[Eeglablist] Fwd: ICA weights not "saved"?

[Keith Yoder]

Since two of us have encountered this issue before, it's probably worth
posting this to list.

---------- Forwarded message ----------
From: Keith Yoder <kjyoder at gmail.com>
Date: Thu, Aug 13, 2009 at 5:21 PM
Subject: Re: [Eeglablist] ICA weights not "saved"?
To: Andrew Hill <andrewhill at ucla.edu>


In our case, we have to restrict the amount of data that ICA is working with
to produce sensible components, but this is certainly driven in large-part
by our paradigm.  We are having subjects play a video game that consists of
four distinct minigames, each of which has one or more cognitive tasks built
into it (for more information see http://autismcollaborative.org).
This has given us very large datasets (128 channels recording for an hour).
 I thought this would be great - ICA would have a *lot* of data to work
with.  However, this means that the specfic events we are interested in
comprise a very small proportion of the EEG data individually.  For
instance, a go-no-go task in one of the minigames occurs less than 60 times
per game.  If I'm assuming that the salient effects in the EEG occur within
a three second window at each event, then I'm expecting ICA to pick out
components that may only be active for for a total of 3 minutes out of the
hour.
 -Keith

On Thu, Aug 13, 2009 at 4:40 PM, Andrew Hill <andrewhill at ucla.edu> wrote:

> hi Keith,
> thanks for the reply - explicitly saving after ICA weights are calculated
> is what i was missing.
>
> on running the ICA on epochs versus continuous recordings.. I was under the
> impression that Arno and Scott encouraged running ICA on the "longest" data
> chunks possible, and that running them on epochs would actually produce
> artifacts at the edges of the epoch.. are you not finding this?
>
> best,
> andrew
>
> On Aug 11, 2009, at 5:41 AM, Keith Yoder wrote:
>
> I had a similar experience at first.  Here's my resolution:
> Once Run ICA is finished, manually save the component-weighted datset.
>  This can be done by either 'File -> save current dataset(s) as' or, if you
> are using a script, the pop_saveset(EEG) command.
> A rough outline of our process:
> Import the dataset
> Highpass filter the dataset
> Save that filtered dataset
> Extract epochs
> Save the epoched datasets
> Run ICA
> Save the component-weighted epoched datasets
> We've found that running ICA before epoching the datasets yields
> nonsensical results.  Additionally, if you epoch the data first, then the
> ICA algorithm will only have to work on the data surrounding your event of
> interest.  This will can greatly reduce the amount of time required for the
> algorithm to identify components.  If your events are widely spaced
> throughout the dataset, you may even be able to include all of your channels
> at once.
> Hope this helps!  If you have any questions, feel free to let me know.
> -Keith
> Research Aide
> Belmonte Autism Lab
> Cornell University
>
> On Mon, Aug 10, 2009 at 3:49 PM, Andrew Hill <andrewhill at ucla.edu> wrote:
>
>> hi folks,
>>
>> i might be misunderstanding something about how the "Run ICA" works.
>> i've been running ICA against .set files, half of the channels at a
>> time to avoid out of memory errors... i got through all my .sets, and
>> then extracted epochs from all the .set files, assuming the ICA
>> weights would be inherited by the epochs.
>>
>> the problem is, when i open one of these .set files, it appears that
>> the data file doesn't actually have ICA weights saved into it. am i
>> doing something wrong??  is there some step to explicitly save the
>> weights with the .set file that i need to do?
>>
>> these weights don't appear to be saved even with the "write ICA
>> activations to disk" option checked...
>>
>> i cannot load more than one .set into memory at a time... and i cannot
>> process more than half the channels in one Run ICA go.. so what am i
>> missing?  why are these not saved to the .set?
>>
>> thanks,
>> andrew
>>
>> _______________________________________________
>> 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
>>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20090819/55dd831d/attachment.html>