GroupBy case results in malformed DataFrame

```
In [19]: d
Out[19]: 
  group  zeros  ones label
0    g1      0     1    l1
1    g2      0     1    l2

In [20]: d.groupby(['group']).mean()
Out[20]: 
       ones  zeros
group             
g1        1      0
g2        1      0

In [21]: d.groupby(['group']).mean().values
Out[21]: 
array([[ 0.,  1.],
       [ 0.,  1.]])
```
