Allow Keyword Aggregation in DataFrame.agg and Series.agg

[TomAugspurger]

Followup to #26399

In [2]: df = pd.DataFrame({"A": [1, 2, 1, 2], "B": [1, 2, 3, 4]})

In [3]: df
Out[3]:
   A  B
0  1  1
1  2  2
2  1  3
3  2  4

In [4]: df.agg(foo=("B", "sum"))

Expected Output

In [13]: df.agg({"B": {"foo": "sum"}})
/Users/taugspurger/Envs/dask-dev/lib/python3.7/site-packages/pandas/core/frame.py:6284: FutureWarning: using a dict with renaming is deprecated and will be removed in a future version
  result, how = self._aggregate(func, axis=axis, *args, **kwargs)
Out[13]:
      B
foo  10

without the warning. Similar for Series.agg

In [16]: df.B.agg({"foo": "sum"})  # allow foo="sum"
Out[16]:
foo    10
Name: B, dtype: int64

[TomAugspurger]

Is this a blocker for 0.25.0? Is anyone interested in working on it? I'm not.

cc @zertin, @smcateer, @pirsquared, who seemed interested in this for the GroupBy.agg context (this is for DataFrame/Series.agg).

[zertrin]

I am not sure what this follow-up corresponds to.

As blocker, I remember the multiple lambda problem, for which I believe you said you had a WIP.

Which of the blocker remaining issues this issue corresponds to? The description is unclear to me.

[TomAugspurger]

Multiple lambdas is at #26905 (tagged for 0.25, but needs review).

This issue is about supporting the new named aggregation syntax in DataFrame.agg, rather than just DataFrameGroupBy.agg (likewise for Series). Ideally, these two wouldn't diverge in behavior.

In [2]: df = pd.DataFrame({"A": [1, 2, 1, 2], "B": [1, 2, 3, 4]})

In [3]: df
Out[3]:
   A  B
0  1  1
1  2  2
2  1  3
3  2  4

In [4]: df.agg(foo=("B", "sum"))
      B
foo  10

[TomAugspurger]

Would be nice for 0.25, but not a blocker. Pushing.

[christopherzimmerman]

@TomAugspurger I'm going to look into this, just had a question about desired behavior.

Currently, if you aggregate different functions on different series with DataFrame.agg, you still get a Series back:

In [16]: df = pd.DataFrame({"A": [1, 2, 1, 2], "B": [1, 2, 3, 4]})

In [17]: df
Out[17]:
   A  B
0  1  1
1  2  2
2  1  3
3  2  4

In [18]: df.agg({"A": "sum", "B": "mean"})
Out[18]:
A    6.0
B    2.5
dtype: float64

For the example you provided above, it would seem the result would be something like:

In [18]: df.agg(foo=("A", "sum"), bar=("B", "mean"))
Out[18]:
       A    B
foo    6  NaN
bar  NaN  2.5

When something like:

foo    6.0
bar    2.5
dtype: float64

Would be the closest to current behavior. Any thoughts?

[TomAugspurger]

You're probably right about the expected output being a Series there.