ENH: rolling apply multiple columns or whole dataframe

[mb-hz]

Pandas version checks

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

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

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd

value4 = 10
weight3 = 2
weight2 = 3
data = {'value': [np.nan, np.nan, np.nan, value4],
        'weight': [np.nan, np.nan, weight3, weight2]}
df = pd.DataFrame(data)

df.rolling(window=2, min_periods=1).apply(lambda x: x['value'][0] * x['weight'][0], raw=False)

Issue Description

If you do for i in df.rolling(window=2, min_periods=1): print(i), you can see that i includes both columns. However, when you use df.rolling with df.apply function, the function can not recognise both columns.

Expected Behavior

I expect the rolling function can return multiple columns as it shows in for loop print, into apply function after it, when we use dataframe instead of series or array as the input.

Installed Versions

INSTALLED VERSIONS

commit : 2e218d1
python : 3.11.1.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.19045
machine : AMD64
processor : Intel64 Family 6 Model 154 Stepping 4, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : English_United Kingdom.1252

pandas : 1.5.3
numpy : 1.23.5
pytz : 2022.7.1
dateutil : 2.8.2
setuptools : 65.5.0
pip : 23.0.1
Cython : 0.29.33
pytest : 7.2.2
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.9.2
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.1.2
IPython : 8.12.0
pandas_datareader: None
bs4 : 4.12.0
bottleneck : None
brotli : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : 3.7.1
numba : None
numexpr : None
odfpy : None
openpyxl : 3.1.2
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : 1.10.1
snappy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
zstandard : None
tzdata : None

[topper-123]

That's what the parameter method="table" is for. It is however only implemented when given engine="numba" in the method, so is a bit of a hit & miss for custom functions given to apply.

So, this is currently more of an limitation of the current API. and changing it,so method="table" also works for non-numba functions, could be worth an investigation but would be an enhancement proposal, not a bug fix IMO.

Is rolling.apply part of your work on fixing groupby.apply, @rhshadrach?

[mb-hz]

Hi topper-123, thank you so much for your reply.

The info for reference of numba is limited. I suppose this code should work, replacing the last row of code in the question.
df.rolling(window=2, min_periods=1, method="table").apply(lambda x: x, raw=True, engine='numba'), which returns an error. ValueError: cannot broadcast source array for assignment

[topper-123]

I am not familiar with the restrictions on when numba works with rolling.apply, but yeah I tried that too and it didn't work, I haven't checked why...

I looked through the issue list, and there is already #40374, which started out as a enhancement proposal, but ended up. as a doc issue, but is still open, though the related doc issue has been resolved. So IDK, can you rephrase this to be an enhancement proposal, requesting to method="table" in rolling work outside of numba also? Then I'll close the other as completed.

[rhshadrach]

Is rolling.apply part of your work on fixing groupby.apply, @rhshadrach?

Yes, it is!

[topper-123]

Is rolling.apply part of your work on fixing groupby.apply, @rhshadrach?

Yes, it is!

Nice. Would we then be able to roll over a DataFrame like requested here?

[rhshadrach]

Yes - I am aiming to add user-control as to whether the columns or the entire frame is passed to the UDF across the board, so in particular in rolling.

[mroeschke]

I am not familiar with the restrictions on when numba works with rolling.apply, but yeah I tried that too and it didn't work, I haven't checked why...

For context, this was done because the current window aggregations (including apply) in cython were only written to accepts 1D inputs. At the time, it wasn't in my scope to rewrite them to 2D