spectralPAM {sClust}

Spectral-PAM clustering

Description

The function, for a given similarity matrix, will separate the data using a spectral space.It is based on the Jordan and Weiss algorithm. This version uses K-medoid to split the clusters.

Usage

spectralPAM(W, K, flagDiagZero = FALSE, verbose = FALSE)

Arguments

W

Gram Similarity Matrix.

K

number of cluster to obtain.

flagDiagZero

if True, Put zero on the similarity matrix W.

verbose

To output the verbose in the terminal.

Value

returns a list containing the following elements:

• cluster: a vector containing the cluster

• eigenVect: a vector containing the eigenvectors

• eigenVal: a vector containing the eigenvalues

Author(s)

Emilie Poisson Caillault and Erwan Vincent

Examples

### Example 1: 2 disks of the same size
n<-100 ; r1<-1
x<-(runif(n)-0.5)*2;
y<-(runif(n)-0.5)*2
keep1<-which((x*2+y*2)<(r1*2))
disk1<-data.frame(x+3*r1,y)[keep1,]
disk2 <-data.frame(x-3*r1,y)[keep1,]
sameTwoDisks <- rbind(disk1,disk2)
W <- compute.similarity.ZP(scale(sameTwoDisks))
res <- spectralPAM(W,K=2,flagDiagZero=TRUE,verbose=TRUE)
plot(sameTwoDisks, col = res$cluster)
plot(res$eigenVect[,1:2], col = res$cluster, main="spectral space",
     xlim=c(-1,1),ylim=c(-1,1)); points(0,0,pch='+');
plot(res$eigenVal, main="Laplacian eigenvalues",pch='+'); 
abline(h=1,lty="dashed",col="red")

### Example 2: Speed and Stopping Distances of Cars
W <- compute.similarity.ZP(scale(iris[-5]))
res <- spectralPAM(W,K=2,flagDiagZero=TRUE,verbose=TRUE)
plot(iris, col = res$cluster)
plot(res$eigenVect[,1:2], col = res$cluster, main="spectral space",
     xlim=c(-1,1),ylim=c(-1,1)); points(0,0,pch='+');
plot(res$eigenVal, main="Laplacian eigenvalues",pch='+'); 
abline(h=1,lty="dashed",col="red")