Jonck -
Welcome. As you read this list day after day, you will find that
it's mostly about the mechanics of how to get R to "do its thing"
... and sometimes about "what is the name of the function that
does so-and-so". This last is because the reverse-indexing problem
is inherently a difficult one (we can alphabetize the function
names, but how do you "alphabetize" what they do ?) and R users all
recognize that the collective memory is a BETTER, more efficient
way to handle the reverse indexing problem than any algorithm !
(It takes us, as users, less time to read and respond to questions
as they come up, than it would to design and write an automated
system.) (Er, I digress.)
Advice on how to do the data analysis for a specific problem is in
a very different category ... because the person directly facing a
problem has a LOT more knowledge, about that problem and about
constraints on the kind of answer they want, than can possibly be
communicated in a few sentences of email. It's very hard to give
good statistical guidance without quite a bit of back-and-forth
communication about the problem.
So, you're on your own about choosing a "clustering method".
Do you mean that you want to further cluster the nodes of the
grid returned by Kohonen's SOM ? (I don't even understand the
question clearly.) Well, then you'll need to think about what
criteria you want to use to do that ... and only then think
about what methods might be appropriate.
As to the technical part, help("SOM"), help("batchSOM")
make it
clear that both functions do have a return value. They do NOT
"only give a visual representation of the topological mapping."
It's your problem, have fun with it ! Keep in touch.
- tom blackwell - u michigan medical school - ann arbor -
On Wed, 16 Apr 2003, Jonck van der Kogel wrote:
> Hello everyone,
> I'm new to this list, so let me introduce myself: my name is Jonck van
> der Kogel and I am a graduate student at the Erasmus University of
> Rotterdam. I am currently working on my thesis which is on the use of
> artificial intelligence for large data-sets.
> To do an analysis of a certain data-set I want to use Kohonen's SOM
> algorithm. However, as I understand it, the SOM algorithm as it's
> implemented in R only gives a visual representation of the topological
> mapping. I want to do further analysis with this mapping and thus I
> need to do a clustering of the topological mapping produced by the SOM.
> I was wondering wether anyone could give me some advice on which
> clustering method in R is most suited for clustering the map produced
> by the SOM algorithm.
>
> Thanks very much, Jonck