Package org.nlpub.watset.graph

Class ChineseWhispers<V,E>

java.lang.Object

org.nlpub.watset.graph.ChineseWhispers<V,E>

Type Parameters:

V - the type of nodes in the graph

E - the type of edges in the graph

All Implemented Interfaces:

Clustering<V>

public class ChineseWhispers<V,E>
extends Object
implements Clustering<V>

Implementation of the Chinese Whispers algorithm.

See Also:

Biemann (TextGraphs-1)

Field Summary

Fields

Modifier and Type	Field	Description
protected org.jgrapht.Graph<V,E>	graph	The graph.
protected int	iterations	The number of iterations.
static int	ITERATIONS	The default number of Chinese Whispers iterations.
protected Map<V,Integer>	labels	The mapping of nodes to labels.
protected Random	random	The random number generator.
protected int	steps	The number of actual algorithm iterations.
protected NodeWeighting<V,E>	weighting	The node weighting approach.

Constructor Summary

Constructors

Constructor	Description
ChineseWhispers(org.jgrapht.Graph<V,E> graph, NodeWeighting<V,E> weighting)	Create an instance of the Chinese Whispers algorithm.
ChineseWhispers(org.jgrapht.Graph<V,E> graph, NodeWeighting<V,E> weighting, int iterations, Random random)	Create an instance of the Chinese Whispers algorithm.

Method Summary

Modifier and Type	Method	Description
void	fit()	Run the algorithm to induce the parameters of the clusters.
Collection<Collection<V>>	getClusters()	Return a collection of clusters, each cluster is a collection of objects.
int	getIterations()	Return the number of iterations specified in the constructor
int	getSteps()	Return the number of iterations actually performed during fit().
static <V,E> Function<org.jgrapht.Graph<V,E>,Clustering<V>>	provider(NodeWeighting<V,E> weighting)	A factory function that sets up the algorithm for the given graph.
static <V,E> Function<org.jgrapht.Graph<V,E>,Clustering<V>>	provider(NodeWeighting<V,E> weighting, int iterations, Random random)	A factory function that sets up the algorithm for the given graph.
protected Map<Integer,Double>	score(org.jgrapht.Graph<V,E> graph, Map<V,Integer> labels, NodeWeighting<V,E> weighting, V node)	Score the label weights in the given neighborhood graph, which is a subgraph of graph.
protected int	step(List<V> nodes)	Perform one iteration of the algorithm.

Methods inherited from class java.lang.Object

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

Field Detail

ITERATIONS

public static final int ITERATIONS

The default number of Chinese Whispers iterations.

See Also:

Constant Field Values

graph

protected final org.jgrapht.Graph<V,E> graph

The graph.

weighting

protected final NodeWeighting<V,E> weighting

The node weighting approach.

iterations

protected final int iterations

The number of iterations.

random

protected final Random random

The random number generator.

labels

protected Map<V,Integer> labels

The mapping of nodes to labels.

steps

protected int steps

The number of actual algorithm iterations.

Constructor Detail

ChineseWhispers

public ChineseWhispers(org.jgrapht.Graph<V,E> graph,
                       NodeWeighting<V,E> weighting,
                       int iterations,
                       Random random)

Create an instance of the Chinese Whispers algorithm.

Parameters:

graph - the graph

weighting - the node weighting approach

iterations - the number of iterations

random - the random number generator

ChineseWhispers

public ChineseWhispers(org.jgrapht.Graph<V,E> graph,
                       NodeWeighting<V,E> weighting)

Create an instance of the Chinese Whispers algorithm.

Parameters:

graph - the graph

weighting - the node weighting approach

Method Detail

provider

public static <V,E> Function<org.jgrapht.Graph<V,E>,Clustering<V>> provider(NodeWeighting<V,E> weighting)

A factory function that sets up the algorithm for the given graph.

Type Parameters:

V - the type of nodes in the graph

E - the type of edges in the graph

Parameters:

weighting - the node weighting approach

Returns:

a factory function that sets up the algorithm for the given graph

provider

public static <V,E> Function<org.jgrapht.Graph<V,E>,Clustering<V>> provider(NodeWeighting<V,E> weighting,
                                                                                              int iterations,
                                                                                              Random random)

A factory function that sets up the algorithm for the given graph.

Type Parameters:

V - the type of nodes in the graph

E - the type of edges in the graph

Parameters:

weighting - the node weighting approach

iterations - the number of iterations

random - the random number generator

Returns:

a factory function that sets up the algorithm for the given graph

fit

public void fit()

Description copied from interface: Clustering

Run the algorithm to induce the parameters of the clusters.

Specified by:

fit in interface Clustering<V>

step

protected int step(List<V> nodes)

Perform one iteration of the algorithm.

Parameters:

nodes - the list of nodes

Returns:

whether any label changed or not

getClusters

public Collection<Collection<V>> getClusters()

Description copied from interface: Clustering

Return a collection of clusters, each cluster is a collection of objects. Usually this method is called after the Clustering.fit() method.

Specified by:

getClusters in interface Clustering<V>

Returns:

clusters

score

protected Map<Integer,Double> score(org.jgrapht.Graph<V,E> graph,
                                          Map<V,Integer> labels,
                                          NodeWeighting<V,E> weighting,
                                          V node)

Score the label weights in the given neighborhood graph, which is a subgraph of graph. This method sums the node weights corresponding to each label.

Parameters:

graph - the neighborhood graph

labels - the map of graph nodes to their labels

weighting - the node weighting approach

node - the target node

Returns:

a mapping of labels to sums of their weights

getIterations

public int getIterations()

Return the number of iterations specified in the constructor

Returns:

the number of iterations

See Also:

ChineseWhispers(Graph, NodeWeighting, int, Random)

getSteps

public int getSteps()

Return the number of iterations actually performed during fit(). Should be no larger than the value of getIterations().

Returns:

the number of iterations