variationInformation: Variation of Information
In laderast/Consense: Consense: a package for evaluating multiple clusterings

Description Usage Arguments Details Value Author(s) References See Also Examples

Given two clusterings, this function calculates the Variation of Information between them. Variation of Information can be thought of as what information content is lost and what information content is gained by choosing one clustering over the other clustering.

1	variationInformation(clustering1, clustering2)

`clustering1`	A clustering as defined by the cluster output functions of this package. See `outputcluster` for more details.
`clustering2`	same as clustering 1.

The Variation of Information (from Meila) is a quantity based on the entropy H(C) of a clustering C, the entropy H(C') of the second clustering C', and the mutual information I(C,C') between these two clusterings. See Meila for more information on how entropy and mutual information are calculated.

Variation of Information can be thought of as what information content is lost and what information content is gained by choosing one clustering over the other clustering. Unlike more conventional methods of counting cluster membership (such as Jaccard Index), the value can be interpreted when there are different numbers of clusters in the two clusterings.

The Variation of information is calculated by the following equation: H(C) + H(C') - 2I(C, C')

variation

Variation of Information. A value that ranges from 0-1. The more the clusterings are in agreement, the closer this value is to 0.

Ted Laderas (laderast@ohsu.edu)

Meila, M. Comparing Clusterings. Technical Report 418. 2002, University of Washington Statistics Department: Seattle.

jaccard

    data(choresults)
    clusts <- choresults$clusters
    #calculate Variation of Information from clusterings
    variationInformation(as.data.frame(clusts[[c("UPGMACOR")]]), 
    as.data.frame(clusts[[c("UPGMAEUC")]]))