Anthony Blaom, PhD

Also, we want the actual model traits to reflect those saved in the model registry, which is created by using a Project.toml with all the packages in it.

My sense is that the degree of code complexity you need to add to make MLJFlux.jl work is not worth any extra functionality you buy into, and so interfacing MLJ...

To clarify an earlier point regarding the form of predictions. It is not absolutely necessary that predictions be distributions. I noticed that prediction is VERY slow at present, so am...

FYI This timing is fairly reproducible on my machine: ```julia X = MLJBase.table(rand(Float32, 100, 3)); y = coerce(rand("abc", 100), Multiclass); model = LaplaceClassification(); fitresult, _, _ = MLJBase.fit(model, 0, X,...

Thanks @tlienart for looking into this. This looks completely generic. That is, we could apply this idea to any supervised model, not just the linear ones in this package. Do...

Okay, there is still a problem. For some reason the vectors, although consisting now of `Float32` are still wrapped as `AbstractFloat`: ```julia julia> X.Column2 5-element Vector{AbstractFloat}: 1.0f0 2.0f0 3.0f0 4.0f0...

If you need "my trick" you are already doing something suboptimal. I didn't check this, but maybe the problem is not `recode` at all but that your dictionary (or `Pair`...

Okay, thanks. I think easiest is for me to separately fix the dict type in #276, which I've not merged yet.

Although resolved, I'm re-opening this issue because the test suite currently lacks a catch for re-occurrence and this already led to an issue at #281.

Not sure I understand the question. I am proposing that new tests be added to the test suite to catch errors like the one encountered in the opening post. Nothing...