Hong Ge (are you on Disourse?) recently posted this comment in the probprog
channel on Slack:
…I think it is quite helpful to consider carefully how we, as a community, can work together to build a better platform for probabilistic programming and (Bayesian) inference. For example, within the Turing team, we made an effort to create a shared interface for MCMC sampling methods. It allowed us to decouple inference algorithms from the Turing DSL. I found this approach very helpful since it compartmentalises complexity between various components such as the DSL and HMC implementation – separating Turing’s HMC code into its package made maintaining and improving HMC algorithms much more comfortable. Extrapolating form these experiences, what might be interesting to consider as a community, is to co-develop a shared base library
AbstractPPL
for various approaches to probabilistic programming such as tracing and source rewriting. It would allow different PPLs to talk to each other, and share some codebase in some instances.
I think there’s a lot of potential in this idea, but it will take lots of discussion to work through.
The comment will scroll away soon, so I’m capturing it here so we don’t lose it.