Andreas Noack

@zaikunzhang I should add that to generate the output ```julia Dict{Float64, Int64} with 2 entries: 0.9329 => 483 0.9301 => 17 ``` that shows the two different solutions, I'm adding...

FWIW, here are the optimization traces for a good and a bad fit 

@palday I can see that you added them in https://github.com/JuliaStats/MixedModels.jl/commit/34899cf6d95abb2df8c961c9de5a6a500351ebff. What do they do? I tried to set them zero and it doesn't seem to make a difference.

> It is not a matter of converging to a saddle point as much as it is a matter of converging on the boundary. Ah. I didn't notice ealier that...

I don't remember if we have an issue for this but in general, there isn't a closed form solution to the truncated `mean` so this would have to be computed...

If somebody with access to SAS could try out the SAS macro, it would be a useful data point to know if the SAS macro fails in the same way...

Is there anything holding this one back at this point?

There is indeed a significant difference between direct package dependencies and test dependencies in that the latter can't break the environment alone, so having wrong compat (e.g. no) bounds on...

@DilumAluthge has anything changed with respect to > AutoMerge, doesn't currently require packages to have [compat] entries for test-only deps or is it status quo?

The new implementation stores the rows of the hat matrix (and a bit more) for each vertex when constructing the Loess fit. That requires much more memory than the old...