Rewrite apps in python (keeping the core library in C++)

[papadop]

Advantages:

• Better test the python interface which should be reliable now.
• We can defer input/output to python, so there is less code to maintain. In particular, matrix IO could go away after a while, thus removing the dependency on matio and hdf5.
• Similarly, most of the higher level matrix operations are in apps. It needs to be checked, but I think that most of OpenBLAS/MKL/... can go away and be differed to python.
• vtk and nifiti can probably be handled the same way.
• In turn, this will greatly simplify cmake stuff.
• And this is probably, at very low cost.

Drawbacks:

• python will become a hard dependency. We have to check with our matlab users whether this is acceptable or not...

[agramfort]

@ftadel what do you think? would it be a deal breaker for many brainstorm users?

[ftadel]

We've done all the work to call Python functions from Brainstorm using the Matlab-Python integration. This way we can call functions from MNE-Python or other libraries:

• https://neuroimage.usc.edu/brainstorm/MnePython
• https://github.com/brainstorm-tools/brainstorm3/blob/master/toolbox/process/functions/process_mne_maxwell_py.m
• https://github.com/brainstorm-tools/brainstorm3/blob/master/toolbox/process/functions/process_fooof_py.m

However, setting up a Python environment is probably too complicated or time-consuming for many of the Brainstorm users working with Windows. In clinical contexts (100% Windows, and sometimes still Win7), it's sometimes difficult for people to use tools that are not fully graphical and plug-and-play.

I understand why it would be technically better for OpenMEEG to be interfaced in Python. But if you consider that a good share of OpenMEEG users are coming from Brainstorm, then this might not be the most appropriate choice, at least not until Python is available natively on all the OS.

If the future versions of OpenMEEG are Python-based only, I guess we'd have to stick to the old compiled versions for Brainstorm, at least for the standard users. For advanced users, we could write a Python wrapper to access the new version (I hope you'd help me write, document and maintain this code).

The most versatile solution would be: a C++ library + a Python wrapper + a Matlab wrapper... (all the arguments listed above are also valid for Matlab)

Possible interfacing options:

• https://fr.mathworks.com/help/matlab/cpp-mex-file-applications.html
• https://fr.mathworks.com/help/matlab/ref/loadlibrary.html
• https://fr.mathworks.com/help/matlab/call-cpp-library-functions.html

[agramfort]

I had the same feeling here. Python running in matlab is very much a niche setup and I doubt it will keep the user base of openmeeg stable in the matlab/brainstorm ecosystem. We need a Python wrapper in openmeeg but it should not become the only way to do things.

my 2c

Message ID: @.***>

[vhs304305]

Please spend time to fix the existing bugs and improve the documentation first. That would most likely increase and stabilize the user base.

[papadop]

@vhs304305: Which bugs are you referring to ? If you have specific bugs or documentation improvement suggestions, please feel issues....

[papadop]

Just for the record, I did a simple test with the simplest app: minverser. I have not fully tested it yet (I'm unsure of the type=argparse.FileType('w')), but with -h, it provides a nice output, and without the inner code, it gives a nice output.... Copy/pasting the code (replacing args with actual file names) seems to write a correct file).

This gives:

import sys
import time
import argparse
import openmeeg as om

parser = argparse.ArgumentParser(description='Compute the inverse of an head matrix.')
parser.add_argument('source_file', metavar='head_matrix', type=open, nargs=1, help='file name of the head matrix')
parser.add_argument('dest_file', metavar='inverse_head_matrix', type=argparse.FileType('w'), nargs=1, help='file name to store the inverse head matrix')

args = parser.parse_args()
print('-------------------------------------------')
print('Command line:')
print(*sys.argv)
print('-------------------------------------------')

start = time.time()

head_matrix = om.SymMatrix()
head_matrix.load(args.source_file)
head_matrix.invert()
head_matrix.save(args.dest_file)

end = time.time()

print('-------------------------------------------')
print('Elapsed Time: ', end - start, ' s.')
print('-------------------------------------------')