rclust.kmedoids: k-Medoids Clustering Manifold-valued Data

In kyoustat/RiemBaseExt: Extension to 'RiemBase' Package

Description

k-Medoids is a generally applicable clustering algorithm as long as we have concept of dissimilarity. We adopt pam algorithm by cluster package. See pam for more details.

Usage

rclust.kmedoids(input, k = 2, type = c("extrinsic", "intrinsic"))

Arguments

input	a S3 object of riemdata class. See riemfactory for more details.
k	the number of clusters.
type	type of distance, either "intrinsic" or "extrinsic".

Value

an object of class pam. See pam for details.

Examples

## generate 50 points near (0,0,1)  and
##          50 points near (0,0,-1) on Sphere S^2 
ndata = 50
theta = seq(from=-0.99,to=0.99,length.out=ndata)*pi
tmpx  = cos(theta) + rnorm(ndata,sd=0.1)
tmpy  = sin(theta) + rnorm(ndata,sd=0.1)

## wrap it as 'riemdata' class
data  = list()
for (i in 1:ndata){
  tgt = c(tmpx[i],tmpy[i],1)
  data[[i]] = tgt/sqrt(sum(tgt^2)) # project onto Sphere
}
for (i in 1:ndata){
  tgt = c(tmpx[i],tmpy[i],-1)
  data[[i+ndata]] = tgt/sqrt(sum(tgt^2)) # project onto Sphere
}
data = RiemBase::riemfactory(data, name="sphere")

## compare k-medoids with intrinsic distance for different k's
k2 <- rclust.kmedoids(data, k=2, type="intrinsic")
k3 <- rclust.kmedoids(data, k=3, type="intrinsic")
k4 <- rclust.kmedoids(data, k=4, type="intrinsic")

## let's visualize the results via MDS in R^2
pdist = stats::as.dist(RiemBase::rbase.pdist(data))
dat2d = stats::cmdscale(pdist, k=2)

opar <- par(no.readonly=TRUE)
par(mfrow=c(1,3))
plot(dat2d[,1], dat2d[,2], col=k2$clustering, pch=19, main="k=2")
plot(dat2d[,1], dat2d[,2], col=k3$clustering, pch=19, main="k=3")
plot(dat2d[,1], dat2d[,2], col=k4$clustering, pch=19, main="k=4")
par(opar)

kyoustat/RiemBaseExt documentation built on March 28, 2020, 2:08 a.m.