[Eeglablist] Massimiliano Papera removing ICA component

Massimiliano Papera massimiliano.papera at googlemail.com
Thu Oct 11 05:22:00 PDT 2012


I see,

However, I guessed that already actually but is good to know that s the
right way. Thank you so much.
About the running the ICA. I ran the ICA on epochd data, sinche on the
forum I saw people sayig it does not matter (e.g. ICA agnostic).
I ran:
first ICA
utilities to reject epochs based on ICA analysis (unstereotyped artifcats)
second run ICA
visually removing bad component (hopefully containing stereotyped artifact
such as blinks etc)


is that bad?

Why I need to run the ICA with the option pca afterwards? any idea?

Thank you


On Wed, Oct 10, 2012 at 9:01 PM, Tarik S Bel-Bahar
<tarikbelbahar at gmail.com>wrote:
Greetings Massimiliano,

I am not sure if this is the most complete response, but hopefully this
will help you a little.
I think that the "amount of data changed" in the tutorial text you sent
refers to "how many ICs" have been removed. It should be modified and
expanded to be a bit more clear.

Perhaps an expert on the list has a better idea, but it seems that if you
had
256 channels, and then ran ICA normally, and then removed 4 ICs, then you
are left with 252 ICs but 256 channels.
If you run ICA again using the data and ICs that are left,
as per the tutorial instructions, you should then set PCA to 252, so that
ICA generates the correct number of ICs.
it should then deliver 252 ICs in the new ICA decomposition.

Note that the above suggestion assumes (as per the tutorial)
that you are not epoching the data after dropping the ICs and that you are
not epoching the data before doing your second ICA.

Note that you cannot hurt anything by trying a few different ICA runs. It's
better to make some mistakes
and familiarize yourself with the processing.

Please also check the eeglab list archives for a variety of conversations
about number of ICs,
and the best ways to run ICA under various conditions, and for various
needs.

Bocca di lupe!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20121011/6a75b29a/attachment.html>


More information about the eeglablist mailing list