[Eeglablist] EEG.icaact vs. ICAcomp

[hughgs at duke.edu]

Version 5.03b.  I suspect it has to do with removal of the mean.

On Thu, 7 Dec 2006, arno wrote:

> Dear George,
>
> what version of EEGLAB are you using? In earlier versions than 5.0b,
> EEGLAB was not handling the output of Fastica properly (the weight
> matrix and the the inverse matrix were inverted). This is probably the
> problem. After you download EEGLAB, you will have to rerun ICA though.
> Please send us more comments at eeglab at sccn.ucsd.edu or to enter bugs,
> go to the Bugzilla web page
>
> https://sccn.ucsd.edu/eeglab/bugzilla/
>
> Best,
>
> Arno
>
> hughgs at duke.edu wrote:
> > I'm trying to understand the interfaces to the fastica algorithm and am
> > having a problem with the difference between EEG.icaact computed by
> > eeg_checkset and the "icasig" output from fastica().
> >
> > It seems that the two should be relatively equivalent but when I check the
> > difference between the two I find that,
> >
> > max(max(abs(EEG.icaact - ICAcomp))) = 3,
> > where ICAcomp is the variable used by pop_runica() when running fastica().
> >
> > I'm using the eeglab_data.set that was shipped with the EEGLAB v5.03 as my
> > data set.  Am I missing something?  Thanks.
> >
> > George S. Hugh
> > 261 Hudson Annex
> > (919)660-5167
> > hughgs at duke.edu
>

George S. Hugh
261 Hudson Annex
(919)660-5167
hughgs at duke.edu