2) pairwise work and an eigen-decomposition of the n×n Laplacian.

Field Summary

Fields

Modifier and Type	Field	Description
private final FeatureBasedClusterer	myClusterer
private final int	myK

Constructor Summary

Constructors

Constructor	Description
SpectralClusterer(int k, DistanceMeasure measure)

Method Summary

Modifier and Type	Method	Description
List<Set<Point>>	cluster(Collection<Point> input)	Partitions the given items into clusters.

Methods inherited from class FeatureBasedClusterer

centroid, cluster, distance, distance, getThreshold, initialiser, isSquared, newAutomatic, newAutomatic, newGreedy, newGreedy, newKMeans, newKMeans, newSpectral, newSpectral, setup

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

myK

private final int myK

myClusterer

private final FeatureBasedClusterer myClusterer

Constructor Details

SpectralClusterer

SpectralClusterer(int k,
 DistanceMeasure measure)

Method Details

cluster

public List<Set<Point>> cluster(Collection<Point> input)

Description copied from interface: ClusteringAlgorithm

Partitions the given items into clusters.

Parameters:

input - The items to cluster; must not be null. May be empty.

Returns:

A list of clusters; each element of the list is a Set<T> representing one cluster and containing its members. Sets are non-empty and pairwise disjoint; ordering is unspecified.