Closed segasai closed 3 years ago
Not a problem. I’ll look into this and see if I can find out what’s going on here. (These bounding distribution edge cases will be the end of me...)
Thanks. I am wondering honestly if the test suite is lacking challenging distributions. I.e. 100d Rosenbrock or funnel, or some almost degenerate high-D valley. I think that would help to catch those things. And if the running time is the issue, presumably individual tests can be scheduled as separate runs on travis, so each one is shorter than 1hr
You're right -- I don't have those tests in there primarily because of the long(er) run times. But given some of these problems, it looks like I should just bite the bullet and throw in a challenging high-D problem.
(not to be critical), but personally, for several challenging problems I had so far in the last couple of years, I had to always switch back to pymultinest, because of some issues with sampling/boundaries/efficiency in dynesty. So It would be good to have a hard distribution testing suite to ensure that everything works like it should.
No offense taken -- dynesty
was the first public package that I made when I was learning how to code properly, and that's led to problems that have come back to bite me ever since! Hopefully, with future software I will have better tests in place to catch these earlier. Sorry that the package has caused you frustration.
Just because I don't want to open yet another issue with this same error. I could provide the data and run setup if that would help, as well as a dump of the sampler after it crashes..
That'd be super. Thanks!
OK. This took long enough but I sent you an email with a script and instructions on how to replicate the errors I'm getting... Hope it helps you!
Hi,
while running the latest master version (with the 'rejecting multiellipsoid region' change), I'm getting this exception. I didn't have it before (with the previously released version). The error seems to occur exactly after the warnign about rejecting the multiellipsoid region.
I'm not sure I'll be able to provide a backtrace on this one, as it was a very long run.