A scikit-learn-like interface for BayesFusion's SMILE.
BayesFusion's SMILE engine is one of the most highly respected packages for Bayesian Modelling and offers a limited-in-time academic license.
The package is written in C/C++ and is closed-source, but headers are available.
This project is all about creating a convenient, python-based, front-end for the
package, inspired by scikit-learn standards.
There's two approaches being tried here. One relies on command-line interaction with custom made executables, where communication goes over disk. This is of course slow and ugly, but also supposed to be very stable.
The other, more challeging approach was to design an actual C-binding to the
C++ library, and then use CTypes (using cython) to interact with the resulting .so direclty.
This works to some extent, but is less stable at times. And, crucially, very
difficult to debug.