Pandas API: Panel.groupby has an older interface

This seems to not conform to the DataFrame.groupby/Series.groupby

should be straightforward to fix this I think, just simply removing it in favor of core/generic/NDFrame.groupby, but would need to fix tests, maybe. The mapper argument is the by argument.

http://stackoverflow.com/questions/24914748/why-does-panel-groupby-not-accept-level-parameter-in-pandas-0-14/24915441#24915441

Comment From: connorbrinton

Attempting to call Panel.mean(axis=foo, level=bar) on a Panel with a MultiIndex for items fails because of this older API for groupby.

Details:

In [16]: results
Out[16]:
<class 'pandas.core.panel.Panel'>
Dimensions: 162 (items) x 3 (major_axis) x 107 (minor_axis)
Items axis: (0, 0) to (26, 5)
Major_axis axis: x to z
Minor_axis axis: 0 to 106

In [17]: results.mean(0, level='ring')
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/home/connor/script.py in <module>()
----> 1 results.mean(0, level='ring')

/usr/local/lib/python2.7/dist-packages/pandas/core/generic.pyc in stat_func(self, axis, skipna, level, numeric_only, **kwargs)
   5306         if level is not None:
   5307             return self._agg_by_level(name, axis=axis, level=level,
-> 5308                                       skipna=skipna)
   5309         return self._reduce(f, name, axis=axis, skipna=skipna,
   5310                             numeric_only=numeric_only)

/usr/local/lib/python2.7/dist-packages/pandas/core/generic.pyc in _agg_by_level(self, name, axis, level, skipna, **kwargs)
   5016
   5017     def _agg_by_level(self, name, axis=0, level=0, skipna=True, **kwargs):
-> 5018         grouped = self.groupby(level=level, axis=axis)
   5019         if hasattr(grouped, name) and skipna:
   5020             return getattr(grouped, name)(**kwargs)

TypeError: groupby() got an unexpected keyword argument 'level'

Comment From: jreback

Panels with MI have very little support. You are highly recommended to use a DataFrame with a MI. (also much easier to reason about).