comp.lang.idl-pvwave archive: archive

Home » Public Forums » archive » Cluster analysis

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

Re: Cluster analysis [message #41891 is a reply to message #11153]

Fri, 03 December 2004 15:30

Chris.Jacobsen
Messages: 1
Registered: December 2004

Junior Member

hfrey@ssl.berkeley.edu (Harald) wrote in message news:<51c9fb99.0412021550.41c2d53d@posting.google.com>...
> The cluster analysis in IDL with clust_wts and cluster finds only 8
> useful clusters. I can set n_clusters to a greater number, however all
> cluster more than 8 useful ones are at locations, that do not contain
> any of the samples. Is there any statistical reason that the number of
> clusters can never exceed 8?
>
> I have a data set with 24 variables and 7680 samples and would expect
> that I can easily find 20 or so different clusters.
>
> Does anybody have better IDL software that does the cluster analysis
> properly?

In our work we started out using the "stock" IDL
cluster routine but we have added to it a bit.
Still, we have not changed the basic algorithm.
We've found that preparation of the data can
make a big difference. If the variation in
variable X is 100 times bigger than the variation
in variable Y, then the clustering (which looks
at simple Euclidian distance) will not see the
variation in Y very well. One approach is
to subtract the mean of each variable, and
apply a scale factor to the data in variable
Y so that it is spread out over the same distance
as in variable X.

With data preparation of that sort, the stock
IDL routine can certainly find more than 8 clusters.

A paper on our work is at
http://xray1.physics.sunysb.edu/~micros/publications/papers/ lerotic_ultramic_2004.pdf

Report message to a moderator

[Message index]

		Cluster analysis By: mph410 on Fri, 27 February 1998 00:00
		Re: Cluster analysis By: David Foster on Mon, 09 March 1998 00:00
		Re: Cluster analysis By: Chris[2] on Tue, 07 December 2004 11:32
		Re: Cluster analysis By: Harald Frey on Mon, 06 December 2004 14:21
		Re: Cluster analysis By: George N. White III on Sun, 05 December 2004 06:43
		Re: Cluster analysis By: Chris.Jacobsen on Fri, 03 December 2004 15:30
		Re: Cluster analysis By: James Kuyper on Fri, 03 December 2004 06:44

Previous Topic:	READCOL compilation error
Next Topic:	Re: Canny Edge Detection

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Mon Oct 27 05:15:11 PDT 2025

Total time taken to generate the page: 0.02731 seconds