Package elki.clustering.kmeans

Class HartiganWongKMeans.Instance

java.lang.Object

elki.clustering.kmeans.AbstractKMeans.Instance

elki.clustering.kmeans.HartiganWongKMeans.Instance

Enclosing class:

HartiganWongKMeans<V extends NumberVector>

protected static class HartiganWongKMeans.Instance
extends AbstractKMeans.Instance

Instance for a particular data set.

Author:

Minh Nhat Nguyen

Field Summary

Fields

Modifier and Type	Field	Description
(package private) double[]	an1	Weights for adding/removing points from a cluster.
(package private) double[]	an2	Weights for adding/removing points from a cluster.
(package private) boolean[]	itran	Updated in quick-transfer
(package private) int[]	live	Live set indicators
(package private) int[]	ncp	In Optimal-transfer-stage, NCP(L) indicates the step at which cluster L is last updated.
private int	optries	Number of attempts to make an optimal transfer.
(package private) WritableDoubleDataStore	r1s	The value [NC(L1) * D(I,L1)^2] / [NC(L1) -1] will be remembered and will remain the same for Point I until cluster L1 is updated.
(package private) WritableIntegerDataStore	secondary	Second nearest cluster.

Fields inherited from class elki.clustering.kmeans.AbstractKMeans.Instance

assignment, clusters, diststat, isSquared, k, key, means, relation, varsum

Constructor Summary

Constructors

Constructor	Description
Instance(Relation<? extends NumberVector> relation, NumberVectorDistance<?> df, double[][] means)	Constructor.

Method Summary

Modifier and Type	Method	Description
private double	cacheR1(DBIDIter it, NumberVector vec, int l1)	Compute and cache the R1 value.
protected Logging	getLogger()	Get the class logger.
private int	initialAssignToNearestCluster()	Step 1: For each point I, find its two closest centers, IC1(I) and IC2(I).
private void	initialize()	Initialize AN1, AN2, ITRAN, NCP
protected int	iterate(int iteration)	Main loop function.
private int	optimalTransfer()	In this stage, there is only one pass through the data.
private int	quickTransfer()	Step 6: the quick transfer (QTRAN) phase.
private void	transfer(DBIDRef it, NumberVector vec, int l1, int l2)	Transfer a point from one cluster to another.

Methods inherited from class elki.clustering.kmeans.AbstractKMeans.Instance

assignToNearestCluster, buildResult, buildResult, computeSquaredSeparation, copyMeans, distance, distance, distance, initialSeperation, meansFromSums, movedDistance, recomputeSeperation, recomputeVariance, run, sqrtdistance, sqrtdistance, sqrtdistance

Methods inherited from class java.lang.Object

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

Field Detail

secondary

WritableIntegerDataStore secondary

Second nearest cluster. IC2(I).

r1s

WritableDoubleDataStore r1s

The value [NC(L1) * D(I,L1)^2] / [NC(L1) -1] will be remembered and will remain the same for Point I until cluster L1 is updated.

ncp

int[] ncp

In Optimal-transfer-stage, NCP(L) indicates the step at which cluster L is last updated. In the Quick-transfer-stage, NCP(L) is equal to the step at which cluster L is last updated plus M (number of points).

live

int[] live

Live set indicators

itran

boolean[] itran

Updated in quick-transfer

an1

double[] an1

Weights for adding/removing points from a cluster.

an2

double[] an2

Weights for adding/removing points from a cluster.

optries

private int optries

Number of attempts to make an optimal transfer.

Constructor Detail

Instance

public Instance(Relation<? extends NumberVector> relation,
                NumberVectorDistance<?> df,
                double[][] means)

Constructor.

Parameters:

relation - Data relation

df - Distance function

means - Initial means

Method Detail

iterate

protected int iterate(int iteration)

Description copied from class: AbstractKMeans.Instance

Main loop function.

Specified by:

iterate in class AbstractKMeans.Instance

Parameters:

iteration - Iteration number (beginning at 1)

Returns:

Number of reassigned points

initialAssignToNearestCluster

private int initialAssignToNearestCluster()

Step 1: For each point I, find its two closest centers, IC1(I) and IC2(I). Assign it to IC1(I).

Returns:

Number of reassigned points

initialize

private void initialize()

Initialize AN1, AN2, ITRAN, NCP

optimalTransfer

private int optimalTransfer()

In this stage, there is only one pass through the data. Each point is reallocated, if necessary, to the cluster that will induce the maximum reduction in wcss.

Returns:

Number of reassigned objects

quickTransfer

private int quickTransfer()

Step 6: the quick transfer (QTRAN) phase. Each point is tested in turn to see if it should be reallocated to the cluster which it is most likely to be transferred to IC2(I) from its present cluster IC1(I). Loop through the data until no further change is to take place.

cacheR1

private double cacheR1(DBIDIter it,
                       NumberVector vec,
                       int l1)

Compute and cache the R1 value.

Parameters:

it - Point id

vec - Data vector

l1 - Center id

Returns:

R1

transfer

private void transfer(DBIDRef it,
                      NumberVector vec,
                      int l1,
                      int l2)

Transfer a point from one cluster to another.

Parameters:

it - Point id

vec - Vector

l1 - First cluster

l2 - Second cluster

getLogger

protected Logging getLogger()

Description copied from class: AbstractKMeans.Instance

Get the class logger.

Specified by:

getLogger in class AbstractKMeans.Instance

Returns:

Logger