[Eeglablist] Clustering Problem

[wibral at mpih-frankfurt.mpg.de]

Dear Listusers,

I am trying to use the new clustering tools in EEGLAB.
I build a study with 19 subjects and one condition. Since ICA was done
once per subject I have left the session information blank, as indicated
in the tutorial. I build the preclustering information for the ERP with
10 dimensions. I have left the number of ICs per subject unconstrained
(~55-60). Now, if I try to cluster these components (~1000) using
kmeans, all works well if i specify an unnaturally small number of
clusters (say 10).
If I increase that number to ~20 I get an error message:
EEGLAB error in function pop_clustedit() at line 258:
Error using ==> length
Too many input arguments.

If I increase the number of clusters even further, I repeatedly get
things like '..created an empty cluster' and the ensuing division by
zero when Kmeans tries to compute the standard deviation for that empty
cluster, which is logical.

My questions is, whether the above things are bugs, or whether I should
try to increase the dimensionality of the feature space, as the expected
number of clusters is high. I suspect it to be actually around 50 or
more.

Thanks for your help and best regards,
Michael Wibral