ENH: Improved (JMP-like) Box Plot Hierarchy Labels

[tty2-tony]

For a lot of work in statistics, box plots are still one of the most useful tools for at-a-glance comparison of experimental groups. JMP has a great default handler for plotting box plots when including multiple grouping levels for the data, including sane handling for missing data. Please see this example for what a simple case looks like:

In terms of 'sane defaults' and making a lot of use without a lot of work, I think this would be a great model for future box plot work for pandas given its rich GroupBy functionality. When creating a box plot based on a multi-level grouping in pandas/matplotlib, the only obvious method I can see is to use labels which are tuples of each of the values for the various group levels. This makes it hard to get much out of the plot visually for grouped data set.

JMP has a variety of other features for box plots that would be very useful in the future, but I believe this first visual enhancement would be more powerful for the most users.

[jreback]

this might be nice to add to: https://github.com/mwaskom/seaborn as well

[tty2-tony]

I'll copy it over there and potentially leave it here for discussion, or however you all see fit.

Thanks!

[dragoljub]

I like this idea! JMP has one of the best plotting systems and handles all kinds of data very well.

[ghost]

mwaskom/seaborn#64 already had this in the works. good stuff.

I for one am glad to have pandas' plotting superceded and upstaged by our new python viz overlords.