Tyler Fairnington
Tyler Fairnington
While I can't help with the sampling speed, The problem of getting nonsensical outputs is (likely) due to these two factors: for your test notebooks you have a prior on...
@soichiro-hattori Thanks for the suggestion! I have tried both with and without and for multiwavelength fitting didn't notice a change in timing (a quick test earlier gives 34:32 for dense_mass=False,...
"And, not 100% sure but I think the long computational time is from sampling being slow and not a jit issue! If you diagnose your warmup you might see that...
Going to close this issue as most of this was performed on a CPU but using a strong GPU solves multi-thousand parameter fits in minutes, so the sampling speed is...
Not sure of the implications but just wanted you to be aware that there is a minor error in the 2018 papers that was noted in 2019 in regard to...
I'm also coming across this issue, thanks @hposborn for the current fix! I imagine the only workaround to this is if marginal calls something like pm.Potential(pm.logp(pm.MvNorm(
Hi @hposborn , I believe the solution to this is to indeed model the classical way of lc+GP = observed, but this is accessed through setting the GP mean as...
Just wondering if this PR is ready to be merged or if the power2 is implemented elsewhere, thanks!