Penelope Yong

I'm very sorry - I don't think I have the knowledge to review this. Happy to try to look at specific questions if you have any though.

For context, here are the method ambiguities: https://github.com/TuringLang/Turing.jl/actions/runs/10161073643/job/28098796083?pr=2290#step:8:305 They arise from three functions: 1. `tilde_assume`; this should have been fixed in https://github.com/TuringLang/DynamicPPL.jl/pull/636 and https://github.com/TuringLang/Turing.jl/pull/2299 2. `bundle_samples`; these are fixed in...

Having mulled it over a bit, I think my preferred option would just be to swallow the breaking change and bump the minor version, as otherwise we will end up...

OrderedDict for pointwise_loglikelihoods was introduced in DynamicPPL 0.22.4 (https://github.com/TuringLang/DynamicPPL.jl/pull/475), and the compat bounds for DynamicPPL here go all the way down to 0.21, so I guess technically the sorting is...

(And thank you, @zeyus!)

I think this was probably fixed by https://github.com/TuringLang/docs/pull/426.

#885 will solve this (once merged to main) as prior and likelihood are cleanly separated.

It's `@addloglikelihood (; logprior=0.0, loglikelihood=mylogprob())`. Besides, just `@addlogprob mylogprob()` will add to the log-likelihood, so in fact it does directly implement your original suggestion. 0.37 of DynamicPPL will be quite...

Just as a small update, I've been working (mainly for my own benefit!) on implementing Metropolis–Hastings, and writing it up as I go along. I'm currently hosting this on my...

ps. since this issue is so wide-ranging, and covers both the sampling side of things as well as the probabilistic DSL, I'm going to move this to the docs repo....