Model-based machine learning + Edward


The upcoming book Model-Based Machine Learning by John Winn and Chris Bishop presents machine learning in a model based, probabilistic framework.

The book contains in detail several examples (skill ranking, email filtering, time series modeling) in a Bayesian probabilistic framework.

All the examples are solved using message passing or belief propagation inference with Infer.NET

Could this be done with Edward? I think it would be very strong to extend the examples folder following the MBML book.

Your thoughts?


As with many things, it only requires someone to put in the time. It’s certainly possible with our current algorithms. That said, Kevin Murphy is preparing a new version of his textbook and is using Edward; Dave Blei is preparing a graphical model textbook (not released in a few years) that also uses Edward. It would be nice if someone (either Chris Bishop or any enthusiastic Edward user) to convert the examples.