Problem with random generator

sergevic · April 10, 2020, 3:06am

I’m completely puzzled with a problem concerning my code:

using LinearAlgebra
using DeterminantalPointProcesses

L=Symmetric(rand(100,100))
variables=collect(1:100)
dpp=DPP(eigen(L))
M0=rand(dpp,1)
function proposal(Mt::Array{Int64,1})
    compl=setdiff(variables,Mt)
    Lcompl=L[compl,compl]
    dpp_compl=DPP(eigen(Lcompl))
    M1=compl[rand(dpp_compl,1)[1]]
    return M1
end

When I run map(x->proposal(M0[1]),1:3) I get

3-element Array{Array{Int64,1},1}:
 [2, 3, 4, 5, 6, 8, 10, 11, 12, 13  …  84, 87, 88, 89, 93, 94, 97, 98, 99, 100]
 [2, 3, 4, 5, 6, 8, 10, 11, 12, 13  …  84, 87, 88, 89, 93, 94, 97, 98, 99, 100]
 [2, 3, 4, 5, 6, 8, 10, 11, 12, 13  …  84, 87, 88, 89, 93, 94, 97, 98, 99, 100]

but I’m supposed to obtain 3 different arrays, because of the random command rand(dpp_compl,1)[1] .
When I try to run

Mt=M0
compl=setdiff(variables,Mt[1])
Lcompl=L[compl,compl]
dpp_compl=DPP(eigen(Lcompl))
compl[rand(dpp_compl,1)[1]]
compl[rand(dpp_compl,1)[1]]
compl[rand(dpp_compl,1)[1]]

I obtain 3 different arrays and that’s what I should obtain. I’ completely puzzled because the first code should produce different arrays just like the second code! Has anyone an idea of what’s wrong??

dpsanders · April 10, 2020, 4:44am

It is returning a Vector of Vectors, i.e. each component of the result is one of the Vectors that you wanted.

Oscar_Smith · April 10, 2020, 4:48am

I think the complaint is that the three entries are the same.

rfourquet · April 10, 2020, 4:55am

I didn’t run the code, but it looks likely to come from the fact that in your mapcommand, the function returns a constant (i.e. doesn’t depend on its argument x.

WschW · April 10, 2020, 5:19am

Each time you call DPP the DeterminantalPointProcesses it creates contains a MersenneTwister created with the same seed.

github.com

alshedivat/DeterminantalPointProcesses.jl/blob/80ffff8c9e33b1cfd43fcfe7379f8ffd80ba7240/src/types.jl#L12


      
          abstract type PointProcess end;
          
          

          
# specific types
          mutable struct DeterminantalPointProcess <: PointProcess
              L::Symmetric
              Lfact::Eigen
              size::Int
              rng::AbstractRNG
          
          
    function DeterminantalPointProcess(L::Symmetric; seed::Int=42)
                  Lfact = eigen(L)
                  new(L, Lfact, length(Lfact.values), MersenneTwister(seed))
              end
          
          
    function DeterminantalPointProcess(Lfact::Eigen; seed::Int=42)
                  L = Symmetric((Lfact.vectors .* Lfact.values') * Lfact.vectors')
                  new(L, Lfact, length(Lfact.values), MersenneTwister(seed))
              end
          end

Since the rand method provided by the package uses that MersenneTwister you will always get the same sequence if you create the DeterminantalPointProcesses without setting the seed to a different number each time you create one.

sergevic · April 10, 2020, 5:00pm

The problem is that the three vectors are the same when they don’t have to be!

sergevic · April 10, 2020, 5:01pm

Yes exactly!

sergevic · April 10, 2020, 5:01pm

The argument x is just here to settle the numer of times I repeat the function proposal.

sergevic · April 10, 2020, 5:04pm

Ahhhhhhh yes! Didn’t see that the seed was inside the rand command, since usually I set a seed myself outside! That should do the trick, then!