[Eeglablist] relationship between no of channels and no of components

[Arnaud Delorme]

Dear Piotr,

this is most likely because some of your channels are linear  
combinations of other channels. The runica function will first compute  
the rank of the data matrix and reduce the dimensionality of it if  
necessary.

The rank of a common reference EEG data array is usually equal to the  
number of channel (minus 1 if the reference is listed in the data  
channels). There was a Matlab bug in the first versions of Matlab 7 on  
64-bit machines that caused the rank function to return inaccurate  
results when using single precision float data. This is why we now  
forces the EEG data to be converted to double precision before running  
ICA (although the data will remain in single precision if not enough  
memory is available). It might be that you re not using the latest  
version of Matlab and EEGLAB.

Refer also to the following bug repports:

https://sccn.ucsd.edu/eeglab/bugzilla/show_bug.cgi?id=317

https://sccn.ucsd.edu/eeglab/bugzilla/show_bug.cgi?id=191

https://sccn.ucsd.edu/eeglab/bugzilla/show_bug.cgi?id=187

Hope this helps,

A. Delorme

On Oct 13, 2009, at 6:22 AM, Piotr Jaskowski wrote:

> Dear All,
>                I run ICA with 32-channel segmented data. I expected  
> to get 32 ICA components. In fact, I’ve got only 22 components and a  
> following message:
>
> Input data size [22,51775] = 22 channels, 51775 frames/nAfter PCA  
> dimension reduction,
>   finding 22 ICA components using extended ICA.
>
> Could anybody explain me where the remaining 10 components  
> disappeared?
>
> Best regards,
> Piotr
> ---------------------------------------------------
> Prof. Piotr Jaskowski, Ph.D.
> Department of Cognitive Psychology
> University of Finance and Management
> Pawia 55, 01-030 Warszawa, Poland
> http://cogn.vizja.pl
>
>
>
>
>
> --
>
> Specjalna oferta ksi??ek i podr?cznik?w: http://oferta.vizja.net.pl
>
>
>
>
> <ATT00001.txt>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20091016/89573885/attachment.html>