Package elki.clustering.silhouette

Class FastMSC.Instance

java.lang.Object

elki.clustering.silhouette.FastMSC.Instance

Direct Known Subclasses:

FasterMSC.Instance

Enclosing class:

FastMSC<O>

protected class FastMSC.Instance
extends java.lang.Object

FastMSC clustering instance for a particular data set.

Author:

Erich Schubert

Field Summary

Fields

Modifier and Type	Field	Description
protected WritableDataStore<FastMSC.Record>	assignment	Distances and nearest medoids.
protected DistanceQuery<?>	distQ	Distance function to use.
protected DBIDs	ids	Ids to process.
protected WritableIntegerDataStore	output	Output cluster mapping.

Constructor Summary

Constructors

Constructor	Description
Instance(DistanceQuery<?> distQ, DBIDs ids, WritableIntegerDataStore assignment)	Constructor.

Method Summary

Modifier and Type	Method	Description
protected double	assignToNearestCluster(ArrayDBIDs means)	Assign each object to the nearest cluster.
protected double	doSwap(ArrayDBIDs medoids, int b, DBIDRef j)	Assign each object to the nearest cluster when replacing one medoid.
protected double	findBestSwap(DBIDRef j, double[] ploss)	Compute the loss change when choosing j as new medoid.
protected double	run(ArrayModifiableDBIDs medoids, int maxiter)	Run the FastMSC optimization phase.
DoubleDataStore	silhouetteScores()	Get the silhouette scores per point (must be run() first)
protected void	updateRemovalLoss(double[] losses)	Update the share removal loss data
protected void	updateThirdNearest(DBIDRef j, FastMSC.Record rec, int m, double bestd, DBIDArrayIter miter)	Update the third nearest in the record.

Methods inherited from class java.lang.Object

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

Field Detail

ids

protected DBIDs ids

Ids to process.

distQ

protected DistanceQuery<?> distQ

Distance function to use.

assignment

protected WritableDataStore<FastMSC.Record> assignment

Distances and nearest medoids.

output

protected WritableIntegerDataStore output

Output cluster mapping.

Constructor Detail

Instance

public Instance(DistanceQuery<?> distQ,
                DBIDs ids,
                WritableIntegerDataStore assignment)

Constructor.

Parameters:

distQ - Distance query

ids - IDs to process

assignment - Cluster assignment

Method Detail

run

protected double run(ArrayModifiableDBIDs medoids,
                     int maxiter)

Run the FastMSC optimization phase.

Parameters:

medoids - Initial medoids list

maxiter - Maximum number of iterations

Returns:

final medoid Silhouette

assignToNearestCluster

protected double assignToNearestCluster(ArrayDBIDs means)

Assign each object to the nearest cluster.

Parameters:

means - Cluster medoids

Returns:

loss

findBestSwap

protected double findBestSwap(DBIDRef j,
                              double[] ploss)

Compute the loss change when choosing j as new medoid.

Parameters:

j - New medoid

ploss - Loss array

Returns:

Shared loss term

doSwap

protected double doSwap(ArrayDBIDs medoids,
                        int b,
                        DBIDRef j)

Assign each object to the nearest cluster when replacing one medoid.

Parameters:

medoids - Cluster medoids

b - Medoid position index

j - New medoid

Returns:

medoid silhouette

updateThirdNearest

protected void updateThirdNearest(DBIDRef j,
                                  FastMSC.Record rec,
                                  int m,
                                  double bestd,
                                  DBIDArrayIter miter)

Update the third nearest in the record.

Parameters:

j - Current object

rec - Current record

m - Medoid id replaced

bestd - distance to medoid

miter - Medoid iterator

updateRemovalLoss

protected void updateRemovalLoss(double[] losses)

Update the share removal loss data

Parameters:

losses - Removal loss storage

silhouetteScores

public DoubleDataStore silhouetteScores()

Get the silhouette scores per point (must be run() first)

Returns:

Silhouette scores