ENH: allow `ewm()` to use a DatetimeIndex if `times` isn't provided

[gsiano]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

rolling() allows passing a timedelta window when the DataFrame has a DatetimeIndex. ewm() requires passing the times parameter when using a timedelta halflife with a DatetimeIndex. I think ewm() should default to using the index as times if it is a DatetimeIndex and halflife is a timedelta. This way it would be more similar to the rolling() behavior, and in my experience I've only ever seen df.index passed as the times argument. It seems counterintuitive to me to have to explicitly pass the index to ewm() when working with times and timedeltas.

Feature Description

If the halflife parameter to ewm() is a timedelta and times is not explicitly passed, use the index if it is a DatetimeIndex.

#54181

Alternative Solutions

The alternative would be to keep the existing behavior.

Additional Context

Current:

>>> now = pd.Timestamp.now()
>>> idx = [now + pd.Timedelta(seconds=i) for i in range(1, 4)]
>>> df = pd.DataFrame({'a': [1,2,3]}, index=idx)
>>> df.ewm(halflife=pd.Timedelta(seconds=1)).mean()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/gsiano/.local/lib/python3.10/site-packages/pandas/core/generic.py", line 11743, in ewm
    return ExponentialMovingWindow(
  File "/home/gsiano/.local/lib/python3.10/site-packages/pandas/core/window/ewm.py", line 386, in __init__
    raise ValueError(
ValueError: halflife can only be a timedelta convertible argument if times is not None.
>>> df.ewm(halflife=pd.Timedelta(seconds=1), times=df.index).mean()
                                   a
2023-07-18 10:45:18.007264  1.000000
2023-07-18 10:45:19.007264  1.666667
2023-07-18 10:45:20.007264  2.428571
>>> 

Proposed:

>>> now = pd.Timestamp.now()
>>> idx = [now + pd.Timedelta(seconds=i) for i in range(1, 4)]
>>> df = pd.DataFrame({'a': [1,2,3]}, index=idx)
>>> df.ewm(halflife=pd.Timedelta(seconds=1)).mean()
                                   a
2023-07-18 10:46:46.589891  1.000000
2023-07-18 10:46:47.589891  1.666667
2023-07-18 10:46:48.589891  2.428571