Machine learning

Current implementation of Go binding can not specify options.

GPUOptions struct is in internal package. And go generate doesn't work for protobuf directory. So we can't specify GPUOptions for NewSession.

🚀 Feature

Support to pickle a jitted function (or at least throw a TypeError when using protocol 0 and 1).

Motivation

Trying to pickle a jitted function either raises TypeError: cannot pickle 'torch._C.ScriptFunction' object when protocol>1 or far worse when using protocol=0 or protocol=1 python 3.9.5 dies with:

terminate called after throwing an instance of 'std::runt

We should be using pkg_resources (or importlib.resources if our min Python version is 3.7) instead of uses of __file__.

$ get grep '__file__' sklearn/
sklearn/__check_build/__init__.py:    local_dir = os.path.split(__file__)[0]
sklearn/datasets/_base.py:    module_path = dirname(__file__)
sklearn/datasets/_base.py:    module_path = dirname(__file__)
sklearn/datasets/_base.py:    

I received this error message:

  MethodError: convert(::Type{Union{}}, ::Vector{Float32}) is ambiguous. Candidates:
    convert(::Type{T}, a::AbstractArray) where T<:Array in Base at array.jl:532
    convert(T::Type{var"#s79"} where var"#s79"<:BitArray, a::AbstractArray) in Base at bitarray.jl:577
    convert(T::Type{var"#s832"} where var"#s832"<:SparseArrays.SparseVector, m::AbstractVect

Currently many more Python projects like dask and optuna are using Python type hints. With the Python package of xgboost gaining more and more features, we should also adopt mypy as a safe guard against some type errors and for better code documentation.