[R] Fwd: MDS problems [ajtee@ajtee.uklinux.net]

[Prof Brian Ripley]

If you `lie', you may find that the routines fail to converge, especially 
sammon -- it is not a good idea.

On 26 Mar 2004, Jari Oksanen wrote:

> On Fri, 2004-03-26 at 15:02, Adam Tee wrote:
> > Hi all,
> > 
> > 
> > I'm trying to perform an MDS of some data that I have.  When I use
> > cmdscale everything is fine and I get some interesting results however,
> > the tends to be low.
> > 
> > What I wnat to do is compare this with the Non-Metric MDS using isoMDS  
> > or sammon. However, when I try using these I get the following message.
> > 
> > Error in isoMDS(x.dist) : zero or negative distance between objects 2  
> > and 4
> > 
> The error message is clear: You have some identical sites so that their
> distance is zero. I think the canonical solution is to remove the
> duplicate cases from the data before calculating the dissimilarities. If
> your data frame is called X:
> 
> Xuniq <- unique(X)
> x.dist <- dist(Xuniq)
> 
> (or, as a one-liner: x.dist <- dist(unique(X))
> 
> A dirty solution is to lie to isoMDS. The following will do:
> 
> isoMDS(x.dist + 1e-5)
> 
> But you better not tell anybody that you did this dirty trick.
> 
> cheers, jari oksanen
> 

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595