[JOSS] make functions run deterministic

[matbesancon]

Following the convention from Julia stdlib's Random and various packages from the JuliaStats ecosystem, every function which performs a random sampling should take in parameter the Random Number Generator (which is usually a subtype of Random.AbstractRNG)

[kirui93]

@matbesancon Is this really necessary for functions like rand() and randn()?

[matbesancon]

That's precisely for these functions. You would take a RNG in all functions and pass it to the rand and randn calls

[kirui93]

Yes. I have seen what you mean.
This means that I need to use the package Random for this.
Thank you. I will implement immediately.

[matbesancon]

Linking to the review: openjournals/joss-reviews#1912

[kirui93]

@matbesancon I will close this issue then because I have already included using Random RNG in the package through MersenneTwister.

[kirui93]

According to Julia's documentation, every function which performs a random sampling should take in a parameter the Random Number Generator (which is usually a subtype of Random.AbstractRNG. Also, random number generation uses the Mersenne Twister library via MersenneTwister objects. We employed this idea and included this functionality for all the parts of the code where we generate random numbers.

[matbesancon]

https://github.com/kirui93/ScenTrees.jl/blob/master/src/StochPaths.jl

This is not quite right, the rng as a global variable makes each run depend on the number of calls of a function. The rng should be an argument passed to functions instead

[kirui93]

https://docs.julialang.org/en/v1/stdlib/Random/

I used the concept in the above julia documentation. Another option was to use Random.seed!().

[kirui93]

Both MersenneTwister and Random.seed!() works in a similar manner. Consider the below,

julia> MersenneTwister(01102019) == Random.seed!(01102019)
true

[kirui93]

In https://github.com/kirui93/ScenTrees.jl/blob/master/src/StochPaths.jl,
rng = MersenneTwister(0101219)
and I have passed it as an argument of randn() in all the functions.

[kirui93]

Hello @matbesancon, do you suggest using as in the following example instead of the above?

julia> rand(rng::AbstractRNG, 3,2)
3×2 Array{Float64,2}:
 0.496169  0.449182 
 0.732     0.875096 
 0.299058  0.0462887

[matbesancon]

In any function that needs rng, it should be an argument of this function:

GaussianSamplePath1D()

# replace with 
GaussianSamplePath1D(rng)