com.chemaxon.clustering.wards

Class LanceWilliamsAlgorithm

java.lang.Object

com.chemaxon.clustering.wards.LanceWilliamsAlgorithm

@Beta
 @PublicAPI
public final class LanceWilliamsAlgorithm
extends Object

Agglomerative hierarchical clustering implementation for Lance-Williams algorithm family.

Licensing: this class is part of the JKlustor suite; it can be used with valid JKlustor license.

Please note that this class is marked with @Beta annotation, so it can be subject of incompatible changes or removal in later releases.

See Also:

http://en.wikipedia.org/wiki/Ward%27s_method#Lance.E2.80.93Williams_algorithms

Method Summary

Modifier and Type	Method and Description
static double[][]	calculateDissimilarityMatrix(DissimilarityInput input, SubProgressObserver po) Distance matrix calculation.
static IDBasedHierarchicClustering	cluster(DissimilarityInput input, LanceWilliamsMerge merge, ProgressObserver po) Invoke clustering.
static IDBasedHierarchicClustering	cluster(double[][] dissim, LanceWilliamsMerge merge, SubProgressObserver po) Invoke clustering.

Methods inherited from class java.lang.Object

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

Method Detail

calculateDissimilarityMatrix

public static double[][] calculateDissimilarityMatrix(DissimilarityInput input,
                                                      SubProgressObserver po)

Distance matrix calculation.

Parameters:

input - Input dissimilarity space

po - Progress observer

Returns:

Distance matrix, with [<TARGET>][<QUERY>] indices

cluster

public static IDBasedHierarchicClustering cluster(DissimilarityInput input,
                                                  LanceWilliamsMerge merge,
                                                  ProgressObserver po)
                                           throws CancellationException

Invoke clustering.

The current implementation requires O(n2) storage and O(n2) pairwise input comparison invocation. Over the stored dissimilarity values it requires O(n3) simple numerical comparison steps.

Parameters:

input - Input dissimilarity space

merge - Cluster distance update function

po - Observing progress observer. Two ProgressObserver.subTask(java.lang.String, long)} will be associate to the distance matrix calculation and to the clustering iterations

Returns:

Clustered hierarchy, preferred levels set as cluster distances

Throws:

CancellationException - Upon cancellation through the given ProgressObserver

IllegalArgumentException - When less than 2 input specified

IllegalStateException - When distance inconsistency found during merging: the distance of a merged cluster and any other cluster must be not less than the distance of the merging clusters

LicenseException - when appropriate license is not available

cluster

public static IDBasedHierarchicClustering cluster(double[][] dissim,
                                                  LanceWilliamsMerge merge,
                                                  SubProgressObserver po)
                                           throws CancellationException

Invoke clustering.

The current implementation requires O(n2) storage and O(n2) pairwise input comparison invocation. Over the stored dissimilarity values it requires O(n3) simple numerical comparison steps.

Parameters:

dissim - Distance matrix, with [<TARGET>][<QUERY>] indices

merge - Cluster distance update function

po - Observing to track progress. Will be closed upon finishing.

Returns:

Clustered hierarchy, preferred levels set as cluster distances

Throws:

CancellationException - Upon cancellation through the given ProgressObserver

IllegalArgumentException - When less than 2 input specified

IllegalStateException - When distance inconsistency found during merging: the distance of a merged cluster and any other cluster must be not less than the distance of the merging clusters

LicenseException - when appropriate license is not available