st--

It'd be good to make clear to readers/users what the limitations/assumptions of the examples are, so when they try it out they know that bad performance might be due to...

@IanButterworth yes it'd be great if `@deprecate` had a more explicit warning against usage as you suggest. That's unlikely to happen anytime soon, so how about the following instead? Add...

> > That's unlikely to happen anytime soon > > Why? If people like the idea we could add the feature for 1.8 Well, "soon" is relative - extrapolating from...

Hi, was there any decision as to what's actually going to happen? As a user who would like to use FillArrays's structured arrays as well as autodiff through those arrays,...

> I just checked and UniformScaling is supported and can be used at least with some methods (mean_and_var errors since diagind is not defined): Could we not make that non-breaking...

> Could we not make that non-breaking by also having a constructor that converts the UniformScaling into an appropriately-sized ScalMat? Looks like we can, @devmotion I added the example you...

See https://github.com/JuliaGaussianProcesses/ApproximateGPs.jl/pull/126 for the required AD fix

Ah good point :|

Maybe an alternative would be just to show train/test curves to demonstrate that you need to do _something_.

A Haskell coder would say that's just how functions work :joy: x -> f -> lik But I do agree it's a bit cumbersome here. What was your motivation for...