R package available for "Adjusted Mutual Information"? [closed] - r

Question

Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 6 years ago.
Improve this question
I am searching for a package to calculate the adjusted mutual information between two clusterings. I have only found some python code via google. Is there any built-in R package or function that can be clustering the data via mutual information?

Answer 1

Here is a link
https://github.com/defleury/adjusted_mutual_information
It says "contains code for the fast & parallelized calculation of Adjusted Mutual Information (AMI), Normalized Mutual Information (NMI) and Adjusted Rand Index (ARI) between clusterings in R."
For small clusters, here are 3 functions. Function f_rez() takes as input 2 vectors in which numbers says the partition of that element and returns AMI. It takes about 30 s for 3 pairs of clusters of length N = 11.117, on a dual core non-parallel.
f_nij <- function(v1,v2,l1,l2){ #contingency table n(i,j)=t(i,j)
m <- matrix(0,l1,l2)
for (i in 1:length(v1)){
m[v1[i],v2[i]] <- m[v1[i],v2[i]] +1
}
m
}
f_emi <- function(s1,s2,l1,l2,n){ #expected mutual information
s_emi <- 0
for(i in 1:l1){
for (j in 1:l2){
min_nij <- max(1,s1[i]+s2[j]-n)
max_nij <- min(s1[i],s2[j])
n.ij <- seq(min_nij, max_nij) #sequence of consecutive numbers
t1<- (n.ij / n) * log((n.ij * n) / (s1[i]*s2[j]))
t2 <- exp(lfactorial(s1[i]) + lfactorial(s2[j]) + lfactorial(n - s1[i]) + lfactorial(n - s2[j]) - lfactorial(n) - lfactorial(n.ij) - lfactorial(s1[i] - n.ij) - lfactorial(s2[j] - n.ij) - lfactorial(n - s1[i] - s2[j] + n.ij))
emi <- sum(t1*t2)
s_emi <- s_emi + emi
}
}
return(s_emi)
}
f_rez <- function(v1,v2){
library(infotheo)
s1 <- tabulate(v1);
s2 <- tabulate(v2);
l1 <- length(s1)
l2 <- length(s2)
N <- length(v1)
tij <- f_nij(v1,v2,l1,l2) #contingency table n(i,j)=t(i,j). this would be equivalent with table(v1,v2)
mi <- mutinformation(v1,v2) #function for Mutual Information from package infotheo
h1 <- -sum(s1*log(s1/N))/N
h2 <- -sum(s2*log(s2/N))/N
nmi <- mi/max(h1,h2) # NMI Normalized MI
emi <- f_emi(s1,s2,l1,l2,N) # EMI Expected MI
ami <- (mi-emi)/max(h1,h2) #AMI Adjusted MI
return(c(nmi,ami))
}

Answer 2

I found here a matlab code for Adjusted Mutual Information(AMI). And according to this thread on stackoverflow, it is possible to translate .m file into .r file, though this isn't trivial, at least it is a way to get a function of AMI for R.
The R package "CLUE" provided Normalized Mutual Information(NMI), a "less-good" version of AMI. According to this paper "A Novel Approach for Automatic Number of Clusters Detection in Microarray Data based on Consensus Clustering" by Nguyen Xuan Vinh and Julien Epps, it seems that the Adjusted Rand Index(ARI) is a good substitution of AMI.
Luckily for ARI, there are a few R packages that implemented this funciton, such as, RRand() funciton in Phyclust package, RandIndex() function in flexclust package, adjustedRandIndex() function in mclust package, and a.rand.index2() in clustergas package.
Hope this helps.