Closed akashgpt closed 3 years ago
It looks like you are running the 1.1 version, my suggestion is to try the one from github, it has so many various fixes (concerning slice sampling, and in general stability), that I think it should help. Also running 500k live-points is not a good idea. I doubt anything more than 10k is meaningful.
Thank you for the suggestion @segasai. I was indeed using dynesty v1.1. I am using the Github version now. Simulations are taking longer and only the 'rslice' simulation has finished so far (run time >15 hrs) but I have not gotten any of the errors I mentioned above, yet. I'll update here if that changes. Thanks again!
I am currently using dynesty for a 29-D problem (with pretty simple likelihoods and priors) but have been running into the following two issues (I have previously used dynesty successfully on a different problem but that had less than 10 dimensions) -
1) I get the "RuntimeError: Slice sampling appears to be stuck!" error with sample='auto'. I saw that someone else too pointed this out. Based on the discussion there I have been intentionally using more than enough 'live points' (or so I guess; =20k or 500k) but it is not helping, unfortunately. Here's how I call dynesty and the error I eventually get (similar error when running with 500k live points):
2) I also tried using sample="rslice" as you have suggested previously. In this case, I do get results when I assume lower accuracy for log-likelihood but not when I increase accuracy. Basically, I am trying to make 29 functions to go as close to '0' as possible and thus define a Gaussian likelihood for each with \mu=0 (the output likelihood is of course the log of the product of these 29 likelihoods). Then the \sigma of each of these Gaussians ends up defining how accurate the posteriors are/how close to '0' the 29 functions are. Lower the \sigma higher the accuracy (sorry if this is fairly obvious but thought of explicitly mentioning in case not). Here's how I call dynesty and the error I eventually get (similar error when running with 500k live points):
I get multiple instances of the following lines
and the following error eventually:
Please let me know if you might have any advice or comments on this (perhaps I am doing something really silly!). Thank you so much!