Groupby aggregations could ignore non-numeric columns when axis=1

[hayd]

Perhaps the following groupby aggregation should work only the numeric columns, as they would when using the dataframe:

In [1]: df = pd.DataFrame({'bar': {0: 1, 1: 1, 2: 1}, 'foo': {0: 0, 1: 1, 2: 2}, 'foo1': {0: 1, 1: 2, 2: 3}, 'hello': {0: 'a', 1: 'a', 2: 'a'}}, columns=['bar', 'foo', 'foo', 'hello'])

In [2]: df
Out[2]:
   bar  foo  foo hello
0    1    0    1     a
1    1    1    2     a
2    1    2    3     a

In [3]: df.mean()  # hello is ignored
Out[13]:
bar    1
foo    1
foo    2
dtype: float64

In [4]: df.groupby(level=0, axis=1).mean()
---------------------------------------------------------------------------
DataError                                 Traceback (most recent call last)
<ipython-input-4-7c2612a8fbda> in <module>()
----> 1 df.groupby(level=0, axis=1).mean()

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/core/groupby.pyc in mean(self)
    351         """
    352         try:
--> 353             return self._cython_agg_general('mean')
    354         except GroupByError:
    355             raise

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/core/groupby.pyc in _cython_agg_general(self, how, numeric_only)
   1569
   1570     def _cython_agg_general(self, how, numeric_only=True):
-> 1571         new_blocks = self._cython_agg_blocks(how, numeric_only=numeric_only)
   1572         return self._wrap_agged_blocks(new_blocks)
   1573

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/core/groupby.pyc in _cython_agg_blocks(self, how, numeric_only)
   1616
   1617         if len(new_blocks) == 0:
-> 1618             raise DataError('No numeric types to aggregate')
   1619
   1620         return new_blocks

DataError: No numeric types to aggregate

From this SO question, where I gave very hacky workaround.

cc #3683 @jreback was this the question you were talking about? This ones related but in the sense of coming up against non unique problems... Thought I should mention it here anyway.

[jreback]

this is the question, but still breaks (even on my new branch).... will look at this soon

[jreback]

linking to #3679

[keir]

Any update on this @jreback? Did your branch ever go anywhere?

[jreback]

you are welcome to do a pr if u would like

[WillAyd]

Still a problem, though note that this only fails due to axis=1

[NumberPiOso]

Nowadays running the example results in a FutureWarning at both operations.

In [3]: df.mean()
FutureWarning: Dropping of nuisance columns in DataFrame reductions (with 'numeric_only=None') is deprecated; in a future version this will raise TypeError.  Select only valid columns before calling the reduction.


In [4]:  df.groupby(level=0, axis=1).mean() 
FutureWarning: Dropping invalid columns in DataFrameGroupBy.mean is deprecated. In a future version, a TypeError will be raised. Before calling .mean, select only columns which should be valid for the function.
  df.groupby(level=0, axis=1).mean()
Out[4]: 
Empty DataFrame
Columns: [bar, foo, hello]
Index: []

This change was introduced by PR #41480. So, nowadays groupby aggregations should NOT ignore non-numeric columns. Closing this issue.