de.jungblut.clustering

Class OnePassExclusiveClustering

java.lang.Object

de.jungblut.clustering.OnePassExclusiveClustering

public final class OnePassExclusiveClustering
extends Object

A one pass exclusive clustering algorithm. As the name suggests, the clustering algorithm will iterate once over a constructed kd-tree and find nearest neighbours inside a distance threshold. The found neighbours are going to be put into a bitset and will be omitted from search in the following kd-tree searches. Found clusters are checked against a minimum size and maybe discarded when not reaching the configured threshold. This is considered a very fast algorithm, it can be used instead of CanopyClustering.

Author:

thomas.jungblut

Constructor Summary

Constructors

Constructor and Description
OnePassExclusiveClustering(double t1) Constructs a one pass clustering algorithm.
OnePassExclusiveClustering(double t1, int k, int minSize, boolean mergeOverlaps) Constructs a one pass clustering algorithm.

Method Summary

Modifier and Type	Method and Description
List<de.jungblut.math.DoubleVector>	cluster(List<de.jungblut.math.DoubleVector> values, boolean verbose) Cluster the given items.

Methods inherited from class java.lang.Object

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

Constructor Detail

OnePassExclusiveClustering

public OnePassExclusiveClustering(double t1)

Constructs a one pass clustering algorithm. With unlimited maximum number of neighbours retrieved and a minimum cluster size of 2.

Parameters:

t1 - the maximum distance of neighbourhood.

OnePassExclusiveClustering

public OnePassExclusiveClustering(double t1,
                                  int k,
                                  int minSize,
                                  boolean mergeOverlaps)

Constructs a one pass clustering algorithm.

Parameters:

t1 - the maximum distance of neighbourhood.

k - the maximum number of neighbours to retrieve inside the t1 threshold.

minSize - the minimum size of a cluster.

mergeOverlaps - if true, overlapping found centers by t1 distance will be merged.

Method Detail

cluster

public List<de.jungblut.math.DoubleVector> cluster(List<de.jungblut.math.DoubleVector> values,
                                                   boolean verbose)

Cluster the given items.

Parameters:

values - the vectors to cluster.

verbose - if true, outputs progress to STDOUT.

Returns:

a list of centers that describe the given vectors.