Closed sdvillal closed 9 years ago
Data transfer back to python world is really slow and not numpy aware. Until it improves, this is not really an option.
But at the moment it is the best option for code evaluation, faster than matlab_wrapper, more robust than pymatbridge, provides the best experience for error reporting and can be nicely combined with file-based data transfer. Install cannot be automated and it is only matlab 2014b+, but we will stick with it while still supporting matlab_wrapper and pymatbridge.
Our current solution to pimping matlab uses pymatbridge for code dispatch and oct2py for data transfer. How does it compare with the new (R2014b) matlab <-> python interface? This new feature allows both:
In case it would be competitive (handle properly type conversion, is numpy aware, make faster in-memory transfers, control better memory de-allocation, it is not too buggy, it is not a pain to debug...) we could create a new MatlabEngine implementation using it.
It seems to me that this is not a separate toolbox.