On 1 Mar 2001, at 7:11, niels Waller wrote:
>
> The (slow) SOM code can be found on my web page (towards the bottom)
>
> http://peabody.vanderbilt.edu/depts/psych_and_hd/faculty/wallern/
Thank you, I have tested it (well, on toy examples) and it works just fine
- as a clustering tool. How 'slow' it is I don't know - as in, I
have not
tested it on large problems - but in any event the Kaski* paper (below)
suggests some possible speedups.
SOM, I think, is a bit of a moving target, but as well as the clustering
aspect there appears to be also a data visualisation aspect which is not
represented in your code as far as I could see. Paraphrasing Kaski, one
implements a regular 2d grid on which 'units' are located : each unit
has
a 'reference vector' (a vector of means), the 'winning unit'
for a given
input vector (based on Euclidean distance in the 'measurement space')
has
its reference vector updated as do its neighbours in 'grid space' (with
weights approc proportional to 1/distance in the GRID SPACE) . Eventually
this process organizes the data on the grid in some meaningful fashion.
You can then use this grid as an 'ordered groundwork' on which you can
display a bunch of things .. the original measurement variables, cluster
densities etc. It has a certain appeal imho.
* Samuel Kaski's thesis "Data exploration using self-organizing
maps" is at http://www.cis.hut.fi/~sami/thesis.ps
John Aitchison
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._