jsat.clustering.evaluation

Class AdjustedRandIndex

java.lang.Object

jsat.clustering.evaluation.AdjustedRandIndex

All Implemented Interfaces:

ClusterEvaluation

public class AdjustedRandIndex
extends Object
implements ClusterEvaluation

Adjusted Rand Index (ARI) is a measure to evaluate a cluster based on the true class labels for the data set. The ARI normally returns a value in [-1, 1], where 0 indicates the clustering appears random, and 1 indicate the clusters perfectly match the class labels, and negative values indicate a clustering that is worse than random. To match the ClusterEvaluation interface, the value returned by evaluate will be 1.0-Adjusted Rand Index so the best value becomes 0.0 and the worse value becomes 2.0.
NOTE: Because the ARI needs to know the true class labels, only evaluate(int[], jsat.DataSet) will work, since it provides the data set as an argument. The dataset given must be an instance of ClassificationDataSet

Author:

Edward Raff

Constructor Summary

Constructors

Constructor and Description
AdjustedRandIndex()

Method Summary

Modifier and Type	Method and Description
ClusterEvaluation	clone()
double	evaluate(int[] designations, DataSet dataSet) Evaluates the clustering of the given clustering.
double	evaluate(List<List<DataPoint>> dataSets) Evaluates the clustering of the given set of clusters.

Methods inherited from class java.lang.Object

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

Constructor Detail

AdjustedRandIndex

public AdjustedRandIndex()

Method Detail

evaluate

public double evaluate(int[] designations,
                       DataSet dataSet)

Description copied from interface: ClusterEvaluation

Evaluates the clustering of the given clustering.

Specified by:

evaluate in interface ClusterEvaluation

Parameters:

designations - the array that stores the cluster assignments for each data point in the data set

dataSet - the data set that contains all data points

Returns:

a value in [0, Inf) that indicates the quality of the clustering.

evaluate

public double evaluate(List<List<DataPoint>> dataSets)

Description copied from interface: ClusterEvaluation

Evaluates the clustering of the given set of clusters.

Specified by:

evaluate in interface ClusterEvaluation

Parameters:

dataSets - a list of lists, where the size of the first index indicates the the number of clusters, and the list at each index is the data points that make up each cluster.

Returns:

a value in [0, Inf) that indicates the quality of the clustering.

clone

public ClusterEvaluation clone()

Specified by:

clone in interface ClusterEvaluation

Overrides:

clone in class Object