Mlj.jl: Adopt and update XGBoost.jl wrappers

[ChrisRackauckas]

XGBoost.jl is a very good part of any ML library. The library is currently under-maintained, with a lot of issues installing on Windows. I was wondering if this project would consider adopting some of the maintenance of XGBoost.jl and help fix some of the build issues, ideally update it to use BinaryBuilder.

https://github.com/dmlc/XGBoost.jl/issues/64

[ViralBShah]

Perhaps @staticfloat might already have kicked off some BB work (but I am not fully sure).

[staticfloat]

Yes, status here: https://github.com/JuliaPackaging/Yggdrasil/issues/10#issuecomment-484678477

[ViralBShah]

I would love to add that usage of MLJ.jl is likely to greatly increase if we can use XGBoost.jl, given how popular it is.

[ablaom]

Thanks @ViralBShah for the spring cleaning at XGBoost.jl. I hope interest in maintenance can be revived.

BTW, MLJ already supports XGBoost:

julia> using MLJ

julia> models()["XGBoost"]
3-element Array{Any,1}:
 "XGBoostCount"     
 "XGBoostRegressor" 
 "XGBoostClassifier"

See also this example. The MLJ implementation is a wrap of XGBoost.jl.

If the MLJ team at Turing grows as hoped, we may be able to find someone to push maintenance of XGBoost.jl along.

[ViralBShah]

The immediate steps are to get it into BinaryBuilder and make the tests pass.

[aviks]

This should probably be closed now. [edit] XGBoost releases are now current, and most installation issues have been resolved thanks to jll artifacts.

[OkonSamuel]

cc @ablaom. can this be closed?