[Eeglablist] About running ica on several machines

[Tarik S Bel-Bahar]

Hi Matthew, notes below. all the best.
I am not sure but running a single ICA decomposition across several
machines might be tricky or problematic. With a saavy programmer, you
should be able to modify the code to, for example, have a different bias
file name.
I hope at least one of the suggestions below is useful to you. You will
most likely get better answers from eeglab developers and other users who
have solved issues similar to yours for their needs.

1. parallelized version of the code that can make the most use of
parallelized matlab processing on a single computer
2. explore cuda based ICA
3. consider running multiple instances of matlab (on separate computers)
4. use a gpu/processor bank/farm to send the "ica job" to
5. saavy users can also likely benefit from cloud-based services, of course
at some cost
some national programs that make free processors available to some
researchers e.g., canada's sharknet I believe)
6. check out EEGGIFT documentation for possible details regarding your
question
7. why not just run let each of your "machines" process one EEG file at a
time. You are already ahead if you have "multiple machines" available :)


















On Sun, Feb 11, 2018 at 8:58 PM, Matthew Ma <matthewhang92 at gmail.com> wrote:

> Dear all,
>
> I have been using eeglab for a while for preprocessing stuff, it is very
> useful so far. I have been noticing the processing speed of ica was quite
> low (I have already used binica and done downsampling of my data). I am
> thinking if I can make use of several machines which can be used to do the
> preprocessing.
>
> As I noticed, the binica will generate the sphere and weight matrix as a
> file, also with the bias file. However, the bias file is always produced
> with the same file name, this gives rise to my question. As my work
> directory is located in another shared server(I access it using sshfs), I
> am wondering if the ica to produce wrong results if I run ica in several
> machines simutaneously (due to the same name of bias file...). I know the
> easiest solution is to duplicate the work directory to every local
> machine...but that seems a little bit stupid to do so.
>
> Looking forward to your reply! Thank you very much.
>
> Best regards,
> Matthew
>
> _______________________________________________
> 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/20180214/f58459bb/attachment.html>