Pairplot for MCMCChains

[yebai]

We should mention

https://github.com/sefffal/PairPlots.jl

which is quite helpful for understanding MCMC convergence issues.

cc @sefffal do you want to contribute a doc piece on this?

[sefffal]

Hi @yebai ! I’d welcome including PairPlots in the documentation. I don’t have the bandwidth to add it myself, but may I suggest we just add calls to pairplot in the existing tutorials?

[penelopeysm]

Very nice package!

From a logistical point of view, in Turing's docs, we should not include code from packages that depend on TuringLang packages (i.e. reverse dependencies). There are a couple of reasons for this. The first is that it's not our codebase and it's not our responsibility to document how to use it (e.g. if there are updates to PairPlots, we would be the ones who would have to fix the example to work). The other, more pragmatic, reason for this is compatibility bounds — if we release a new version of MCMCChains, and the reverse dep (PairPlots) doesn't update compatibility right away*, we will be blocked from including the newest version of MCMCChains in our docs.

For these exact reasons (described also in #576), we already removed Pathfinder from the docs: #579

I would be strongly in favour of adding some explanatory text and a link to the PairPlots.jl docs, though. (We still have that for Pathfinder.)

* And this is not to say that I expect other people to update compat bounds right away, or that I was somehow annoyed at Pathfinder for not doing so. In fact, it's quite the contrary — the point that I'm making is that we don't get to decide when reverse-dep compat bounds get updated.