BUG: Aggregating over an integer on an empty DataFrame removes the index name

[chrisaycock]

Suppose I have this simple DataFrame:

In [91]: df = pd.DataFrame({'ref':[1, 2, 3], 'symbol':['GOOGL', 'IBM', 'MSFT'], 'price':[585.93, 180.72, 42.03]}, columns=['ref', 'symbol', 'price'])

In [92]: df
Out[92]:
   ref symbol   price
0    1  GOOGL  585.93
1    2    IBM  180.72
2    3   MSFT   42.03

In [93]: df.dtypes
Out[93]:
ref         int64
symbol     object
price     float64
dtype: object

If I aggregate over an object and then reset the index, everything comes back as expected:

In [95]: df.groupby('symbol').first().reset_index()
Out[95]:
  symbol  ref   price
0  GOOGL    1  585.93
1    IBM    2  180.72
2   MSFT    3   42.03

Similarly with aggregating over an integer:

In [96]: df.groupby('ref').first().reset_index()
Out[96]:
   ref symbol   price
0    1  GOOGL  585.93
1    2    IBM  180.72
2    3   MSFT   42.03

Now let's say I have an empty DataFrame. Aggregating over an object produces what I expect:

In [97]: df.query('price > 1000').groupby('symbol').first().reset_index()
Out[97]:
Empty DataFrame
Columns: [symbol, ref, price]
Index: []

But aggregating over an integer gives me index as the column name instead of the expected ref!

In [98]: df.query('price > 1000').groupby('ref').first().reset_index()
Out[98]:
Empty DataFrame
Columns: [index, symbol, price]
Index: []
          ^^^^^

Interestingly, setting the index in a non-aggregating function does the right thing:

In [100]: df.query('price > 1000').set_index('ref').reset_index()
Out[100]:
Empty DataFrame
Columns: [ref, symbol, price]
Index: []

So aggregating over an integer in an empty DataFrame removes the index name. This was discovered in pandas 0.14.0.

[jreback]

hmm...could be that the name is lost, ok, marking as a bug

if you would like to dig in and see if you figure out would be great

[jreback]

ahh, yes I fixed that in #7580 (it has an implicit test for the name).

would you like to submit a PR including the tests that you have for both issues?