Douglas Bates

The mechanism for handling the GLM family and link functions is imported from the `GLM.jl` package so I would first look there to extend the available families. From `src/MixedModels.jl` ```julia...

Some timings on my laptop ```julia (jl_6i3bqi) pkg> st Status `/private/var/folders/d9/t94kd5m50rl1r3bchncn62sm0000gn/T/jl_6i3bqi/Project.toml` [ff71e718] MixedModels v4.34.0 julia> dat = MixedModels.dataset(:insteval) Arrow.Table with 73421 rows, 7 columns, and schema: :s String :d String...

There may be a way of cutting down on the cost of the `rankUpdate!` operation that evaluates `L[2,2] - L[2,1] * L[2,1]` , which accounts for the largest amount of...

@ajinkya-k I should have been more clear that I was speaking of the case where L[2,1] is sparse. Using RFP for L[2,2] only makes sense if L[2,2] is very large...

So I have a question about strategy. The bottleneck when using TriangularRFP for (usually) the [2,2] block of L which implies that the [2,1] block will be a SparseMatrixCSC is...

For the time being I will do the simplified indexing part and not do the reordering of levels part. However, we can recommend reordering the levels before creating the model.

I wrote the enclosed to do the reordering of levels in a `CategoricalArray` by incidence. (I'm not a big fan of `CategoricalArrays.jl` but it seemed the easiest way to accomplish...