Normalized Compression Distance

ncd.R

# Normalized Compression Distance
# http://en.wikipedia.org/wiki/Normalized_Compression_Distance
# Quick and dirty implementation 


ncd <- function(file1, file2) {
  both <- ""
  file1 <- toString(file1)
  file2 <- toString(file2)
  comp1 <- memCompress(file1, type=c(c("bzip2")))
  comp2 <- memCompress(file2, type=c(c("bzip2")))
  lencomp1 = length(comp1)
  lencomp2 = length(comp2)
  both <- paste(both, file1, file2, sep="")
  compboth <- memCompress(both, type=c(c("bzip2")))
  lencompboth = length(compboth)
  result <- (lencompboth - min(lencomp1, lencomp2))/max(lencomp1,lencomp2)
  result
}

library(tm)
romany <- Corpus(DirSource("romany"), readerControl = list(language = "cz"))


# matrix
mydata.vectors <- character(0)
for(i in 1:10) {
    for(x in 1:10) {
        ncdromany <- ncd(romany[[i]],romany[[x]])
        mydata.vectors <- c(ncdromany,mydata.vectors)
    }
}

a <- matrix(rev(mydata.vectors), nrow=10)
names.vectors <- character(0)
for(x in 1:10) { names.vectors <- c(ID(romany[[x]]),names.vectors) }
dimnames(a) <- list(rev(names.vectors), rev(names.vectors))


d <- dist(a, method = "euclidean")
fit <- hclust(d, method="ward") 
plot(fit) # display dendogram