R help - Oct 2005 - Dendrogram for many cases

Dear All,

I have a cluster object based on a 
dissimilarity matrix from about 1,100 
cases and wish to know whether anyone 
can think  of any tips to display some 
form of graphical output which would 
give some sense of the similarity 
between the cases.

A standard form of dendrogram would be 
fine, but with so many cases the 
dendrogram on the standard devices 
(R-2.20 on NT4) is very compact in the 
x-dimension. I wonder whether there is 
any way that the dendrogram can be 
subdivided into discrete pieces?

Failing that, is there any other means 
of graphically representing the 
dissimilarity matrix. I am only 
interested in the low order 
dissimilarity rather than high order 
structure between these cases.

A further constraint is that the NT4 box 
is well bolted down in that it has no 
means by which data can be transfered 
to, or from it.

Cheers,

David.


#########################################
Dr. David Lucy
School of Mathematics
JCMB
King's Buildings
Edinburgh University
Edinburgh
EH9 3JZ

tel:    0131 650 5086
e-mail: dlucy at maths.ed.ac.uk
web:    http://www.maths.ed.ac.uk/~dlucy/
#########################################

David,

Sounds as if you're looking for cut.dendrogram().

My solution (with c. 250 cases) has been to color the terminals so 
patterns can be seen even when there are too many terminals to label. I 
don't think you can do that easily with plot.hclust() or plot.dendrogram() 
so I posted a hacked version of plot.dendrogram() to R-devel last week. 
Subsequently I was also pointed to a package that does a better job of it 
than my hacked function. It's called A2R, and is not on CRAN but can be 
downloaded from: http://addictedtor.free.fr/packages.

Walton
> Date: Wed, 26 Oct 2005 11:23:26 +0100
> From: David Lucy <dlucy at maths.ed.ac.uk>
> Subject: [R] Dendrogram for many cases
> To: r-help at stat.math.ethz.ch
> Message-ID: <435F591E.9010902 at maths.ed.ac.uk>
> Content-Type: text/plain; charset=us-ascii; format=flowed
> 
> Dear All,
> 
> I have a cluster object based on a 
> dissimilarity matrix from about 1,100 
> cases and wish to know whether anyone 
> can think  of any tips to display some 
> form of graphical output which would 
> give some sense of the similarity 
> between the cases.
> 
> A standard form of dendrogram would be 
> fine, but with so many cases the 
> dendrogram on the standard devices 
> (R-2.20 on NT4) is very compact in the 
> x-dimension. I wonder whether there is 
> any way that the dendrogram can be 
> subdivided into discrete pieces?
> 
> Failing that, is there any other means 
> of graphically representing the 
> dissimilarity matrix. I am only 
> interested in the low order 
> dissimilarity rather than high order 
> structure between these cases.
> 
> A further constraint is that the NT4 box 
> is well bolted down in that it has no 
> means by which data can be transfered 
> to, or from it.
> 
> Cheers,
> 
> David.
######################## .signature ########################
# Walton A. Green                    walton.green at yale.edu #   
# 139 Caulkinstown Road     P. O. Box 208109, Yale Station #
# Sharon, Connecticut 06069   New Haven, Connecticut 06520 #
# (860) 364-5100                            (203) 640-8122 #
#################### 60 characters wide ####################