Bagging algorithm for hclust

bagHclust.R

library(reshape2)

#Bagging hierarchical clustering

bagHClust<-function(data, n, k, size, outlier.th) {

clus.bs<-NULL

for (i in 1:n) {

  n<-nrow(data)
  bs.ind<-sample.int(n, size, replace=FALSE)
  hc <- hclust(dist(data[bs.ind,]), "ave")
  ct<-cutree(hc, k)
  add<-data.frame(iter=rep(i, size), ind=names(ct), cluster=ct)
  clus.bs<-rbind(clus.bs, add)
}

#Cartesian products

m<-merge(clus.bs, clus.bs, by=c("iter", "cluster"))
d<-dcast(m, ind.x~ind.y, length)
dm<-d[,-1]/diag(as.matrix(d[,-1]))
dm.rep<-dm
diag(dm.rep)<-0
most<-apply(dm.rep,2,max)
most[most<0.75]

disim<-as.dist(1-dm)
h<-hclust(disim, "ave")

return(h)

}

bhc<-bagHClust(USArrests, 100, 8, 40)
plot(bhc)

hc<-hclust(dist(USArrests), "ave")
plot(hc)