Deprecating the trellis rplot module (GH3445)

[jorisvandenbossche]

Closes #3445

---

Start for #3445. Then the current examples should get a seaborn or ggplot alternative.

Questions:

• explicitely refer to seaborn and ggplot as alternatives? (as there may be other packages that could feel neglected that way?)
• I now raise the warning on import of the module (import pandas.tools.rplot as rplot). The only thing is that the warning will be raised in the tests this way

[TomAugspurger]

@TomAugspurger Did you already have some examples of how to do it in seaborn? I vaguely can remember you posted it once in an issue?

@jorisvandenbossche I redid all the examples using seaborn. A quick search didn't turn it up, but I'll dig for it later.

[tacaswell]

There are ways to both suppress and check for warnings in the tests, we have a couple of examples of it in mpl's test suite.

[shoyer]

I'm not sure ggplot is ready for prime time. I would probably only recommend Seaborn at this point.

[jorisvandenbossche]

@shoyer yes, I could agree, as I noticed while trying to convert some of the examples to gpplot this was not easy (some of the functionality of rplot is not yet included in ggplot). I will update the PR

[jankatins]

I agree too: ggplot currently doesn't handle categorical data very well if faceting is used (actually that was my motivation for working on categorical data in pandas :-) ).

[jorisvandenbossche]

OK, I will remove ggplot for now from the explanation, and if @TomAugspurger finds his seaborn examples, we can add that. But @JanSchulz, if you want to add some ggplot examples when there is some progress, certainly do!

[jreback]

@jorisvandenbossche @TomAugspurger let's do this for 0.16.0 (pls add a refernce in #6581) as well

[jorisvandenbossche]

I made some examples of how to do it with seaborn, for each example in de rplot docs now: http://nbviewer.ipython.org/gist/jorisvandenbossche/103876bff4a5f42c9933#

@TomAugspurger Can you have a looks if this looks good? Or if there are better ways? (only for the last example (4 variables) I did not found a fully equivalent seaborn-way of doing it)

@jreback For the actual PR, I assume adding seaborn as a doc dependency is not that handy, so I better include all the figures as images?

[TomAugspurger]

Joris, I remember talking to Michael about that last plot and it was agreed that seaborn didn't have a way to do that plot directly. It was also agreed that this is a good thing, since that chart is a mess :) Thanks for redoing this

[jorisvandenbossche]

I fully agree about that mess :-)

The rest is looking OK?

[shoyer]

you might add hue_kws={"marker": ["^", "v", "s", "*"]} if you like the symbols.... or not :).

[jorisvandenbossche]

Are the different symbols then additional to the different colors? Or in place of?

[shoyer]

The symbols are matched with colors 1 to 1. See the Seaborn docs on plotting on data aware grids for details.

On Wed, Mar 4, 2015 at 11:23 AM, Joris Van den Bossche
notifications@github.com wrote:

• ScaleRandomColour(column)
• g = sns.FacetGrid(tips_data, row="sex", col="smoker", hue="day")
  Are the different symbols then additional to the different colors? Or in place of?

---

Reply to this email directly or view it on GitHub:
https://github.com/pydata/pandas/pull/9357/files#r25804584

[shoyer]

This looks great to me!

[jreback]

@jorisvandenbossche I think just need a release note in the deprecations sections

[jorisvandenbossche]

@jreback ah yes!