Feature Request: JMP-Style Grouped Boxplots about seaborn HOT 3 CLOSED

Actually in 0.3 there is going to be a more flexible faceting system that will make this very easy:

I've thought about making the groupby argument for boxplot() take a list of columns and break down the boxes hierarchically that way, but I think the slightly more verbose approach of using FacetGrid is ultimately preferably to hardcoding a lot of nested Pandas operations in the individual plotting functions.

from seaborn.

ooh, I like the simple structure of mapping a plot function onto the FacetGrid class. So far as I'm concerned this is the 'killer abstraction' that will simplify so much of my own ad-hoc matplotlib faceting code.

I poked through your facet_grid branch and I love that the class accept either a row argument or a col_wrap argument: Easy, robust facet-wrapping was the biggest reason I couldn't completely drag myself away from ggplot2... Three cheers for seaborn! I'm lookings forward to trying this out.

I wonder, is it possible using this structure to apply another "layer" of data on top of the graph with a second call to g.map(...), perhaps as an example just to add a horizontal black line for "sex grand mean' to both plots? If so, I could see this developing into a pretty powerful method of building up plots...

e.g.

g = sns.FacetGrid(data, ...)
g.map(sns.boxplot, ...)
g.map(my_own_plot_grand_mean_hline_func, color='black',  ...)

from seaborn.

Yep, I think that should work. I'll pink you when I have it ready for a PR, I'll want people to play with it a bit before I merge to master.

from seaborn.