DOC: move Trellis plotting under Plotting

[jreback]

As this is not really a fleshed out section, kind of odd to have it as a top-level

@TomAugspurger
@jorisvandenbossche

[TomAugspurger]

I noticed that the other day. I was wondering if I messed it up when I did the reorganization. I'll move it under plotting.

[jorisvandenbossche]

I know the PR is already merged, but some comments:

• I agree that this is not worth a toplevel link in the docs index, but I am personally in favor of evolving the docs towards smaller files (so splitting existing files, eg in subfolders) instead of merging them into even larger files than they already are. But this is a little bit a seperate issue as this needs a reorganisation of the docs index (so OK for the PR).
• About this plotting interface, I think it is not really actively developed/maintained/enhanced by anyone? And given that this functionality is now much better covered in things like seaborn and ggplot, and @jreback called the docs section on this 'not really fleshed out' which is even an understatement I think. Shouldn't we rather deprecate it and eventually remove it from the pandas code base? (instead of promoting it to out main plotting doc page) See also the previous discussion in Deprecating the trellis rplot module #3445 and the idea of adding some showcases of seaborn/ggplot/... to our vis docs (I don't directly find where this was dicussed recently)

[TomAugspurger]

All good points. It does really feel buried at the end of that massive page.

Thoughts on putting a warning about it possibly being deprecated in the future. I can also try to do all the examples in seaborn to show a side by side and help people transition away. I don't use ggplot, but it would presumably be capable.

[jorisvandenbossche]

If we deprecate it, that would be a great way to document it (current example with pandas trellis interface -> how to do it in seaborn/ggplot)
I think the trellis interface was modelled a bit after ggplot, so I would assume it should also be easily translatable to that.

But for the actual discussion on deprecating it, I will reinitiate it in #3445

[jreback]

@jorisvandenbossche @TomAugspurger this all sounds fine

as an aside, I had a discussion with

cc @tacaswell

These guys are in the process of proposing essentially an API that would allow something like this:

df.plot(......., engine='matplolib') (whereby pandas could delegate virtually all of the actual plotting functions to the engine (and where engine could be seaborn/bokeh/ggplot/matplotlib

IOW, they would in theory handle all of the messy details that pandas now handles for example resample time-series and such (via callbacks) for example frequency resetting when zooming.

So pandas would just be the interface and the implementation would move to these engines.

So I think they will open a discussion soon on this.

[jorisvandenbossche]

@jreback That sounds interesting! Although I don't think it is really related to this issue, as you speak of df.plot(..) while this issue is about the rplot.py module

[jreback]

@jorisvandenbossche correct....just wanted to mention (and don't need a new issue at this point)