Generating matrix random variables in Multi-Threaded mode with Distributions


I now understand (from Multi-threading) that I need to generate random seeds for each thread to avoid race conditions if I wish to speed up the generation of independent random variables. One good and straightforward way of doing this is the following code snippet from

using Random
import Future

rseed = let m = MersenneTwister(1)
    [m; accumulate(Future.randjump, fill(big(10)^20, Threads.nthreads()-1), init=m)]

The problem is that I wish to generate Inverse Wishart random variables from the Distributions package but find that I’m unable to do so because of the way the related InverseWishart() function and its extensions are defined. Executing

rand( rseed[1], InverseWishart( 5, [1.0 0; 0 1] ) )

will return

ERROR: ArgumentError: Sampler for this object is not defined

Wonder if anyone would be able to guide me to a workaround? I tried looking the base codes of the function InverseWishart from
to try and work something out but struggled to understand and meaningfully manipulate their definitions to my favour.

1 Like


This works for me on Julia v1.0.3 and v1.1.0 with Distributions v0.17.0.

What versions are you using?

1 Like


It does work now @greg_plowman.

I was using Distributions v0.16.4.

Thanks for the heads up.