search

joshspeagle / dynesty

Dynamic Nested Sampling package for computing Bayesian posteriors and evidences
https://dynesty.readthedocs.io/
MIT License
341 stars 75 forks source link

The loglstar is always -inf<-inf<inf #447

Closed Sunnnsh closed 11 months ago

Sunnnsh commented 1 year ago

Hi, I currently using dynesty to do some sample and I find that the loglstar is always -inf<-inf<inf. I checked the log-likelihood and its value is around 300000~50000000. Is this issue occur because my log-likelihood value is too large? Or how can I solve this issue?

Thanks!

segasai commented 1 year ago

Hi,

I am afraid you don't provide enough information to assess if there is any issue at all. Also there is a question in FAQ on a related issue https://dynesty.readthedocs.io/en/stable/faq.html#sampling-questions

If you think there is an issue, please provide enough information to understand what you do, what you see and why/if you think something is wrong.

Sunnnsh commented 1 year ago

Sorry for less information about my issue,

I am using dynesty for some analysis related to gravitational waves. I have set up the prior ranges for sampling, and I followed the steps outlined in the documentation to start the sampling process. However, I noticed that in the output information from dynesty, the value of loglstar is always -inf < -inf < inf, which suggests that dynesty is obtaining -inf during the computation of log-likelihood. I have checked my log-likelihood within the prior ranges, and its values range from approximately -300,000 to -50,000,000. Could it be that the magnitude of my log-likelihood values is causing dynesty to evaluate it as -inf during the computation process?

Thanks!

segasai commented 11 months ago

Sorry for delayed reply, but it seems that this specific question is addressed in FAQ "There are inf values in my lower/upper log-likelihood bounds! Should I be concerned?"