[Eeglablist] How to make an ICA run in less than a week?

Wed Jul 25 12:41:41 PDT 2012

Hi Baris,

In order to overcome the time costs of ICA I have been using EEGLab and a modified Batch plugin on the Shared Hierarchical Academic Research Computer Network (SHARCNet, www.sharcnet.ca<http://www.sharcnet.ca/>, it is on my to-do-list to explore similar international HPC resources). While this does not necessarily speed up the ICA calculation it allows me to run up to 64 simultaneous Matlab workers (shared across users). I am also exploring Octave in this context (no licensing limitations).

While presenting this strategy to the HPC community at a recent research day I was enthusiastically approached by someone who witnessed a GPU optimized ICA implemented in R. I have not had a chance to look into it further than finding this sitehttp://brainarray.mbni.med.umich.edu/brainarray/rgpgpu/

See alsohttp://sccn.ucsd.edu/wiki/GPU_and_EEGLAB

___
Sent from an auto-correcting touch screen device.
___
James A. Desjardins, MA
Electrophysiology Technologist
Cognitive and Affective Neuroscience Lab, Psychology Department
Jack and Nora Centre for Lifespan Development Research
Brock University
500 Glenridge Ave.
St. Catharine's, ON. Canada L2S 3A1
905-688-5550 x4676
___
"'Cause you never can tell What goes on down below!
"This pool might be bigger Than you or I know!"

McElligot's Pool
Dr. Seuss 1947

On Jul 24, 2012, at 2:42 PM, "Baris Demiral" <demiral.007 at googlemail.com<mailto:demiral.007 at googlemail.com>> wrote:

Hi James,

1- For frequency of interest of 200Hz, you can down-sample your data
down to 512Hz. But before that, low-pass filter your data with cut off
value of 250Hz or so. Follow Nyquist theorem.
2- You can run ICA on continuous data, but make sure you have marked
AND rejected the bad segments in the EEG data, since bad recording
WILL influence your ICA analysis A LOT. Pre-processing is still
important, even though ICA will eventually be used to reject artifacts
later. SO, observe your data well.
3- Xeon processors should have more than 2 cores I think. Run parallel
MATLABs on the same computer over the different cores for different
sets of subjects.

Arno, and other eeglab gurus, do you have any plans to
implement/combine ICA routines with parallel computing toolbox in
matlab?  Is it possible to create, say, a PARICA? Is this
theoretically possible? For instance, ICA routine can divide the data
into 4 or so and then combine them over different cores/machines?

Cheers,
Baris

On Fri, Jul 13, 2012 at 1:07 PM, James Schaeffer <schaefj3 at gmail.com<mailto:schaefj3 at gmail.com>> wrote:
Hi All,

Thanks for your help.  I have started using binica and it is going a bit
faster.  We have chosen to use a high sampling rate (2048 Hz) because we
thought it might be needed for cross-frequency coupling analysis with high
gamma up to 200 Hz.  If I downsample, will I still be able to see these
relationships?

Thanks again,

James

On Thu, Jul 5, 2012 at 6:24 PM, James Schaeffer <schaefj3 at gmail.com<mailto:schaefj3 at gmail.com>> wrote:

Dear eeglablist,

I am running ICAs on 10 minutes of Biosemi EEG data, collected from 128
channels, and sampled at 2048 Hz. However, the ICAs have been running for a
few days without much progress. Two computers have 8G RAM and an Intel(R)
Xeon(TM) CPU 2.80GHz. One has been running an ICA for 6 days and is on step
33; the other, for 4 days, and is on step 28. Another computer has 12G RAM
and an Intel(R) Xeon(R) CPU 5110 @ 1.60GHz; it has been running for 3 days
and is on step 60. All are running openSUSE 12.1 (x86_64), with eeglab
version 10.2.2.4b. The 'free' command indicates that they are not using any
swap space.  Should it be taking this long?  Is it possible that we have
more than the one copy of the data in ram or that Matlab or Eeglab has
placed something else in ram, or is busy with other processes? Is there
anything I can do to speed up this process? Any help would be greatly
appreciated.

Thanks in advance,
James

_______________________________________________
Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu<mailto: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<mailto:eeglablist-request at sccn.ucsd.edu>

--
Ş. Barış Demiral, PhD.
Department of Psychiatry
Washington University
School of Medicine
660 S. Euclid Avenue
Box 8134
Saint Louis, MO 63110
Phone: +1 (314) 747 1603

_______________________________________________
Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu<mailto: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<mailto:eeglablist-request at sccn.ucsd.edu>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120725/d2e4827e/attachment.html>