-
Notifications
You must be signed in to change notification settings - Fork 23
Description
I am quite new to Turing, but I really like how it composes with basically any Julia library which is fantastic.
For hard problems, I think being able to customize sampling like in e.g. Gen.jl is the way to go. As far as I understand, this is the goal of AdvancedHMC and AdvancedMH, which are getting refactored out of Turing.
I am facing a difficult discrete problem where, currently, my only efficient solution is an implementation using Infer.NET. This library basically compiles programs to factor graphs. I would like to have a more high level implementation in Julia using Turing. I guess the way to go is to use AdvancedMH plus Turing. Would it be possible to get a little Turing example model with custom AdvancedMH sampling documented? I couldn't find any appropriate examples that included a Turing model.