Andreas Noack

I think it might require a little more work than just loosening the signature of `cholpred`. What is ```julia > PaddedView(1,X,(10000,4),(1,2)) |> t -> typeof(t't) ```

There are two differences between the R model and the`GLM.jl` model 1. R automatically excludes a variable. 2. R treats the `Per` variable as a factor. After changing these two,...

I agree with @palday here. In particular, I believe that age is a more meaningful variable here. It doesn't change the general issue though. Numerical rank determination is not clear-cut....

Looks right to me. Could you please add a test case as well?

@kleinschmidt Is it on purpose that the model matrix promotes to `Float64` even though the data columns are `Float32`? It's worth noticing that `lm(Matrix{Float32},Vector{Float32})` actually works so it's not `Float32`...

I think I have a fix for this. The precision can be improved in the inverse probit functions, in the calculation of the the Bernoulli variance and in the residual...

This is not completely fixed by #100 although the example above now works. The problem is only postponed until the separation becomes more severe. In the estimation weight calculation, there...

I think Tables.jl might be more general than needed for `GlmResp`. I like https://github.com/piever/StructArrays.jl which is conceptually a bit simpler and would give us row iteration.

I think it would be nice if we could leave missing values out of this package. I don't know if it is practical but I think we should try to...

I think this is one of the relatively rare cases where you'd need the extra precision that the QR factorization provides relative to the Cholesky. The functionality for using QR...