General Improvements to stingray.modeling

[dhuppenkothen]

A bunch of things that I think could improve the modeling interface:

• make stingray.modeling compatible with emcee3 (issue emcee update #455)
• make sure stingray.modeling supports bounds (issue Bounds and parameter links in Maximum likelihood fitting #187 and issue Model bounds ignored #463); I suspect this might be an issue with the underlying scipy.optimize that I don't understand
• implement support for pymc3 and Hamiltonian Monte Carlo, which should make this much faster
• implement support for autodiff and more modern fitting methods (e.g variational inference)
• Move away from using astropy.models: using the astropy models keeps causing issues primarily for threads and multiprocessing, but also makes it harder to set up functions with logarithmic parameters and complex priors. I'd kind of like to move away from the astropy.models interface, but that requires some more thought.

[PrakharQ]

Hi,
I am currently a 4th year undergraduate at IIT Kanpur and planning to take part in GSoC 2021. I've done many courses involving signal processing and Bayesian inference. I would like to work on this project. Can you guide me from where I should start?