BUG: Groupby rolling count with datetime not working correctly

[nrcjea001]

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Note: Please read this guide detailing how to provide the necessary information for us to reproduce your bug.

Code Sample, a copy-pastable example

df = pd.DataFrame({"column1": range(6), 
                   "column2": range(6), 
                   'group': 3*['A','B'], 
                   'date':pd.date_range(end="20190101", periods=6)})

## Without Closed Parameter
df.groupby('group').rolling('3d',on='date')['column1'].count()
## Output
group   
A      0    1.0
       2    2.0
       4    3.0
B      1    1.0
       3    2.0
       5    3.0
Name: column1, dtype: float64

## With Closed Parameter
df.groupby('group').rolling('3d',on='date',closed='left')['column1'].count()
## Output
ValueError: closed only implemented for datetimelike and offset based windows

Problem description

In Pandas 1.1.1, although .sum() and .mean() are working correctly, the rolling groupby .count() is not working correctly with datetime. Furthermore, when closed parameter is added, ValueError is produced.

Expected Output

## Without Closed Parameter
group   
A      0    1.0
       2    2.0
       4    2.0
B      1    1.0
       3    2.0
       5    2.0
Name: column1, dtype: float64
## With Closed Parameter
group   
A      0    NaN
       2    1.0
       4    1.0
B      1    NaN
       3    1.0
       5    1.0
Name: column1, dtype: float64

Output of pd.show_versions()

INSTALLED VERSIONS

commit : f2ca0a2
python : 3.8.5.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.18362
machine : AMD64
processor : Intel64 Family 6 Model 142 Stepping 11, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : English_United States.1252

pandas : 1.1.1
numpy : 1.18.5
pytz : 2020.1
dateutil : 2.8.1
pip : 20.2.1
setuptools : 49.2.1.post20200802
Cython : 0.29.21
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.11.2
IPython : 7.17.0
pandas_datareader: None
bs4 : 4.9.1
bottleneck : None
fsspec : 0.8.0
fastparquet : None
gcsfs : None
matplotlib : 3.3.0
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : 1.0.0
pytables : None
pyxlsb : None
s3fs : None
scipy : 1.5.0
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
numba : 0.48.0

[TomAugspurger]

@nrcjea001 can you check what commit introduced the change, to see if this was deliberate?

[nrcjea001]

@TomAugspurger, don't think this was deliberate. If we take the sum and divide by the mean (which is equivalent to the count), we achieve the expected output. However, when using the rolling count function, we do not get the expected output. There has also been some speed improvements to the sum and mean code, while the count is considerably slower (see here). Not sure if this is related.

In regards to the ValueError, the same issue appeared for the mean and sum and was fixed, but not the count (see here).

[TomAugspurger]

Thanks. Are you interested in working on this?

…

On Sep 6, 2020, at 03:21, nrcjea001 ***@***.***> wrote:  @TomAugspurger, don't think this was deliberate. If we take the sum and divide by the mean (which is equivalent to the count), we achieve the expected output. However, when using the rolling count function, we do not get the expected output. There has also been some speed improvements to the sum and mean code, while the count is considerably slower (see here). Not sure if this is related. In regards to the ValueError, the same issue appeared for the mean and sum and was fixed, but not the count (see here). — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[nrcjea001]

Very sorry, I do not have the skill set for this. But thanks for asking

[simonjayhawkins]

first bad commit: [6875a05] BUG: RollingGroupby with closed and column selection no longer raises ValueError (#35639)

https://github.com/simonjayhawkins/pandas/runs/1092697367?check_suite_focus=true

[simonjayhawkins]

@mroeschke are you still planning to work on this?

[mroeschke]

@simonjayhawkins yes but unsure if I'll be able to get to this by 1.1.3. Need to clean up the count code path first to make this easier to debug xref #36649

[simonjayhawkins]

@simonjayhawkins yes but unsure if I'll be able to get to this by 1.1.3.

not a problem. can move to 1.1.4

@mroeschke Thanks for the update.