Turing.jl now support passing RNG to sample as a keyword argument. Although, all the randome process in AdvancedHMC.jl are correctly controlled by the passed RNG, the initialization, from Turing.jl, is not: https://github.com/TuringLang/Turing.jl/blob/master/src/inference/hmc.jl#L267. What's the best way to pass RNG to this?
xukai92
All 7 comments
In the future, which I know is not super helpful right now, the sample interface will handle the rng argument in a first-class way. The file I linked below basically forces all samplers to accept rng as the first argument, though as far as I am aware only the HMC samplers use it. The particle samplers, IS, and MH all only use the argument for dispatch and discard the actual value.
cpfiffer
on 28 Jul 2019
This should be easy to fix once your PR is ready.
trappmartin
on 31 Jul 2019
@trappmartin do you mean it's better for me to leave this issue aside in the current PR and resolve it later?
xukai92
on 31 Jul 2019
Yeah, I think that's wise.
cpfiffer
on 1 Aug 2019
871 is merged.
xukai92
on 1 Aug 2019
Can we close this?
trappmartin
on 12 Sep 2019
xukai92
on 12 Sep 2019
Related issues
hessammehr
路
4Comments
mateuszbaran
路
5Comments
scheidan
路
5Comments
Vaibhavdixit02
路
4Comments
trappmartin
路
3Comments
Most helpful comment
Yeah, I think that's wise.