org.openimaj.ml.clustering.incremental

Class IncrementalSparseClusterer

java.lang.Object

org.openimaj.ml.clustering.incremental.IncrementalSparseClusterer

All Implemented Interfaces:

Clusterer<SparseMatrix>, DataClusterer<SparseMatrix,IndexClusters>, SparseMatrixClusterer<IndexClusters>

Direct Known Subclasses:

IncrementalLifetimeSparseClusterer

public class IncrementalSparseClusterer
extends Object
implements SparseMatrixClusterer<IndexClusters>

An incremental clusterer which holds old SparseMatrix instances internally, only forgetting rows once they have been clustered and are relatively stable. The criteria for row removal is cluster stability. The defenition of cluster stability is maximum f1-score achieving a threshold between clusters in the previous round and the current round. Once one round of stability is achieved the cluster is stable and its elements are removed.

Author:

Sina Samangooei (ss@ecs.soton.ac.uk)

Field Summary

Fields

Modifier and Type	Field and Description
protected double	threshold

Constructor Summary

Constructors

Constructor and Description
IncrementalSparseClusterer(SparseMatrixClusterer<? extends IndexClusters> clusterer, int window)
IncrementalSparseClusterer(SparseMatrixClusterer<? extends IndexClusters> clusterer, int window, double threshold)

Method Summary

Methods

Modifier and Type	Method and Description
protected Map<Integer,IntDoublePair>	calculateStability(IndexClusters c1, IndexClusters c2, gnu.trove.set.TIntSet inactiveRows)
IndexClusters	cluster(SparseMatrix data)
protected void	detectInactive(IndexClusters oldClusters, IndexClusters newClusters, gnu.trove.set.TIntSet inactiveRows, List<int[]> completedClusters) Given the old and new clusters, make a decision as to which rows are now inactive, and therefore which clusters are now completed
int[][]	performClustering(SparseMatrix data)

Methods inherited from class java.lang.Object

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

Field Detail

threshold

protected double threshold

Constructor Detail

IncrementalSparseClusterer

public IncrementalSparseClusterer(SparseMatrixClusterer<? extends IndexClusters> clusterer,
                          int window)

Parameters:

clusterer - the underlying clusterer

window -

IncrementalSparseClusterer

public IncrementalSparseClusterer(SparseMatrixClusterer<? extends IndexClusters> clusterer,
                          int window,
                          double threshold)

Parameters:

clusterer - the underlying clusterer

window -

threshold -

Method Detail

cluster

public IndexClusters cluster(SparseMatrix data)

Specified by:

cluster in interface DataClusterer<SparseMatrix,IndexClusters>

Parameters:

data - the data to be clustered

Returns:

the clusters produced

detectInactive

protected void detectInactive(IndexClusters oldClusters,
                  IndexClusters newClusters,
                  gnu.trove.set.TIntSet inactiveRows,
                  List<int[]> completedClusters)

Given the old and new clusters, make a decision as to which rows are now inactive, and therefore which clusters are now completed

Parameters:

oldClusters -

newClusters -

inactiveRows -

completedClusters -

calculateStability

protected Map<Integer,IntDoublePair> calculateStability(IndexClusters c1,
                                            IndexClusters c2,
                                            gnu.trove.set.TIntSet inactiveRows)

performClustering

public int[][] performClustering(SparseMatrix data)

Specified by:

performClustering in interface Clusterer<SparseMatrix>

Returns:

Given data items, cluster them by index