Fix passing of proposals and bounds in the dynesty 3 api - #1117
Fix passing of proposals and bounds in the dynesty 3 api#1117JasperMartins wants to merge 5 commits into
Conversation
|
I'm unsure about the failing jax-backend test. I don't think it is related to this MR 🤔 |
ColmTalbot
left a comment
ColmTalbot
left a comment
There was a problem hiding this comment.
This looks good to me. Once we get the CI fixed, I think we can get this merged in.
5fc3625 to
458d749
Compare
ColmTalbot
left a comment
ColmTalbot
left a comment
There was a problem hiding this comment.
I think, actually, this has broken the default behaviour when not using the live-point sampling.
The docs job timing out is because the default bound for the uniform sampling is now no bound, which is horribly inefficient.
| f"estimated chain length. An average of {2 * internal_sampler.nact} " | ||
| f"steps will be accepted up to chain length {internal_sampler.maxmcmc}." | ||
| ) | ||
| elif kwargs["bound"] == "live": |
There was a problem hiding this comment.
I think we need to reinstate this clause.
There was a problem hiding this comment.
Alternatively, bound = "none" could be dropped entirely as a default keyword argument for the dynesty sampler. That would make "live-multi" (or whatever the upstream default bounding option is if it ever changes) the default option. If it is preferred that the behavior of "live" stays the default, I'll reinstate the clause, otherwise I'll push a fix that removes the default option entirely.
mj-will
added
sampling
4 participants
This MR fixes the issues discussed in #1065.
Closes #1065
Moved from ColmTalbot#4