[Eeglablist] ICA: number of components and channels

[Tarik S Bel-Bahar]

​Hello Irene, some quick comments below. Best wishes.


​
​******************​

​1. you could try reducing with the PCA flag in the runica function. this
would effectively give you "fewer components" than the "number of channels"
that go into ICA. Check out runica documentation in matlab, and online
tutorial info on running ICA if you haven't had a chance to yet. However
there are some issues with a PCA vs.ICA approach, for some researchers the
differences between the two approaches are very important, whereas other
researchers use PCA whenever they think it's necessary to "reduce
dimensionality".

2. Have not heard of an option where one can tell eeglab's ICA function to
ignore some channels. Check and see in the  AMICA distribution, it might
have some advanced functions like that.

3. I think within eeglab you have to have the all and only the channels you
want ICA to use inside the data before starting ICA, and nothing else. In
other words, it may not be easy to just ask ICA to skip or not include some
channels.

4. Remember that after ICA eeglab needs the EEG data to have the same
number of channels that were there when ICA is run. I think also any file
you want to apply the ICA solution to needs to have the same channels as
the file that went into ICA.

5. I would suggest removing and saving the special channels you want to
keep but that you want ICA to ignore. So for ICA, don't have those channels
"in" the dataset.
Perhaps you can analyze or add those channels in later if you need to.

6. Please refer to pass eeglablist discussions and Makoto's pipeline for
recommendations on how to maintain the rank of the data before feeding it
to ICA (usually by removing one channel from the whole set if you have done
re-referencing).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20160317/84cd7c41/attachment.html>