BUG: RollingGroupby ignored as_index=False

[mroeschke]

• closes BUG: groupby.rolling: Originial index is not being preserved when using date_part of DatetimeIndex and as_index key word seems to have no effect #39433
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

[phofl]

Does this fix #31007? Haven't looked closely but seen this in the past

[mroeschke]

@phofl unfortunately doesnt look like it

In [1]: import pandas as pd
   ...:
   ...: data = {
   ...:     'groupby_col': ['A', 'A', 'A', 'A', 'A', 'B', 'B', 'B', 'B', 'B', ],
   ...:     'agg_col': [1, 1, 0, 1, 0, 0, 0, 0, 1, 0],
   ...: }
   ...: df = pd.DataFrame(data)
   ...: df.groupby(['groupby_col'], as_index=False).rolling(4).agg({'agg_col': 'mean'})
Out[1]:
               agg_col
groupby_col
A           0      NaN
            1      NaN
            2      NaN
            3     0.75
            4     0.50
B           5      NaN
            6      NaN
            7      NaN
            8     0.25
            9     0.25

agg still goes partially through groupby code I think to calculate the index result.

[phofl]

Pitty, thanks for checking

[rhshadrach]

What happens here when the groupby is on an explicit list, e.g. in your test use groupby(["A", "A", "B", "B"]) instead. What is groupby_keys in this case?

[mroeschke]

Here's the result

In [4]: df.groupby(["A", "A", "B", "B"], as_index=False).rolling(window=2, min_periods=1).mean()
> /Users/matthewroeschke/pandas-mroeschke/pandas/core/window/rolling.py(580)_apply()
-> result_index_names = groupby_keys + grouped_index_name
(Pdb) groupby_keys
[None]
(Pdb) c
Out[4]:
           level_0    num
date
2018-01-01       A  100.0
2018-01-02       A  150.0
2018-01-01       B  150.0
2018-01-02       B  200.0

In [5]: df.groupby(["A", "A", "B", "B"], as_index=False).mean()
Out[5]:
     num
0  150.0
1  200.0

Not sure if the normal groupby has the expected result but appears that groupby.rolling brings the list into the dataframe as a column

[jreback]

can you add the multi-key groupby here as well (parameterize if you can)

df_res2 = df.groupby([df.id, df.index.weekday], as_index=False).rolling(window=2, min_periods=1).mean()
df_res3 = df.groupby([df.id]).rolling(window=2, min_periods=1).mean()
df_res4 = df.groupby([df.id], as_index=False).rolling(window=2, min_periods=1).mean()

e.g. 2 & 4 (we likley have 1 & 3 covered, but wouldn't object to those included as well)

[mroeschke]

Hmm all these tested (except the as_index=True cases which are tested everywhere else). So you just want it parameterized?

[jreback]

if they are elsewhere then don't (just the cases that is covering in this PR are fine). parameterize if you can.

[jreback]

thanks @mroeschke