[Eeglablist] How to deal with corrupted ICA decompositions

[Makoto Miyakoshi]

Arno, other possibility of becoming rank-deficient is bridging across
channels. What cap did you use?

I recommend you check the rank using 3-sec length of data or around (needs
to be adjusted; longer the data, lower the rank is estimated). I still
think it is the case of rank deficiency, although I don't understand why
'pca', 20 in 32ch dataset did not solve the problem. If the data rank is
lower than 20, then you can't say it's ICA's fault.

By the way did you show spectral plot of the decomposed activities? You'll
find an interesting plot...

Makoto

On Fri, Jan 30, 2015 at 9:46 AM, Arnaud Delorme <arno at ucsd.edu> wrote:

> The data was not full rank but necessary corrections were done (PCA of
> number of channels-1). Full rank is a common issue though and one that
> should be checked first.
>
> This is another problem which might be specific to the EEG amplifier and
> to Infomax ICA in which the first two components become quite noisy.
>
> Arno
>
> On Jan 29, 2015, at 5:30 PM, Makoto Miyakoshi <mmiyakoshi at ucsd.edu> wrote:
>
> Arno, check if your original data was full rank. I've repeatedly seen this
> before, and I found it happens when the data are not full ranked. By the
> way, when you run matlab rank() don't use too long data, just give it short
> data long... you may also want to test it by changing the length (1:1000,
> 1:10000, 1:100000...) In the middle of this wiki page
> http://sccn.ucsd.edu/wiki/Makoto%27s_preprocessing_pipeline I put an
> example of such case.
>
> Makoto
>
> On Tue, Jan 20, 2015 at 3:03 PM, Arnaud Delorme <arno at ucsd.edu> wrote:
>
>> I have recently added a small section to the EEGLAB wiki that deals with
>> working with ICA decomposition where the two first components activity is
>> extremely noisy (which seems to be a bug of Infomax). This does not happen
>> often, but I thought I should share it with the list.
>>
>>
>> http://sccn.ucsd.edu/wiki/Chapter_09:_Decomposing_Data_Using_ICA#How_to_deal_with_.22corrupted.22_ICA_decompositions
>>
>> Arno
>> _______________________________________________
>> 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
>>
>
>
>
> --
> Makoto Miyakoshi
> Swartz Center for Computational Neuroscience
> Institute for Neural Computation, University of California San Diego
>
>
>


-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20150130/b3da3835/attachment.html>