Brian Chen

NNlib is used in a couple of places in https://github.com/JuliaML/MLUtils.jl/blob/main/src/utils.jl, but I don't think those would be too difficult to change or vendor the functions used.

The biggest blocker is still what to use in place of `NNlib.scatter` for https://github.com/JuliaML/MLUtils.jl/blob/09c87f7097536384cea0a132aa0012679df18175/src/utils.jl#L201. Vendoring `scatter` won't help since it depends on KernelAbstractions. It'd also be worth redoing the import...

This could really use a MWE. Without code I'm not sure an answer will be forthcoming.

How much can we reduce import times without removing core deps (FLoops can probably go though since it's mostly used for syntax sugar)? I don't think this package has any...

What about precompilation? I don't think any packages in JuliaML use it?

I thought it was part of https://github.com/SciML/DifferentialEquations.jl/issues/786, but reading back through https://github.com/SciML/DifferentialEquations.jl/issues/786#issuecomment-1000786329 it appears invalidations + Requires were more important for import times. Has anyone tried running `@time_imports` from 1.9...

Some finer-grained timing locally suggests that the majority of import overhead for BangBang and FoldsThreads is self time and not from dependencies (unless there is significant invalidation happening, but that...

Part of the motivation is that APIs like https://juliaml.github.io/TableTransforms.jl/stable/transforms/builtin/#TableTransforms.Scale and https://invenia.github.io/FeatureTransforms.jl/stable/api/#FeatureTransforms.StandardScaling already exist. It doesn't make a lot of sense to reinvent the wheel there, especially since you'd have to...

Good idea! Added the relevant label so this doesn't go forgotten.

Are there any communication channels open between JuliaML and JuliaStats? MLBase has essentially seen 0 development since 2018, so it would be good to know if anyone still has plans...