Closed devmotion closed 3 years ago
It seems the gdemo
test is not very stable, it failed in https://github.com/TuringLang/AdvancedHMC.jl/runs/3077264457 as well. I check if it helps to increase the number of samples but maybe it would be better to tune some parameter of the HMC algorithm?
Oh wait, it seems the rng
is not passed to sample
, so actually the test is stochastic. Let's make it less random first.
Fixes deprecations of AbstractMCMCParallel and of MvNormal constructors with scalars and vectors as variance (https://github.com/JuliaStats/Distributions.jl/pull/1362).