David Widmann
David Widmann
I think on the master branch we could do e.g. ```julia julia> function Distributions._rand!(rng::Distributions.AbstractRNG, d::Distributions.VectorOfUnivariateDistribution, x::AbstractMatrix{ rand(d, 10) 3×10 Matrix{Float64}: 0.397305 0.155416 1.69355 0.284072 1.55921 2.1799 1.26069 2.30843 0.153868 1.69613...
Did you compare it with https://github.com/JuliaStats/Distributions.jl/pull/1908?
It hasn't been approved yet.
This will break Julia < 1.9. It's fine if it's listed in `deps`, in newer Julia versions >= 1.9 it will still be only a weak dependency since it's listed...
Wasn't SnoopPrecompile designed for this precompilation scenario?
In SciML we discussed adding Preferences-based precompilation statements to allow users to switch on/off precompilation in a somewhat granular way, depending on their use cases.
Requires a new release of MLJBase
I like this a lot. I wonder though if it would be better to let `check_model` only return a Boolean. I think this might be simpler and less confusing for...
> Maybe do a check_model_and_extras (not the best name) or something? And then have check_model just wrap this? Yes, that would be an option. An alternative could be `check_model(; return_trace...
Yeah, it's not something I'd generally do. But I have the feeling it's somewhat OKish for functions that are a) only/mainly used in the REPL and b) expected to be...