Closed sylvaticus closed 1 year ago
Perhaps @tlienart would like to comment. I can't think of a reason to have that in __init__
and generally eval
in __init__
is a bad idea. Most MLJ implementations do not put this in __init__
.
on the top of my head I can't think of a reason either. Looking at https://github.com/JuliaAI/MLJScikitLearnInterface.jl/blob/dev/src/MLJScikitLearnInterface.jl there is an init
but it's to import sklearn, looking at https://github.com/JuliaAI/MLJDecisionTreeInterface.jl/blob/dev/src/MLJDecisionTreeInterface.jl there is a metadata_pkg
in the module but no init.
The latter (DecisionTree) can probably be used as a model by all devs. Long story short I think that it should be safe to remove it from init.
In BetaML I have, for reasons I can't recall (most probably just because I have copied it from some other package), the
MLJModelInterface.metadata_pkg()
function inside the__init__
function, but this seems to create some problems.Is it safe to just remove the MLJ model registration from
__init__
?