de.lmu.ifi.dbs.elki.evaluation.paircounting

Class PairCountingFMeasure

java.lang.Object

de.lmu.ifi.dbs.elki.evaluation.paircounting.PairCountingFMeasure

public class PairCountingFMeasure
extends Object

Compare two clustering results using a pair-counting F-Measure. A pair are any two objects that belong to the same cluster. Two clusterings are compared by comparing their pairs; if two clusterings completely agree, they also agree on every pair; even when the clusters and points are ordered differently. An empty clustering will of course have no pairs, the trivial all-in-one clustering of course has n^2 pairs. Therefore neither recall nor precision itself are useful, however their combination -- the F-Measure -- is useful.

Constructor Summary

Constructors

Constructor and Description
PairCountingFMeasure()

Method Summary

Methods

Modifier and Type	Method and Description
static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> double	compareClusterings(R result1, S result2) Compare two clustering results.
static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> double	compareClusterings(R result1, S result2, boolean noiseSpecial, boolean hierarchicalSpecial) Compare two clustering results.
static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> double	compareClusterings(R result1, S result2, double beta) Compare two clustering results.
static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> double	compareClusterings(R result1, S result2, double beta, boolean noiseSpecial, boolean hierarchicalSpecial) Compare two clustering results.
static Triple<Integer,Integer,Integer>	countPairs(PairSortedGeneratorInterface first, PairSortedGeneratorInterface second) Compare two sets of generated pairs.
static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> Triple<Integer,Integer,Integer>	countPairs(R result1, S result2) Compare two sets of generated pairs.
static double	fMeasure(int inBoth, int inFirst, int inSecond, double beta) Computes the F-measure of the given parameters.
static <R extends Clustering<M>,M extends Model> PairSortedGeneratorInterface	getPairGenerator(R clusters, boolean noiseSpecial, boolean hierarchicalSpecial) Get a pair generator for the given Clustering

Methods inherited from class java.lang.Object

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

Constructor Detail

PairCountingFMeasure

public PairCountingFMeasure()

Method Detail

getPairGenerator

public static <R extends Clustering<M>,M extends Model> PairSortedGeneratorInterface getPairGenerator(R clusters,
                                                                                      boolean noiseSpecial,
                                                                                      boolean hierarchicalSpecial)

Get a pair generator for the given Clustering

Type Parameters:

R - Clustering result class

M - Model type

Parameters:

clusters - Clustering result

noiseSpecial - Special handling for "noise clusters"

hierarchicalSpecial - Special handling for hierarchical clusters

Returns:

Sorted pair generator

compareClusterings

public static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> double compareClusterings(R result1,
                                                                                                          S result2,
                                                                                                          double beta,
                                                                                                          boolean noiseSpecial,
                                                                                                          boolean hierarchicalSpecial)

Compare two clustering results.

Type Parameters:

R - Result type

M - Model type

S - Result type

N - Model type

Parameters:

result1 - first result

result2 - second result

beta - Beta value for the F-Measure

noiseSpecial - Noise receives special treatment

hierarchicalSpecial - Special handling for hierarchical clusters

Returns:

Pair counting F-Measure result.

compareClusterings

public static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> double compareClusterings(R result1,
                                                                                                          S result2,
                                                                                                          double beta)

Compare two clustering results.

Type Parameters:

R - Result type

M - Model type

S - Result type

N - Model type

Parameters:

result1 - first result

result2 - second result

beta - Beta value for the F-Measure

Returns:

Pair counting F-Measure result.

compareClusterings

public static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> double compareClusterings(R result1,
                                                                                                          S result2,
                                                                                                          boolean noiseSpecial,
                                                                                                          boolean hierarchicalSpecial)

Compare two clustering results.

Type Parameters:

R - Result type

M - Model type

S - Result type

N - Model type

Parameters:

result1 - first result

result2 - second result

noiseSpecial - Noise receives special treatment

Returns:

Pair counting F-1-Measure result.

compareClusterings

public static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> double compareClusterings(R result1,
                                                                                                          S result2)

Compare two clustering results.

Type Parameters:

R - Result type

M - Model type

S - Result type

N - Model type

Parameters:

result1 - first result

result2 - second result

Returns:

Pair counting F-1-Measure result.

countPairs

public static <R extends Clustering<M>,M extends Model,S extends Clustering<N>,N extends Model> Triple<Integer,Integer,Integer> countPairs(R result1,
                                                                                                                           S result2)

Compare two sets of generated pairs. It determines how many objects of the first set are in both sets, just in the first set or just in the second set.

Type Parameters:

R - Result type

M - Model type

S - Result type

N - Model type

Parameters:

result1 - first result

result2 - second result

Returns:

Returns a Triple that contains the number of objects that are in both sets (FIRST), the number of objects that are just in the first set (SECOND) and the number of object that are just in the second set (THIRD).

countPairs

public static Triple<Integer,Integer,Integer> countPairs(PairSortedGeneratorInterface first,
                                         PairSortedGeneratorInterface second)

Compare two sets of generated pairs. It determines how many objects of the first set are in both sets, just in the first set or just in the second set.

Parameters:

first - first set

second - second set

Returns:

Returns a Triple that contains the number of objects that are in both sets (FIRST), the number of objects that are just in the first set (SECOND) and the number of object that are just in the second set (THIRD).

fMeasure

public static double fMeasure(int inBoth,
              int inFirst,
              int inSecond,
              double beta)

Computes the F-measure of the given parameters.

Returns ((1+beta*beta) * inBoth) / ((1+beta*beta) * inBoth + (beta*beta)*inFirst + inSecond)

Parameters:

inBoth - The number of objects that are in both sets.

inFirst - The number of objects that are in the first set.

inSecond - The number of objects that are in the second set.

beta - The beta values for the f-measure.

Returns:

The F-measure.