I saw the steps for a Bayesian estimation algorithm in a paper, and I was wondering if there’s a way to implement this kind of hybrid MCMC algorithm using the Turing.jl library.
Maybe you have already explored the docs, but just looking at the words mentioned, I wonder if it could be possible using SSMProblems.jl / HiddenMarkovModels.jl, AdvancedMH.jl, Turing.Inference.Gibbs. If it requires a more custom sampler, the external samplers page talks about how to connect such things with Turing. Note also the Gen.jl PPL which is designed for finer control of inference.
The algorithm appears highly customized to a degree that I feel that there is no reason to rely on Turing for this?
1 Like