Example notebook: Gaussian process priors

[st--]

No description provided.

[devmotion]

Some high-level comment/idea: Since this is not how we want users to create or work with Gaussian processes, I think it would be nice to have to both the "naive"/basic KernelFunctions approach and the AbstractGP-based implementation together in an example to avoid any incorrect impressions and so that users know how they should do it. However, if both AbstractGPs and KernelFunctions are part of the example, I think the example fits better into either the AbstractGPs repo or the JuliaGaussianProcesses webpage. Otherwise we use packages that depend on KernelFunctions and hence introduce a circular dependency that makes it difficult to update the example without delay if there are breaking changes in KernelFunctions.

[st--]

How about leaving this example in KernelFunctions but clarifying all you're saying in an intro paragraph and linking to the corresponding "proper" example in AbstractGPs docs?

[devmotion]

This should definitely be done if the example is added to KernelFunctions. My worry is that people do not notice and read the paragraph and just copy whatever code there is on this page about Gaussian processes. I assume this is less of a problem with the "correct" code is on the same page as well.

[st--]

My worry is that people do not notice and read the paragraph and just copy whatever code there is on this page about Gaussian processes.

It might be helpful if you expand on what you're worried about so we can discuss likelihood, impact, and mitigations:)?

[devmotion]

I am worried that they copy and use the code but are not aware of how one is supposed to work with Gaussian processes.

[willtebbutt]

In my opinion it would be sufficient to add one of those hint things that you can add in documenter saying something along the lines of "this is an illustrative example, you should probably be using an AbstractGP of some kind or another if you actually want to do something with a GP"

[devmotion]

On the other hand, we can always link to AbstractGPs or the JuliaGaussianProcesses webpage from the documentation, so the example would still be discoverable from the KernelFunctions docs even if it is part of AbstractGPs or the webpage.

[willtebbutt]

LGTM. Just a quick clarification.