BUG AttributeError: 'DataFrameGroupBy' object has no attribute '_obj_with_exclusions'

[nbonnotte]

I guess it will be clearer with an example. First, let's prepare the dataframe:

In [2]: df = pd.DataFrame(columns=['a','b','c','d'], data=[[1,'b1','c1',3], [1,'b2','c2',4]])

In [3]: df = df.pivot_table(index='a', columns=['b','c'], values='d').reset_index()

In [4]: df
Out[28]: 
b  a b1 b2
c    c1 c2
0  1  3  4

Now, the exception raised:

In [5]: df.groupby('a').mean()
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-29-a830c6135818> in <module>()
----> 1 df.groupby('a').mean()

/home/nicolas/Git/pandas/pandas/core/groupby.py in mean(self)
    764             self._set_selection_from_grouper()
    765             f = lambda x: x.mean(axis=self.axis)
--> 766             return self._python_agg_general(f)
    767 
    768     def median(self):

/home/nicolas/Git/pandas/pandas/core/groupby.py in _python_agg_general(self, func, *args, **kwargs)
   1245                 output[name] = self._try_cast(values[mask], result)
   1246 
-> 1247         return self._wrap_aggregated_output(output)
   1248 
   1249     def _wrap_applied_output(self, *args, **kwargs):

/home/nicolas/Git/pandas/pandas/core/groupby.py in _wrap_aggregated_output(self, output, names)
   3529     def _wrap_aggregated_output(self, output, names=None):
   3530         agg_axis = 0 if self.axis == 1 else 1
-> 3531         agg_labels = self._obj_with_exclusions._get_axis(agg_axis)
   3532 
   3533         output_keys = self._decide_output_index(output, agg_labels)

/home/nicolas/Git/pandas/pandas/core/groupby.py in __getattr__(self, attr)
    557 
    558         raise AttributeError("%r object has no attribute %r" %
--> 559                              (type(self).__name__, attr))
    560 
    561     def __getitem__(self, key):

AttributeError: 'DataFrameGroupBy' object has no attribute '_obj_with_exclusions'

Maybe I'm doing something wrong, and it's not a bug, but then the exception raised should definitely be more explicit than a reference to an internal attribute :-)

This attribute, by the way, is (only) referenced in one file and in issue #5264. It might be connected, but the discussion is a bit long and technical.

I'll try to have a look at what's going on.

[jreback]

it should be a better error message, but you are grouping on something which is not a column, your
columns are a multi-index.

In [16]: df.columns
Out[16]: 
MultiIndex(levels=[[u'b1', u'b2', u'a'], [u'c1', u'c2', u'']],
           labels=[[2, 0, 1], [2, 0, 1]],
           names=[u'b', u'c'])

In [17]: df.index
Out[17]: Int64Index([0], dtype='int64')

In [18]: df.columns.values
Out[18]: array([('a', ''), ('b1', 'c1'), ('b2', 'c2')], dtype=object)

what exactly are you trying to do?

[nbonnotte]

I'm trying to group according to the column a, or ('a',''). What would be the proper way?

[jreback]

In [27]: df = pd.DataFrame(columns=['a','b','c','d'], data=[[1,'b1','c1',3], [1,'b2','c2',4]])

In [28]: df
Out[28]: 
   a   b   c  d
0  1  b1  c1  3
1  1  b2  c2  4

In [29]: df.groupby('a').mean()
Out[29]: 
     d
a     
1  3.5

[nbonnotte]

But that's not the result I would expect: with my dumb example, I would like to get the same dataframe.

BTW, if df['a'] works whatever the status of a, wouldn't it be nice to be able to group according to a as well?

[jreback]

what are your expecattions for a result here? pls show an example.

a is not a group in your example

[nbonnotte]

i would like that

b  a b1 b2
c    c1 c2
0  1  3  4
1  1  5  5

after grouping by a and taking the mean, yields

b b1   b2
c c1   c2
a        
1  4  4.5

where the first dataframe is for instance obtained with

In [88]: df = pd.DataFrame(columns=['a','b','c','d'], data=[[1,'b1','c1',3], [1,'b2','c2',4], [2,'b1','c1',5], [2,'b2','c2',5]]).pivot_table(index='a', columns=['b','c'], values='d').reset_index()

In [89]: df
Out[89]: 
b  a b1 b2
c    c1 c2
0  1  3  4
1  2  5  5

In [90]: df['a'] = 1

In [91]: df
Out[91]: 
b  a b1 b2
c    c1 c2
0  1  3  4
1  1  5  5

[jreback]

In [17]: df.groupby([('a','')]).mean()
Out[17]: 
b     b1   b2
c     c1   c2
(a, )        
1      4  4.5

[nbonnotte]

So that was that... I had tried

In [99]: df.groupby(('a', '')).mean()
Out[99]: 
b  a b1 b2
c    c1 c2
   1  5  5
a  1  3  4

(the result of which I quite don't understand, but never mind) but not enclosing it betweens brackets. Thanks!

[jreback]

gr8.

if u are interested in improving he error message on he above case would be great

[nbonnotte]

Sure!

[nbonnotte]

@jreback digging about this issue, I think what is happening here is not so much a problem about reporting as a real bug. Indeed, my example just shows that after all issue #11185 was only partially solved by the PR #11202:

In [3]: df = pd.DataFrame(columns=['a', 'b', 'c', 'd'],
                       data=[[1, 'b1', 'c1', 3]])

In [4]: df.groupby('z').mean()
Out[4]: <pandas.core.groupby.DataFrameGroupBy object at 0x7f57f363d510>

This should produce a KeyError. The fact that a KeyError is not raised then allows for the AttributeError that is the subject of this issue, and is caused by the fact that the list of keys passed (here ['z']) is of the same length as the index, which in turn causes match_axis_length to be True in the following line:

https://github.com/pydata/pandas/blob/b07dd0cbd6d18c55aaa0043d85f42a483eab7dbb/pandas/core/groupby.py#L2210

I'll dig a bit deeper before making a PR

[jreback]

hmm, that does looks like a bug. I agree should give a KeyError (though a bit lower down in the code that where you pointed).