Ben Cottier

I think the reason it doesn't work with `Pkg.test` in particular is that `Pkg.test` ultimately runs [this](https://github.com/JuliaLang/Pkg.jl/blob/e54b3dee952a07153728eb009bfed2c845523e8e/src/Operations.jl#L1421) to remove the julia base environment from `LOAD_PATH`. The reason I made an...

From the git blame of https://github.com/JuliaLang/IJulia.jl/blob/8710cd4bc1adabfacb8d9caec56332e36daa1f90/deps/kspec.jl#L97 I found https://github.com/JuliaLang/Pkg.jl/issues/1815/ which is a reason to not specify `--startup-file` 😕

Ah, I now realise the command to start the kernel is normally set one time in `installkernel`, so that makes it difficult to change the `startup-file` option.

IIRC, yes I was able to `nbconvert` within the Julia REPL. (Sorry, I'm no longer working with Julia so this issue is no longer relevant to me.)

Thanks for the suggestions! To clarify, is there currently no support (on `master` in any package) for tensor multiplication that preserves axiskeys? And is #6 (or similar) your preferred solution?

Just adding that this is also the case for `Diagonal`. ```julia julia> Diagonal(rand(3,)) \ KeyedArray(rand(3,), features=[:a, :b, :c]); ERROR: MethodError: \(::Diagonal{Float64, Vector{Float64}}, ::KeyedArray{Float64, 1, NamedDimsArray{(:features,), Float64, 1, Vector{Float64}}, Base.RefValue{Vector{Symbol}}}) is...

> The Julia 1.6 build behavior shouldn't have changed. If you remove the TimeZones package and re-add it can you reproduce the problem? Also, can you post your `versioninfo()`? Note...

Following https://github.com/invenia/AxisSets.jl/pull/44#discussion_r614292489: Ideally we would standardise how to handle `dim`s and `AxisSets.Pattern`s across Impute.jl, [FeatureTransforms.jl](https://github.com/invenia/FeatureTransforms.jl), and [AxisSets.jl](https://github.com/invenia/AxisSets.jl), in which the former two are both supported. I think the main difference...

I wonder should `FeatureTransforms.apply(M, p; dims=1, inds=1)` return the full array in the first place? I think that would solve it.

Note that even in my example implementation, performance degrades for large enough input: ``` julia> _periodic(sin, 7e16 + 7/4, 7) 0.0 ``` This might be a more fundamental issue that...