groupby boxplot creates more subplots than are needed

[fonnesbeck]

When calling boxplot on a DataFrameGroupBy object, it does not seem to be able to properly infer the number of subplots required to account for the number of groups. For example, the following:

lsl_dr[lsl_dr.test_type==t].groupby('type').boxplot(column='score', grid=False)

generates this plot:

Notice that it generates a 2x2 grid, even though there are only three groups. I thought I could deal with this manually by calling pyplot.subplots(1,3), but the boxplot method does not appear to be able to accept an axis argument like the DataFrame.boxplot method does. Is there a work-around for this?

[cpcloud]

Yep you can loop over the groups:

In [66]: df = DataFrame({'cats': np.random.choice(['tabby', 'grumpy', 'smelly'], size=1000), 
                         'weight': normal(13, 5, size=1000)})

In [67]: fig, axs = subplots(1, 3)

In [68]: for (k, v), ax in itertools.izip(df.groupby('cats'), axs.flat):
   ....:     v.boxplot(column='weight', ax=ax)
   ....:

[fonnesbeck]

That works, thanks. It would be great if DataFrameGroupBy.boxplot were able to take an axis as an argument.

[cpcloud]

What sort of API would you want? Are you thinking axis would take a sequence of axes objects?

[fonnesbeck]

Yeah, or maybe just rows= or cols=.

[jreback]

closed by #7035