[Eeglablist] ICA error

[Tarik S Bel-Bahar]

Hi Erica, this looks like it's a bug but not one that usually comes up. It
seems that one should be able to select a smaller number of channels to be
run.
SInce it is repeatable and you haven't found a solution, please log it into
the eeglab bugzilla. Some of my thoughts below echo Makoto's.

It may not be an error, but rather eeglab basically requesting that the
channels you put in be the same as the number of channels in the dataset.
In the meantime you may want to arrange your data differently to work
around this until there is a fix or explanation.
Some other quick thoughts below, not sure if they will help, thanks for
giving suggestions a try if you have time.
Cheers!


********Extra thoughts for Erika

If you haven't had a chance, to,  you may want to try    a)delete a channel
before ICA to reduce rank after re-referencing
 or b) reduce rank by reducing the dimensionality of the ICA by adding a
"pca" flag in the function to number of channels minus one.

You may be able to solve the problem by either 1) delete a single channel
right before the ICA step (and running 1:63)
or 2) not adding back the reference channel before ICA. Both together may
work too.

What happens when you just don't add in the reference channel? This might
be creating the mismatch.
I would also step through more closely between the channel count and labels
variables in the EEG structure to see if there was a mismatch in eeglab's
expectations.
See also the sub-structures of EEG that update with reference channel
information.
Each step should be doable via GUI, and if you are using matlab scripts
there should be a checked to make sure commands look like those derived
from eegh history of commands run via the eeglab GUI.

Try triple-checking to make sure you "adding in the reference channel" and
"doing average reference" correctly (should all work through the gui)

Further, it's  important to know is how results look when you run the 1:64
ICA.

Please see if following sequence works if you get a chance:
1. Take the original raw file with channel locations loaded (load 63
channels)
2. Remove bad/worst channels and bad/worst periods.
3. Interpolate/add the single reference channel (possibly do after ICA or
not at all)
3. Run ICA on all the remaining good channels (should be at least 45 and
well distributed)
4. Interpolate bad channels (if computing EEG/ERP metrics at channel level
rather than ICA-level.
5. Stay in ICA space, and compare/analyze similar ICs across your
experiment's design.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20180108/db18ca7b/attachment.html>