edu.ucla.sspace.clustering.criterion

Class HybridBaseFunction

java.lang.Object

edu.ucla.sspace.clustering.criterion.HybridBaseFunction

All Implemented Interfaces:

CriterionFunction

Direct Known Subclasses:

H1Function, H2Function

public abstract class HybridBaseFunction
extends Object
implements CriterionFunction

This CriterionFunction implements the basic functionality needed for a majority of the hybrid functions available. It works by first gathering a handful of meta data for the data set, such as the cluster sizes, initial cluster assignments, and initial centroids. It then implements update and requires subclasses to implement functions for determining the change in the criterion score due to moving a data point. Hybrid CriterionFunctions utilize an internal and external criterion function in order to balance between both objectives in order to create a well balanced clustering.

Author:

Keith Stevens

Field Summary

Fields

Modifier and Type	Field and Description
protected int[]	assignments The set of cluster assignments for each cluster.
protected DoubleVector[]	centroids The centroids representing each cluster.
protected int[]	clusterSizes The number of data points found in each cluster.
protected DoubleVector	completeCentroid The summation vector of all data points.
protected double[]	e1Costs The cost computed for each cluster.
protected double[]	i1Costs The cost computed for each cluster.
protected List<DoubleVector>	matrix The Matrix holding the data points.
protected double[]	simToComplete The distance of each centroid to completeCentroid.

Constructor Summary

Constructors

Constructor and Description
HybridBaseFunction()

Method Summary

Methods

Modifier and Type	Method and Description
int[]	assignments() Returns the cluster assignment indices for each data point in the original matrix passed to {@link #setup(Matrix, int[] int) setup).
DoubleVector[]	centroids() Returns the final set of centroids computed for the dataset passed to setup.
int[]	clusterSizes() Returns the number of data points assigned to each cluster.
protected abstract BaseFunction	getExternalFunction() Returns the external CriterionFunction.
protected abstract BaseFunction	getInternalFunction() Returns the internal CriterionFunction.
double	score() Returns the score computed by this CriterionFunction.
void	setup(Matrix m, int[] initialAssignments, int numClusters) Creates the cluster centroids and any other meta data needed by this CriterionFunction.
boolean	update(int currentVectorIndex) Updates the clustering assignment for data point indexed by currentVectorIndex.

Methods inherited from class java.lang.Object

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

Methods inherited from interface edu.ucla.sspace.clustering.criterion.CriterionFunction

isMaximize

Field Detail

matrix

protected List<DoubleVector> matrix

The Matrix holding the data points.

assignments

protected int[] assignments

The set of cluster assignments for each cluster.

centroids

protected DoubleVector[] centroids

The centroids representing each cluster.

clusterSizes

protected int[] clusterSizes

The number of data points found in each cluster.

e1Costs

protected double[] e1Costs

The cost computed for each cluster. This is maintained seperatly so that update can modify only the two relevant clusters being modified.

i1Costs

protected double[] i1Costs

The cost computed for each cluster. This is maintained seperatly so that update can modify only the two relevant clusters being modified.

completeCentroid

protected DoubleVector completeCentroid

The summation vector of all data points.

simToComplete

protected double[] simToComplete

The distance of each centroid to completeCentroid.

Constructor Detail

HybridBaseFunction

public HybridBaseFunction()

Method Detail

setup

public void setup(Matrix m,
         int[] initialAssignments,
         int numClusters)

Creates the cluster centroids and any other meta data needed by this CriterionFunction.

Specified by:

setup in interface CriterionFunction

Parameters:

m - The Matrix holding data points. This will be used as read only.

initialAssignments - The cluster assignments for each data point in m. This is used as read only and discarded.

update

public boolean update(int currentVectorIndex)

Updates the clustering assignment for data point indexed by currentVectorIndex. This returns true if the data point is left in the same cluster and false if it was relocated to another data point.

Specified by:

update in interface CriterionFunction

assignments

public int[] assignments()

Returns the cluster assignment indices for each data point in the original matrix passed to {@link #setup(Matrix, int[] int) setup).

Specified by:

assignments in interface CriterionFunction

centroids

public DoubleVector[] centroids()

Returns the final set of centroids computed for the dataset passed to setup.

Specified by:

centroids in interface CriterionFunction

clusterSizes

public int[] clusterSizes()

Returns the number of data points assigned to each cluster.

Specified by:

clusterSizes in interface CriterionFunction

score

public double score()

Returns the score computed by this CriterionFunction.

Specified by:

score in interface CriterionFunction

getInternalFunction

protected abstract BaseFunction getInternalFunction()

Returns the internal CriterionFunction.

getExternalFunction

protected abstract BaseFunction getExternalFunction()

Returns the external CriterionFunction.