[Eeglablist] Using PCA before ICA decomposition? (Chang Gu)

[Chang Gu]

Hi Stavros,

Thanks a lot for the PCA function!

Before you apply the PCA by princomp(), do you re-reference the data to the
averaged-reference?  We're using 128 recording system, and I found that the
raw data from channels close to the eyes  are much larger(at 50~100 micro
volt scale) than the data from channels near Cz (<1 scale). I suspect the
scale differences between channels may influence the PCA performance such
that small PCs may account for the channels with small scale. By
re-reference the data, channels near Cz change the scale to ~10 micro
volts. Also, Matlab suggests we could use princomp(zscore(X)) to rescale
the data. What do you think using z-score?

best

2012/1/19 stauros dimitriadis <stdimitriadis at yahoo.gr>

> Dear Chang
> Since it is important to reduce the dimensionality of your data by
> adopting PCA, a simple rule is to select
> an appropriate number of principal components regarding the variance of
> your data.
>
> In MATLAB script:
>
> [pc,score,latent,tsquare] = princomp( (time x triasl)  x channel); % concatenate your trials in time (suggested)
>
> r=find(cumsum(latent)./sum(latent) > 0.98);
>
> %r(1) is the total number of components need to capture more than 98 % of the variance of
>
> %your initial data
>
>
> Afterward, proceed with the ICA.
>
>
> I hope that i help you !!
>
>
>
>
> Best regards
>
> Stavros Dimitriadis
>
> http://users.auth.gr/~stdimitr/index.html
>
>
> 1)Electronics Laboratory <http://www.ellab.physics.upatras.gr/content/view/198/191/lang,english/>,  <http://www.ellab.physics.upatras.gr/index.php?lang=english>Department of Physics <http://www.physics.upatras.gr/>,  <http://www.ellab.physics.upatras.gr/index.php?lang=english>University of Patras <http://www.upatras.gr/>
> 2)Artificial Intelligence Information Analysis lab  <http://poseidon.csd.auth.gr/EN/>Department of Informatics <http://www.csd.auth.gr/>
> *Aristotle University of Thessaloniki <http://www.auth.gr/home/>*
>
>
>
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu
>



-- 
Chang Gu
Psychology & Human Development
Vanderbilt University
Nashville, TN
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120120/e80ece8e/attachment.html>