[Neuroimaging] Using Python libs from MATLAB

[Matthew Brett]

Hi,

On Mon, Jun 13, 2016 at 7:56 AM, Marmaduke Woodman
<marmaduke.woodman at univ-amu.fr> wrote:
> hi all,
>
> I'm a dev behind the Virtual Brain (TVB) a neuroimaging simulation library,
> written in Python. As part of an effort to reach out to some users stuck in
> MATLAB, I've been testing MATLAB's recent *official* support for Python
> (2014b+). It is notably better than the many other attempts because it
> appears to use Py's C API to expose objects and methods as MATLAB objects
> and methods instead of just providing eval/exec.
>
> Generally it's working, but there are some workarounds to avoid segfaults,
> errors and lack of stdout/err. I've started to collect a series of tips and
> workarounds and figured there might be common interest in the community for
> a "helper" library so all our projects can benefit from a larger user base,
> and maybe ease the transition for potential Python users.
>
> The most important in order,
>
> - Anything touching HDF5 must use exactly the same version as MATLAB or
> segfault

Ouch - I guess this means it's not possible to use h5py or pytables?

> - Scipy.io.{save,load}mat segfault

That's interesting - it's probably possible to fix that, because that
stuff doesn't call any external libraries.  Do you have any more
details?

> - C++ extension modules whose exception model doesn't match MATLAB's,
> segfault on exception
> - Linalg segfault unless using MKL (which is what MATLAB uses)
> - stdout/err and logging must be redirecting to mexPrintf via a ctypes
> monkey patch
> - Numpy/MATLAB array conversion is O(1) op, but can probably be made zero
> copy with a ctypes + lib mx workaround
>
> Aside from these (for which I sent a service request to MathWorks), it's
> working well and we (TVB) expect our next release to include demo scripts in
> MATLAB.
>
> So, again, if there is common interest in collecting notes and a helper
> library to monkey patch around the segfaults, it'd be great to not do this
> alone ;)

Thanks too for the feedback.   You might also get some interest over
at the numpy / scipy lists.  Although the blas / lapack thing is a bit
of a downer...

Cheers,

Matthew