sp.kmeans: k-Means Clustering for Data on Hypersphere

In kyoustat/RiemSphere: What the package doesas

Description

k-Means Clustering for Data on Hypersphere

Usage

sp.kmeans(
  x,
  k = 2,
  init = c("kmeans", "random"),
  maxiter = 100,
  type = c("intrinsic", "extrinsic")
)

Arguments

x	an (n\times p) row-stacked matrix for \mathbb{S}^{p-1}.
k	the number of clusters to be found.
init	method for setting initial configuration. "random" or "kmeans".
maxiter	maximum number of iterations to be run.
type	type of distance, either "intrinsic" or "extrinsic".

Value

a named list containing

cluster

length-n vector of class membership indices.

Examples

## generate two-cluster data
mymu1 = c(0,0,0,1)  # center of class 1
mymu2 = c(-1,0,0,0) # center of class 2

x1 = rvmf(50, mymu1, kappa=10)
x2 = rvmf(50, mymu2, kappa=10)
xx = rbind(x1,x2)

## apply clustering with different k values
cl2 <- sp.kmeans(xx, k=2)
cl3 <- sp.kmeans(xx, k=3)
cl4 <- sp.kmeans(xx, k=4)

## compute 2-dimensional embedding for visualization
mds2d <- sp.mds(xx, ndim=2)
mdsx  <- mds2d$embed[,1]
mdsy  <- mds2d$embed[,2]

## compare via visualization
opar  <- par(mfrow=c(1,3), pty="s")
plot(mdsx, mdsy, col=cl2$cluster, main="k=2 means", pch=19)
plot(mdsx, mdsy, col=cl3$cluster, main="k=3 means", pch=19)
plot(mdsx, mdsy, col=cl4$cluster, main="k=4 means", pch=19)
par(opar)

kyoustat/RiemSphere documentation built on April 13, 2020, 10:04 a.m.