dpmeans: DP-means Algorithm for Clustering Euclidean Data

In maotai: Tools for Matrix Algebra, Optimization and Inference

Description

DP-means is a nonparametric clustering method motivated by DP mixture model in that the number of clusters is determined by a parameter λ. The larger the λ value is, the smaller the number of clusters is attained. In addition to the original paper, we added an option to randomly permute an order of updating for each observation's membership as a common heuristic in the literature of cluster analysis.

Usage

dpmeans(
  data,
  lambda = 1,
  maxiter = 1234,
  abstol = 1e-06,
  permute.order = FALSE
)

Arguments

data	an (n\times p) data matrix for each row being an observation.
lambda	a threshold to define a new cluster.
maxiter	maximum number of iterations.
abstol	stopping criterion
permute.order	a logical; TRUE if random order for permutation is used, FALSE otherwise.

Value

a named list containing

cluster

an (n\times ndim) matrix whose rows are embedded observations.

centers

a list containing information for out-of-sample prediction.

References

\insertRef

kulis_revisiting_2012maotai

Examples

## define data matrix of two clusters
x1  = matrix(rnorm(50*3,mean= 2), ncol=3)
x2  = matrix(rnorm(50*3,mean=-2), ncol=3)
X   = rbind(x1,x2)
lab = c(rep(1,50),rep(2,50))

## run dpmeans with several lambda values
solA <- dpmeans(X, lambda= 5)$cluster
solB <- dpmeans(X, lambda=10)$cluster
solC <- dpmeans(X, lambda=20)$cluster

## visualize the results
opar <- par(no.readonly=TRUE)
par(mfrow=c(1,4), pty="s")
plot(X,col=lab,  pch=19, cex=.8, main="True", xlab="x", ylab="y")
plot(X,col=solA, pch=19, cex=.8, main="dpmeans lbd=5", xlab="x", ylab="y")
plot(X,col=solB, pch=19, cex=.8, main="dpmeans lbd=10", xlab="x", ylab="y")
plot(X,col=solC, pch=19, cex=.8, main="dpmeans lbd=20", xlab="x", ylab="y")
par(opar)


## let's find variations by permuting orders of update
## used setting : lambda=20, we will 8 runs
sol8 <- list()
for (i in 1:8){
  sol8[[i]] = dpmeans(X, lambda=20, permute.order=TRUE)$cluster
}

## let's visualize
vpar <- par(no.readonly=TRUE)
par(mfrow=c(2,4), pty="s")
for (i in 1:8){
  pm = paste("permute no.",i,sep="")
  plot(X,col=sol8[[i]], pch=19, cex=.8, main=pm, xlab="x", ylab="y")
}
par(vpar)

maotai documentation built on Oct. 25, 2021, 9:06 a.m.