I am attempting to use mne-python in a grid computing application.
The only function that I need is read_forward_solution .
Both mne-python and all the files imported and used by it must be included in the package which is uploaded to each grid computer in order to get it to work.
As a first cut I'm including both numpy and scipy along with mne-python since it imports modules from them.
The whole thing is failing because the linear algebra routines in numpy require a shared compiled library which is not found on the target machines on the grid.
I've fooled around with commenting out the import statements for numpy and scipy functions but the dependencies are not readily eliminated.
Perhaps someone has a better idea about how to go about this.
I would like to get this to work and secondarily to work without requiring numpy or scipy at all.
The reason for that is that I am running on average 35,000 jobs per day.
Each requires uploading the entire package which requires that much more network bandwidth given the size of the assembled packages.
A somewhat optimal solution would be to "skeletonize" the chain of dependencies and required files needed to support the limited functionality required for the application.
Is there a simple algorithm or even a tool for doing this?
Your thoughts and direction would be appreciated.
Thanks.
Regards,
Don
Don Krieger, Ph.D.
Department of Neurological Surgery
University of Pittsburgh
(412)648-9654 Office
(412)521-4431 Cell/Text
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140709/c76076e0/attachment.html