de.lmu.ifi.dbs.elki.evaluation.clustering.pairsegments

Class Segments

java.lang.Object

de.lmu.ifi.dbs.elki.result.AbstractHierarchicalResult

de.lmu.ifi.dbs.elki.result.BasicResult

de.lmu.ifi.dbs.elki.evaluation.clustering.pairsegments.Segments

All Implemented Interfaces:

HierarchicalResult, Result, Iterable<Segment>

@Reference(title="Evaluation of Clusterings \u2013 Metrics and Visual Support",
           authors="Elke Achtert, Sascha Goldhofer, Hans-Peter Kriegel, Erich Schubert, Arthur Zimek",
           booktitle="Proc. 28th International Conference on Data Engineering (ICDE) 2012",
           url="http://dx.doi.org/10.1109/ICDE.2012.128")
public class Segments
extends BasicResult
implements Iterable<Segment>

Creates segments of two or more clusterings.

Segments are the equally paired database objects of all given (2+) clusterings. Given a contingency table, an object Segment represents the table's cells where an intersection of classes and labels are given. Pair Segments are created by converting an object Segment into its pair representation. Converting all object Segments into pair Segments results in a larger number of pair Segments, if any fragmentation (no perfect match of clusters) within the contingency table has occurred (multiple cells on one row or column). Thus for ever object Segment exists a corresponding pair Segment. Additionally pair Segments represent pairs that are only in one Clustering which occurs for each split of a clusterings cluster by another clustering. Here, these pair Segments are referenced as fragmented Segments. Within the visualization they describe (at least two) pair Segments that have a corresponding object Segment.

Reference:
Evaluation of Clusterings – Metrics and Visual Support
Elke Achtert, Sascha Goldhofer, Hans-Peter Kriegel, Erich Schubert, Arthur Zimek
In: Proc. 28th International Conference on Data Engineering (ICDE) 2012

Details on the experimental setup can be found at: wiki/Examples/ClusterEvaluation

Field Summary

Fields

Modifier and Type	Field and Description
private int	actualPairs Pairs actually present in the data set
private List<Clustering<?>>	clusterings Clusterings
private int	clusteringsCount Number of clusterings in comparison
private List<List<? extends Cluster<?>>>	clusters Clusters
private static Logging	LOG Class logger
private int[]	numclusters Number of Clusters for each clustering
private TreeMap<Segment,Segment>	segments The actual segments
private int	totalObjects Total number of objects

Constructor Summary

Constructors

Constructor and Description
Segments(List<Clustering<?>> clusterings, HierarchicalResult baseResult) Initialize segments.

Method Summary

Methods

Modifier and Type	Method and Description
String	getClusteringDescription(int clusteringID) Get the description of the nth clustering.
int	getClusterings() Get the number of clusterings
int	getHighestClusterCount() Returns the highest number of Clusters in the clusterings
int	getPairCount(boolean withUnclusteredPairs) Get total number of pairs with or without the unclustered pairs.
List<Segment>	getPairedSegments(Segment unpairedSegment) Return to a given segment with unpaired objects, the corresponding segments that result in an unpaired segment.
int	getTotalClusterCount() Return the sum of all clusters
Iterator<Segment>	iterator()
private void	makeOrUpdateSegment(int[] path, DBIDs ids, int pairsize)
private void	recursivelyFill(List<List<? extends Cluster<?>>> cs)
private void	recursivelyFill(List<List<? extends Cluster<?>>> cs, int depth, SetDBIDs first, SetDBIDs second, int[] path, boolean objectsegment)
int	size() Get the number of segments
Segment	unifySegment(Segment temp)

Methods inherited from class de.lmu.ifi.dbs.elki.result.BasicResult

getLongName, getShortName

Methods inherited from class de.lmu.ifi.dbs.elki.result.AbstractHierarchicalResult

addChildResult, getHierarchy, setHierarchy

Methods inherited from class java.lang.Object

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

Field Detail

LOG

private static final Logging LOG

Class logger

clusterings

private List<Clustering<?>> clusterings

Clusterings

clusters

private List<List<? extends Cluster<?>>> clusters

Clusters

clusteringsCount

private int clusteringsCount

Number of clusterings in comparison

numclusters

private int[] numclusters

Number of Clusters for each clustering

totalObjects

private int totalObjects

Total number of objects

actualPairs

private int actualPairs

Pairs actually present in the data set

segments

private TreeMap<Segment,Segment> segments

The actual segments

Constructor Detail

Segments

public Segments(List<Clustering<?>> clusterings,
        HierarchicalResult baseResult)

Initialize segments. Add DB objects via addObject method.

Parameters:

clusterings - List of clusterings in comparison

baseResult - used to retrieve db objects

Method Detail

recursivelyFill

private void recursivelyFill(List<List<? extends Cluster<?>>> cs)

recursivelyFill

private void recursivelyFill(List<List<? extends Cluster<?>>> cs,
                   int depth,
                   SetDBIDs first,
                   SetDBIDs second,
                   int[] path,
                   boolean objectsegment)

makeOrUpdateSegment

private void makeOrUpdateSegment(int[] path,
                       DBIDs ids,
                       int pairsize)

getClusteringDescription

public String getClusteringDescription(int clusteringID)

Get the description of the nth clustering.

Parameters:

clusteringID - Clustering number

Returns:

long name of clustering

getPairedSegments

public List<Segment> getPairedSegments(Segment unpairedSegment)

Return to a given segment with unpaired objects, the corresponding segments that result in an unpaired segment. So, one cluster of a clustering is split by another clustering in multiple segments, resulting in a segment with unpaired objects, describing the missing pairs between the split cluster / between the segments. Basically we compare only two clusterings at once. If those clusterings do not have the whole cluster in common, we have at least three segments (two cluster), one of them containing the unpaired segment. A segmentID 3-0, describes a cluster 3 in clustering 1 (index 0) and all clusters 3-x in clustering 2. So we search for all segments 3-x (0 being a wildcard).

Parameters:

unpairedSegment -

Returns:

Segments describing the set of objects that result in an unpaired segment

unifySegment

public Segment unifySegment(Segment temp)

Parameters:

temp - Temporary segment to be unified

Returns:

the segmentID given by its string representation

size

public int size()

Get the number of segments

Returns:

Number of segments

getPairCount

public int getPairCount(boolean withUnclusteredPairs)

Get total number of pairs with or without the unclustered pairs.

Parameters:

withUnclusteredPairs - if false, segment with unclustered pairs is removed

Returns:

pair count, with or without unclusted (non-existant) pairs

getClusterings

public int getClusterings()

Get the number of clusterings

Returns:

number of clusterings compared

getTotalClusterCount

public int getTotalClusterCount()

Return the sum of all clusters

Returns:

sum of all cluster counts

getHighestClusterCount

public int getHighestClusterCount()

Returns the highest number of Clusters in the clusterings

Returns:

highest cluster count

iterator

public Iterator<Segment> iterator()

Specified by:

iterator in interface Iterable<Segment>