ENH: Adding DataFrame.boxplot()

[lodagro]

Added DataFrame.boxplot() in similar way as DataFrame.plot() and DataFrame.hist().
Some syntactic sugar ;-)

Thinking about adding DataFrameGroupBy.boxplot(), now this call is transfered to DataFrame.boxplot().
But i would prefer the behavior to be different. When groupby is used i`m interested to see in a single plot boxplots for one particular column for all groupby keys and not in one plot to have boxpots for all columns for one groupby key. An input argument can be used to so both ways of boxplotting can be used.

[wesm]

I'll think about this a bit. There are two cases:

• Box plots of each column
• Box plot of one column, grouped by the values in another

I guess it would make sense to support both of these use cases in a single function. You could imagine a plot with multiple box plots generated off of multiple columns, would be pretty cool

[wesm]

So I'm skipping your patch and I hacked up a DataFrame.boxplot with a groupby and column selection option, it's very nice! See e.g. http://imgur.com/1oko5

take a look in the latest and try out the new boxplot function...let me know if there's anything else you would want to add

[lbeltrame]

I noticed an inconsistency in the boxplot documentation. It's supposed to return an AxesSubplot instance, but at least without grouping, it returns None.

[lodagro]

Just gave it a spin, this is what i had in mind and looks good.

some comments:

• As cswegger mentioned DataFrame.boxplot has no return call (pandas.tools.plotting.boxplot does), when i create boxplots i usually add two horizontal lines on the plot indicating the data boundaries, getting the ax as return from boxplot makes this easy. I did not try to use the ax input argument.
• Either i don`t understand the column argument or it has some issues.
  repo/scripts/boxplot_test..py is where i started from:
  • df.boxplot(by=['indic', 'indic2'], fontsize=8, rot=90) -- works fine, this is what is in the test script
  • df.boxplot() -- ok
  • df.boxplot(column=['A']) -- i would expect this would only create a boxplot for column A, but boxplot for all coumns are generated
  • df.boxplot(by=['indic', 'indic2'], column=['A']) -- i woudl expect a boxplot similar as in the first example, but only for column A, it gives TypeError: unhashable type: 'list'
• I mainly use MultiIndex together with DataFrame. The by argument of Dataframe.boxplot works on columns, not on the index, workaround is to use DataFrame.delevel(), but then the indices are also boxplotted (if they hold numeric data). Deleveling also offers possibility to do groupby on multiple levels, which is not (yet :-) )possible on the MultiIndex itself.
• subplot layout improved (not always using a rectangular layout), which is nice. DataFrame.hist() could benefit from this also.