stateful iteration

[cscherrer]

Using Iterators.Stateful, we can rewrite our Sobol sequence interface as

export SobolHypercube

struct SobolHypercube{k} <: Hypercube{k}
    seq :: SobolSeq{k}
    value :: Vector{Float64}
    iter :: Iterators.Stateful{Vector{Float64}, Union{Nothing, Tuple{Float64, Int64}}}

    function SobolHypercube(k::Int)
        seq = SobolSeq(k)
        value = Sobol.next!(seq)
        return new{k}(seq, value, Iterators.Stateful(value))
    end
end

export next!

function next!(ω::SobolHypercube)
    Sobol.next!(ω.seq, ω.value)
    ω.iter.nextvalstate = iterate(ω.value)
    ω.iter.taken = 0
    return ω
end

Base.rand(ω::SobolHypercube) = popfirst!(ω.iter)

[cscherrer]

@ParadaCarleton @mschauer how does this look?

[codecov]

Codecov Report

Merging #15 (9fa3e43) into main (8725354) will decrease coverage by 2.12%.
The diff coverage is 43.75%.

@@            Coverage Diff             @@
##             main      #15      +/-   ##
==========================================
- Coverage   42.25%   40.13%   -2.13%     
==========================================
  Files           8        8              
  Lines         142      152      +10     
==========================================
+ Hits           60       61       +1     
- Misses         82       91       +9     

Impacted Files	Coverage Δ
src/hypercube.jl	29.03% <0.00%> (-11.88%)	⬇️
src/optional/Sobol.jl	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8725354...9fa3e43. Read the comment docs.

[cscherrer]

We can abstract this a little, then Sobol and GoldenSequences should be able to share the same code

[ParadaCarleton]

Looks good! Don't see any issues.

[mschauer]

Am I right to assume we only need ever infinite sequences (which perhaps repeat?). In that case I think we should implement our own version of stateful.

[cscherrer]

@mschauer I think I know what you mean, but maybe we should introduce some terminology to be safe. For both Sobol and GoldenSequences, the user selects a dimensionality at construction time (well, GS can do some other fancy things, but let's ignore those for now). If we ask for a k-dimensional sequence, let's call those k elements "inner", so "inner traversal" or "inner iteration".

So we generally step through these k elements, and then "step" or "reset" or next! (or whatever) the sequence. This loads the next k elements. Let's call this reloading an "outer" operation.

Ok, so with that I think the inner iteration will generally be finite (I think it has to be?), but the outer iteration will not.