Matthieu Gomez

Btw I think the slowdown comes from missing_omit that creates a new namedtuple type depending on variables in the formula.

I think it’s still about specialization — it’s just that everything after missing_omit is respecialized to the new dataset. Yes I think the way forward would be to write missing_omit...

cf https://github.com/JuliaData/TableOperations.jl/issues/7

Special casing would be great. One tiny drawback is that `categorical` is a bit verbose for something that common (in Stata, one can simply write `i.x`), but that's a really...

No it does not (if I understand your question correctly). For now, using `categorical` in the formula fails because the package tries to apply the function elementwise. ```julia using DataFrames,...

Still, the issue is what happens when the variable is `Vector{Union{

Note that for now ```julia using StatsModels N = 10_000_000 x = rand(N) + rand([0, missing], N) df = (x = x,) schema(Term(:x), df) #julia(20271,0x116991dc0) malloc: can't allocate region #:***...

yes, just connecting it to https://github.com/FixedEffects/FixedEffectModels.jl/issues/99

I agree with this. A related suggestion is that the RHS of the formula could always return a tuple of terms no matter the number of terms. The current situation...

I also think the current situation is confusing and I would rather have any of the solution mentioned in the thread.