Closed xukai92 closed 1 month ago
Thanks @xukai92 -- it looks good to me.
@mhauru, you might want to take a look at this PR, too, since it is one of the most important packages in TuringLang. The changes in this PR are to prepare several follow-up PRs that implement Riemannian manifold HMC.
I accepted your suggestions. do you want to merge this PR then?
Some tests seem to be failing. Can you take a look?
Also, feel free to open multiple PRs in parallel. We’d like to get at least two reviews before merging.
Do I understand correctly that θ are the "position" parameters and r are the corresponding canonical momentum, and the idea here is to extend the interface so that the metric could depend on θ, although that is still unimplemented?
Yes, that is correct.
We can gradually address your other comments in future PRs.
extend the interface with position-independent fallback as default