ICA with PCA and predetermined mixing matrix

[Christian Hesse]

Hi Jeroen,

at present componentanalysis using a predetermined mixing matrix
requires this to be a square matrix.

Until this is changed you can use the following workaround (hacky but
sound, tried and tested at the FCDC):

1) call componentanalysis to get your comp.topo = [204-by-64] matrix

2) do an orthogonal expansion of matrix comp.topo, this gives you a
204-by-204 matrix where the columns 65:204 form a subspace that is
orthogonal to the subspace formed by the first 64 columns (so by
definition the signals living in each are uncorrelated), and which
corresponds to the subspace of signal that you threw out by doing PCA
and retaining only the first 64 components. There should be a file in
fieldtrip/private called "expand_orthogonal.m" containing a function
which does this for you (if it is not there let me know and I'll
email it you separately).

3) now call componentanalysis again on whichever dataset with
cfg.method = "predetermined mixing matrix" and cfg.topo = expanded
(comp.topo) and just ignore components 65:204 since they are a
mixture of the "garbage" that was ignored by the ICA anyway (the ICA
only separated the signals in the subspace spanned by the first 64
PCs, which in practice means that the ICs you estimate are mutually
statistically independent but only uncorrelated w.r.t. the PCs
65:204 !!!).


Hope that helps,
Christian



On 8 Feb 2008, at 10:52, Jeroen Geuze wrote:

> Hello all,
>
> I am trying to use ICA with a predetermined mixing matrix on my
> data. I have the following situation: I want to use Cluster
> Analysis on the components of two of my datasets. In this case, you
> want to have the same components for both datasets. I have measured
> with 256 electrodes and after removing the bad electrodes, 204 are
> left. I then run ICA with the PCA setting to 64. I run this on my
> two appended datasets. Then I want to use the mixing matrix found
> there to do ICA on each of my two separate datasets. But because I
> go from 204 electrodes to 64 components, the weights matrix is not
> square. Is there a way to use componentanalysis with a non-square
> predetermined mixing matrix? Or is there another way to get the
> components of the appended datasets into the two separate datasets?
> Regards,
>
> Jeroen Geuze
>
>
>
> --
> J. Geuze M.Sc.
> Cognitive Artificial Intelligence
> NICI, Radboud University Nijmegen
> Montessorilaan 3, Room B.02.28
> P.O.Box 9104
> 6500 HE Nijmegen
> The Netherlands
> Tel: +31(0)24 3611938
> J.Geuze at nici.ru.nl
>
> ----------------------------------
> The aim of this list is to facilitate the discussion between users
> of the FieldTrip  toolbox, to share experiences and to discuss new
> ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/
> archives/fieldtrip.html and http://www.ru.nl/fcdonders/fieldtrip.
>



----------------------------------------------------------------------
Christian Hesse, PhD, MIEEE

F.C. Donders Centre for Cognitive Neuroimaging
P.O. Box 9101
NL-6500 HB Nijmegen
The Netherlands

Tel.: +31 (0)24 36 68293
Fax: +31 (0)24 36 10989

Email: c.hesse at fcdonders.ru.nl
Web: www.fcdonders.ru.nl
----------------------------------------------------------------------





----------------------------------
The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/fcdonders/fieldtrip.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20080208/63fd5941/attachment-0001.html>