Update aliastable.jl

[githubtomtom]

enable rand() from non-Float64 mixtures, e.g. MixtureModel{Univariate,Continuous,Normal,Float32}, now only Float64 element type is supported.
a related pull request is made in StatsBase.make_alias_table!().

[codecov-io]

Codecov Report

Merging #1074 into master will not change coverage.
The diff coverage is 100%.

@@           Coverage Diff           @@
##           master    #1074   +/-   ##
=======================================
  Coverage   80.62%   80.62%           
=======================================
  Files         113      113           
  Lines        5611     5611           
=======================================
  Hits         4524     4524           
  Misses       1087     1087

Impacted Files	Coverage Δ
src/samplers/aliastable.jl	84.61% <100%> (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 d521695...c9b38c9. Read the comment docs.

[matbesancon]

Thanks for the PR! Two points:
Should we wait for the PR in StatsBase to be merged?
Could you add some tests with non-Float64 mixtures to ensure everything is working fine?

[githubtomtom]

Should we wait for the PR in StatsBase to be merged?

no response from StatsBase so far... it's my first time PR there ... do u know the expected turnaround time of them?

Could you add some tests with non-Float64 mixtures to ensure everything is working fine?

I tried, but failed. Because I could not modify AliasTable{S} to AliasTable{S, T} :(

[st--]

ping! is there anything holding back this PR? (other than perhaps the corresponding PR in StatsBase, JuliaStats/StatsBase.jl#559 JuliaStats/StatsBase.jl#499, also pinged that one...)

[devmotion]

IIRC we have to wait for the StatsBase PR.

[devmotion]

This also needs tests.

[devmotion]

With StatsBase 0.33.12, sampling from mixture models with component weights of type Float32 works fine (it errors with StatsBase 0.33.11 and was fixed by JuliaStats/StatsBase.jl#499). E.g.,

julia> using Distributions

julia> rand(MixtureModel([Normal(), Normal()], [0.5f0, 0.5f0]), 100)
100-element Vector{Float64}:
...

I'll close this PR since the issue that it tries to solve seems to be fixed by the change in StatsBase.