Add better support for time-series models [GSoC]

My name is Rohit Sanjay, I am a third year undergraduate student studying electronics and communication engineering at Manipal Institute of Technology, Manipal, India.

I have opted for a minor in computational mathematics, and as part of this minor, I have a course called Applied Statistics and Time Series Analysis.

I have previously contributed to projects like Pandas and Numba.

I would like to work on this project and add the suggested time series models - ARIMA and Prophet among many others, which I will add later.


Hello, rohit. Great idea for future development! Time-series support in Pymc3/4 has a lot of things to work on, and adding some good foundation by implementing well the models you listed above would be a very significant step forward.

To start, I suggest you look at a current attempt of building prophet in Pymc3. Luke seems to have done a decent job with it. Also, I would suggest looking at state-space modelling (afaik state-space modelling is the most rigorous basis for flexible time-series), and this for some practical tips.

I am sure mentors will help you out with finer guidance, but make sure to ask any questions here as well in the forum!