I think it’s generally fine.
Explicitly passing an RNG to sim code is useful is you want:
-
Reproducibility
-
Option of using a different type of RNG (e.g. GitHub - JuliaRandom/RandomNumbers.jl: Random Number Generators for the Julia Language.)
2 Likes
I’d also add testing (which is kind of related to reproducibility): StableRNGs.jl is very useful to test packages across multiple minor versions of Julia, since the default RNG can (and did) change the stream of generated numbers from version to version. Having a mechanism to easily pass the RNG is very important.
4 Likes
The other option is to wait until 1.7 and hope https://github.com/JuliaLang/julia/pull/34852 gets in. If it does, (and I understand it correctly) it should fix the speed of rand() without passing the rng (and be faster since it uses a faster rng also)
1 Like
+1 , I’m using that already.
FYI, I just added a section discussing a couple of approaches for using RNG in FLoops.jl documentation: How to do X in parallel? · FLoops. Some approaches are generic (don’t need a particular package) and one requires something like FLoops.jl (esp. a new release https://github.com/JuliaRegistries/General/pull/31964).
6 Likes