R help - Apr 2003 - plot(pam.object) error with R-1.7.0 on Red-Hat 8.0 i686

I don't know if there is some fault in compiling or a bug of the new 
R-1.7.0 version:

cl.pam.2 <- pam(as.dist(1-cor(mel.data)),2)
plot(cl.pam.2)

perform a right partitioning and silhouette plot on the old R-1.6.2
instead 
"Error in clusplot.default(x$diss,...... ;
        x is not numeric"
is the output on the new R-1.7.0.
 Same platform: RH8.0 i686. 
Some suggestions?

A.S.

----------------------------

Alessandro Semeria 
Models and Simulations Laboratory
The Environment Research Center - Montecatini (Edison Group), 
Via Ciro Menotti 48,
48023 Marina di Ravenna (RA), Italy
Tel. +39 544 536811
Fax. +39 544 538663
E-mail: asemeria at cramont.it

Hello Martin!
Here the script to create mel.data from data set freely 
available attached to"Molecular Classification 
of Cutaneous Malignant Melanoma
        by Gene Expression Profiling" (Bittner et al., 2000):
(melanoma.csv is an ASCII comma separeted containing data and 
with first raw= conditions names)

library(mva)
cond<- scan(file="melanoma.csv",what=character(38),
sep=",",nlines=1)
mel<-
matrix(scan("melanoma.csv",sep=",",skip=1),ncol=38,byrow=T)
dim(mel)

mel.bittner <- mel[1:3613,1:31]
cond.bittner <- cond[1:31]
mel.bittner[mel.bittner<0.02] <- 0.02
mel.bittner[mel.bittner>50] <- 50
mel.bittner <- log2(mel.bittner)

mel.bittner.median <- apply(mel.bittner,2,median)
mel.bittner.mean <- apply(mel.bittner,2,mean)
mel.bittner.sd <- apply(mel.bittner,2,sd)
mel.data <- sweep(data.matrix(mel.bittner),2,mel.bittner.median)

Thanks!
A.S.

----------------------------

Alessandro Semeria 
Models and Simulations Laboratory
The Environment Research Center - Montecatini (Edison Group), 
Via Ciro Menotti 48,
48023 Marina di Ravenna (RA), Italy
Tel. +39 544 536811
Fax. +39 544 538663
E-mail: asemeria at cramont.it

Frome here 
http://www.nature.com/nature/journal/v406/n6795/extref/406536ai3.xls 
I copied cells  from 2F to  8069AQ (only values and string, no formatting) 
of 
CUTANEOUSMELANOMA sheet to a new
file with excell97 on winXP than I saved it as melanoma.csv (as comma 
separeted 
value). Now I'll try to set verbose level option to the maximum to look 
something.
Thanke more!
A.S.

----------------------------

Alessandro Semeria 
Models and Simulations Laboratory
The Environment Research Center - Montecatini (Edison Group), 
Via Ciro Menotti 48,
48023 Marina di Ravenna (RA), Italy
Tel. +39 544 536811
Fax. +39 544 538663
E-mail: asemeria at cramont.it

>>>>> "alessandro" == alessandro semeria
<alessandro.semeria at cramont.it>
>>>>>     on Mon, 28 Apr 2003 18:53:11 +0200 writes:
    alessandro> I don't know if there is some fault in compiling
    alessandro> or a bug of the new R-1.7.0 version:

    > cl.pam.2 <- pam(as.dist(1-cor(mel.data)),2)
    > plot(cl.pam.2)

    alessandro> perform a right partitioning and silhouette plot
    alessandro> on the old R-1.6.2 instead "Error in
    alessandro> clusplot.default(x$diss,...... ; x is not
    alessandro> numeric" is the output on the new R-1.7.0.  Same
    alessandro> platform: RH8.0 i686.  Some suggestions?

yes.  
This is a bug in the cluster package I'm maintaining.
(I'll file a bug report myself).
One workaround is to say

    cl.pam.2 <- pam(as.dist(1-cor(mel.data)), k=2, keep.diss = TRUE)
    ##                                            ^^^^^^^^^^^^^^^^^^
    plot(cl.pam.2)

{this is a workaround and unnecessary and *not* recommended for
 the next release  of the cluster package !}


Note that the `keep.diss' is a new argument to pam() {and agnes() and
others},
which is not anymore always TRUE by default -- for a good reason:

This allows quite a bit larger datasets than previously,
and also improves on speed (for memory allocation!!) in these
cases.

BUT as you found above,  I have overlooked to adapt  plot()
method for pam() results to this case.

PS: I think I was able to entirely reproduce what you did, but
the clustering (of the 31 variables) is pretty "bad":
30 variables in 1st cluster; and variable "M93.007" alone in the
2nd cluster.
If you look at   silhouette(cl.pam.2)   {which is implicitly
used in the above plot() statement} you'll realize that the
silhouette width is really not useful for this case {all widths
are 0}.


Martin Maechler <maechler at stat.math.ethz.ch>
http://stat.ethz.ch/~maechler/
Seminar fuer Statistik, ETH-Zentrum  LEO C16	Leonhardstr. 27
ETH (Federal Inst. Technology)	8092 Zurich	SWITZERLAND
phone: x-41-1-632-3408		fax: ...-1228			<><

Hello Martin!
I modified script with your suggestion 
"cl.pam.2 <- pam(as.dist(1-cor(mel.data)), k=2, keep.diss = TRUE)" 
obtaining the same "bad" cluster as your, 
then I performed pam without setting "keep.diss "
and looking to cl.pam.2$cluster I found
the same "good" parition as in R-1.6.2.
Then I think that your workaround is good to 
plot something but "destroy" the  job of pam!
A.S.

----------------------------

Alessandro Semeria 
Models and Simulations Laboratory
The Environment Research Center - Montecatini (Edison Group), 
Via Ciro Menotti 48,
48023 Marina di Ravenna (RA), Italy
Tel. +39 544 536811
Fax. +39 544 538663
E-mail: asemeria at cramont.it