[Eeglablist] PCA option to runica

[Philip Michael Zeman]

If you are using the PCA parameter with runica:

(1) the data is reduced in dimensionality via a very basic implementation of 
PCA. The data returned from the PCA operation will have the dimension 
specified by the PCA parameter.(Note it does not do any rotations of the data 
based on probabilistic information.)

(2) once the data has been reduced in dimension, ICA run on the reduced 
dimension data.

My partner and I muddled with the PCA parameters extensivly over the last few 
months because we were trying to do analysis on exceptionally noisy data.  The 
PCA option provides you with a cut-off to drop 'weak signals/noise'.  As you 
reduce dimension, two things will happen:  (1) you will reduce the possible 
*weak components* (with respect to a PCA rotation) you will find, (2)you will 
reduce the number of 'maximally independent' components you can generate.  For 
us result (2) makes it harder to do clustering as the components across 
subjects may not be consistently similar. ....ie, harder to cluster

~Phil Zeman



University of Victoria
Assisitve Technology Team
http://www.uvic.ca/uvatt



>===== Original Message From pickard at telusplanet.net =====
>Hi,
>
>As I do not have large number of trials, I opted to run the 'pca' option but
>have gotten slightly confused about the output.  I assumed that the output
>would be principal components; however, in chapter 1.9 of the tutorial, 
section
>1.9.2 (the topoplot above section 1.9.3), it shows "an independent
>component ... binica decomposition used PCA".  So, is the output principal
>components or independent components?  How is the output derived?  Sorry if
>this is a dumb question, my background is clinical/behavioural and I'm still
>trying to sort out PCA/ICA etc.
>
>Thanks for any info you can provide.
>
>Patti Sorensen
>PhD Candidate
>University of Lethbridge
>
>
>
>_______________________________________________
>eeglablist mailing list eeglablist at sccn.ucsd.edu
>http://sccn.ucsd.edu/mailman/listinfo/eeglablist
>Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu