Setting random seed in R

[matherealize]

I'm interested in using your package for simulations. However, at the moment it seems that setting a random seed requires a direct call to Julia.

In the readme it is shown to use "set.seed()" in R, but in my example this has no effect since the seed is not passed to Julia.

> bs = bigsimr::bigsimr_setup()
> set.seed(1)
> bs$rmvn(2, rep(0, 5), diag(1, 5))
          [,1]       [,2]       [,3]       [,4]       [,5]
[1,] 0.8107133  0.5801268 -0.9684956 -0.2439811  0.4141377
[2,] 0.1580521 -0.1109701 -0.3932944 -0.7755616 -0.5166628
> set.seed(1)
> bs$rmvn(2, rep(0, 5), diag(1, 5))
         [,1]      [,2]      [,3]       [,4]        [,5]
[1,] 1.027204  1.141811 -1.251099 0.08030231 -0.05882845
[2,] 1.694608 -1.659026 -1.812908 0.06768497  0.34145174

It DOES work if I set the seed in Julia myself:

> rnd = JuliaCall::julia_pkg_import("Random", "Random.seed!")
> rnd$`Random.seed!`(as.integer(1))
Julia Object of type MersenneTwister.
MersenneTwister(UInt32[0x00000001], Random.DSFMT.DSFMT_state(Int32[1749029653, 1072851681, 1610647787, 1072862326, 1841712345, 1073426746, -198061126, 1073322060, -156153802, 1073567984  …  1977574422, 1073209915, 278919868, 1072835605, 1290372147, 18858467, 1815133874, -1716870370, 382, 0]), [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0  …  0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], UInt128[0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000  …  0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000], 1002, 0)
> bs$rmvn(2, rep(0, 5), diag(1, 5))
           [,1]       [,2]       [,3]       [,4]       [,5]
[1,] -0.3170409 -0.5091629  0.1345132  1.0872089 -0.3866632
[2,] -0.4261330  0.1658788 -0.5986539 -0.9809587 -0.1185315
> rnd$`Random.seed!`(as.integer(1))
Julia Object of type MersenneTwister.
MersenneTwister(UInt32[0x00000001], Random.DSFMT.DSFMT_state(Int32[1749029653, 1072851681, 1610647787, 1072862326, 1841712345, 1073426746, -198061126, 1073322060, -156153802, 1073567984  …  1977574422, 1073209915, 278919868, 1072835605, 1290372147, 18858467, 1815133874, -1716870370, 382, 0]), [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0  …  0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], UInt128[0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000  …  0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000, 0x00000000000000000000000000000000], 1002, 0)
> bs$rmvn(2, rep(0, 5), diag(1, 5))
           [,1]       [,2]       [,3]       [,4]       [,5]
[1,] -0.3170409 -0.5091629  0.1345132  1.0872089 -0.3866632
[2,] -0.4261330  0.1658788 -0.5986539 -0.9809587 -0.1185315

It would be great if this could be updated / implemented. Maybe documenting this is already enough as well. Thank you for your work!

Version: 0.11.2 from CRAN
R: 4.1.1

[adknudson]

Hi @matherealize , thanks for opening the issue. You are correct that setting the seed requires a direct call to Julia. For now, I will update the readme to show the proper way to set the seed. Do note that it is probably still a good idea to set the seed in R at the same time to ensure reproducibility.

[matherealize]

I see, thank you for the quick clarification! I will close this for now.