David Widmann

Another quick debugging yields that the problem is that `eltype(vals) = Any` in https://github.com/TuringLang/Turing.jl/blob/9dc8a297aaed4e898beba698491116c28e40c533/src/inference/Inference.jl#L339.

An ugly fix is to replace https://github.com/TuringLang/Turing.jl/blob/9dc8a297aaed4e898beba698491116c28e40c533/src/inference/Inference.jl#L263 with ```julia return names, map(identity, vals) ``` Then `eltype(vals) = Real` at least, which is sufficient for `Chains`. However, this is not a...

Hmm maybe we don't actually want to convert the integers that we obtain for, e.g., `k` to floating point numbers I guess? Then `eltype(vals) = Real` would be fine.

The following seems to solve the issue but promotes everything to a common type: ```julia function _params_to_array(ts::Vector) # obtain set of all sampled parameters names = Set{Symbol}() for t in...

I'll put together a PR.

I completely missed that one has to consider nested non-scalar parameters as well. This seems to make it a bit more difficult, so I have to see if a similar...

Reminds me of https://github.com/TuringLang/IRTracker.jl/blob/master/src/graphviz.jl.

A quick initial comment without having studied your code in detail: It seems the display of the chain errors but not sampling itself, so you can work around the problem...

Another comment: Turing is not compatible with Julia 1.7 yet (did you install Libtask_jll manually to work around the dependency issues?). The main problem though is that you only create...

> I think in the Julia version most time is spent on indexing op. In C++, these ops are single function call (and maybe inlined), while in the Julia version,...