[Eeglablist] ICA does not start running properly

Arnaud Delorme arno at salk.edu
Tue Apr 19 14:40:04 PDT 2005

Dear Maarten,

>after successfully running ICA on epoched data to remove eyemovement 
>artifacts, I want to run ICA again on the filtered epochs to remove 
>other artifact components. I do not want to work with the remaining 
>components of the first ICA as these epochs can differ a lot from the 
>task related epochs I want to use.
Since you seem to have different data epochs (between the first ICA pass 
on all your data, and the second one on your task related epochs), it is 
OK to rerun ICA (it is useless to rerun ICA on the same data epochs; see 
the tutorial for more details). In theory, it should not be necessary to 
change any parameter since you do not use exactly the same data epochs. 
However, since the algorithm does not seem to converge with your data, 
you probably have a large amount of common data epoch, so you should use 
the 'pca' option to reduce the number of dimensions.  If you removed 10 
artifactual components and have 64 electrodes, use the option " 'pca', 54 ".

Note that you choose to run ICA on your task related epochs only. We 
usually use all the data available to perform ICA decomposition and then 
study components that vary between experimental conditions. The reason 
for using as much data as possible is that we realized that ICA (at 
least Infomax ICA, the default option) returns "nicer" components (i.e. 
component which can be modeled using a single equivalent dipole) as the 
amount of data increases. The reason why ICA finds more cognitively 
related components as you increase the amount of data is that cognitive 
components tend to be expressed over the whole data. In contrast, 
electrical artifacts are usually only expressed transiently, so, as you 
increase the amount of data, the relative proportion between cognitive 
and electrical artifacts is increasing. As a consequence, the number of 
component ICA will dedicate to the cognitive components will also 
increase. This is our hypothesis at least.

Hope this help,


>However, for some unclear reason the ICA procedure starts stalling at 
>the "Beginning ICA training..." print. Sometimes it gets to lowering its 
>learning rate and starting again, but I tend to abort the thing after 
>half an hour.
>Running the ICA for removing the eyemovements was no problem at all. It 
>never took more than one ore two minutes. Even when I used it on more 
>epochs than what I'm trying to now.
>What could be the problem?
>eeglablist mailing list eeglablist at sccn.ucsd.edu
>Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu


*Arnaud Delorme, Ph.D.*
Swartz Center for Computational Neuroscience, INC, University of San 
Diego California
La Jolla, CA92093-0961, USA

*Tel* :/(+1)-858-458-1927 ext 15/
*Fax* :/(+1)-858-458-1847/
*Web page*: sccn.ucsd.edu/~arno <http://www.sccn.ucsd.edu/%7Earno>
*To think upon*:

    The ultimate measure of a man is not where he stands in moments of
    comfort and convenience, but where he stands at times of challenge
    and controversy.

        /Martin Luther King/  

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20050419/a5444408/attachment.htm

More information about the eeglablist mailing list