revisedsil {RSKC}

The revised silhouette

Description

This function returns a revised silhouette plot, cluster centers in weighted squared Euclidean distances and a matrix containing the weighted squared Euclidean distances between cases and each cluster center. Missing values are adjusted.

Usage

revisedsil(d,reRSKC=NULL,CASEofINT=NULL,col1="black",
	CASEofINT2 = NULL, col2="red", print.plot=TRUE, 
	W=NULL,C=NULL,out=NULL)

Arguments

d

A numerical data matrix, N by p, where N is the number of cases and p is the number of features.

reRSKC

A list output from RSKC function.

CASEofINT

Necessary if print.plot=TRUE. A vector of the case indices that appear in the revised silhouette plot. The revised silhouette widths of these indices are colored in col1 if CASEofINT != NULL. The average silhouette of each cluster printed in the plot is computed EXCLUDING these cases.

col1

See CASEofINT.

CASEofINT2

A vector of the case indices that appear in the revised silhouette plot. The indices are colored in col2.

col2

See CASEofINT2

print.plot

If TRUE, the revised silhouette is plotted.

W

Necessary if reRSKC = NULL. A positive real vector of weights of length p.

C

Necessary if reRSKC = NULL. An integer vector of class labels of length N.

out

Necessary if reRSKC = NULL. Vector of the case indices that should be excluded in the calculation of cluster centers. In RSKC, cluster centers are calculated without the cases that have the furthest 100*alpha % Weighted squared Euclidean distances to their closest cluster centers. If one wants to obtain the cluster centers from RSKC output, set out = <RSKCoutput>$oW.

Value

trans.mu

Cluster centers in reduced weighted dimension. See example for more detail.

WdisC

N by ncl matrix, where ncl is the prespecified number of clusters. It contains the weighted distance between each case and all cluster centers. See example for more detail.

sil.order

Silhouette values of each case in the order of the case index.

sil.i

Silhouette values of cases ranked by decreasing order within clusters. The corresponding case index are in obs.i

Author(s)

Yumi Kondo <[e-mail]>

References

Yumi Kondo (2011), Robustificaiton of the sparse K-means clustering algorithm, MSc. Thesis, University of British Columbia http://hdl.handle.net/2429/37093

Examples

# little simulation function 
sim <-
function(mu,f){
   D<-matrix(rnorm(60*f),60,f)
   D[1:20,1:50]<-D[1:20,1:50]+mu
   D[21:40,1:50]<-D[21:40,1:50]-mu  
   return(D)
   }


### output trans.mu ###

p<-200;ncl<-3
# simulate a 60 by p data matrix with 3 classes 
d<-sim(2,p)
# run RSKC
re<-RSKC(d,ncl,L1=2,alpha=0.05)
# cluster centers in weighted squared Euclidean distances by function sil
sil.mu<-revisedsil(d,W=re$weights,C=re$labels,out=re$oW,print.plot=FALSE)$trans.mu
# calculation 
trans.d<-sweep(d[,re$weights!=0],2,sqrt(re$weights[re$weights!=0]),FUN="*") 
class<-re$labels;class[re$oW]<-ncl+1
MEANs<-matrix(NA,ncl,ncol(trans.d))
for ( i in 1 : 3) MEANs[i,]<-colMeans(trans.d[class==i,,drop=FALSE])
sil.mu==MEANs
# coincides 

### output WdisC ###

p<-200;ncl<-3;N<-60
# generate 60 by p data matrix with 3 classes 
d<-sim(2,p)
# run RSKC
re<-RSKC(d,ncl,L1=2,alpha=0.05)
si<-revisedsil(d,W=re$weights,C=re$labels,out=re$oW,print.plot=FALSE)
si.mu<-si$trans.mu
si.wdisc<-si$WdisC
trans.d<-sweep(d[,re$weights!=0],2,sqrt(re$weights[re$weights!=0]),FUN="*") 
WdisC<-matrix(NA,N,ncl)
for ( i in 1 : ncl) WdisC[,i]<-rowSums(scale(trans.d,center=si.mu[i,],scale=FALSE)^2)
# WdisC and si.wdisc coincides