search

joshspeagle / dynesty

Dynamic Nested Sampling package for computing Bayesian posteriors and evidences
https://dynesty.readthedocs.io/
MIT License
347 stars 76 forks source link

Removes the automatic reflective boundaries #151

Closed GregoryAshton closed 5 years ago

GregoryAshton commented 5 years ago

In testing using bilby, we found that reflective boundaries can result in odd behaviour at the boundary. We believe this is because the detailed balance of the MCMC steps are no longer satisfied (multiple ways exist to get to the same point). This removes the automated reflective boundaries for non-periodic parameters (but retains periodic boundaries).

GregoryAshton commented 5 years ago

@joshspeagle this was added (from what I recall) by @ColmTalbot . We have discussed this and think the reflective boundaries are problematic. One could imagine having "reflective", "periodic" and "None" boundaries, but I though initially we could just do the simplest thing and remove the problematic case.

joshspeagle commented 5 years ago

Okay, good to know. Happy to merge this in and revert back to the previous behavior.