ev.mi: Computes the Mutual Information Measure

In lejean2000/CLAV: An R Package For Cluster Validation

Description

Mutual information is a Kullback-Leibler type of measure for directed divergence. It is a positive number varying from 0 to log2(C), where C is the number of clusters. Higher values indicate less divergence, hence better clustering results. Please note that MI is a not suitable for k-means clustering validation in highly imbalanced datasets because it cannot capture the inbalance. MI is equivalent to the entropy measure.

Usage

ev.mi(x, y)

Arguments

x	A vector with cluster assignments.
y	A vector with cluster assignments.

Value

A positive number.

Examples

d<-vegan::vegdist (iris[,3:4], method = "euclidean")
c<-cluster::pam (d, 3, diss = TRUE)
ev.mi (c$clustering, unclass(iris$Species))

lejean2000/CLAV documentation built on Feb. 5, 2020, 8:55 p.m.