Stan cannot fit models with discrete parameters, so every model in Stan is either naturally continuous or marginalized. If we want to use this to test samplers, it might be useful to write up a smaller subdirectory src/extension_models or something which contains models that have discrete components, to test gibbs sampling, smc, automatic marginalization, and things like that.
Gibbs sampling with a combination of discrete and continuous parameters, even though it works, is notoriously difficult to get working for anything slightly interesting. Not saying there aren't examples we should be putting in there, but probably pretty far down the priority queue:)
SMC in Turing.jl, though functional, is very, very slow due to how we have to implement it.
Automatic marginalization is not something we support in Turing.jl.
With all that being said, it's technically possible to achieve these things in Turing.jl, so might be things to consider for the faaaar off future:)
Stan cannot fit models with discrete parameters, so every model in Stan is either naturally continuous or marginalized. If we want to use this to test samplers, it might be useful to write up a smaller subdirectory
src/extension_models
or something which contains models that have discrete components, to test gibbs sampling, smc, automatic marginalization, and things like that.