Different program outputs under the same Random.seed!

General Usage
1

I have a Julia program that sometimes produces different results even if I set Random.seed!(1234). What could be the reasons?

Here is one possible reason I have found: In my program, I’m using a function c(x) from a package C I installed. This function also could produce slightly different results (say, a difference of 10^(-8)) for the same input x and the same Random.seed!. What I did was something like the following:

using C
Random.seed!(1234)
@show c(x)

Is it because Random.seed! can’t control the result of the function c? Or is it due to some other numerical reasons?

Thank you!

2

What is this package C? (I couldn’t find it anywhere)

There are many reasons why c(x) could be independent of the seed,
it could call internally some python, FORTRAN or C code, which is not controlled by the Julia seed.

Even pure Julia code could possibly generate it’s own random numbers internally. (However, I would expect that most packages don’t do that.)

So, without knowing more about the package C it is difficult to answer.

2 Likes
3

Pretty much anything, depending on what that package C does (race conditions in threads, calling external code, initializing it its own RNGs, …).

I would debug this by comparing results are various checkpoints in c.

That said, I don’t think it makes too much sense to worry too much about exact reproducibility of results using randomness; the fact that your numbers come from a particular RNG should not matter too much (if it does, the algorithm is problematic).

1 Like