BUG: Pandas rolling window always converts to float

[daviddavo]

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

This example function gives an error if passed as an argument to apply

aux = pd.Series([1,1,1,1,2,2,3,4,], dtype='uint32')
@numba.jit(numba.float64(numba.uint32[:]), nopython=True)
def nunique(arr):
    return len(set(arr))

aux.rolling(2).apply(nunique, raw=True)

This raises the following error:

TypeError: No matching definition for argument type(s) array(float64, 1d, C)

Issue Description

If you try to use a function that receives anything other than a float on a rolling window, it will give you an error. This is because everything is converted to float beforehand.

Furthermore, arithmetic addition is not invertible for floating point¹

Expected Behavior

raw=True should respect types

Installed Versions

INSTALLED VERSIONS

commit : 2e218d1
python : 3.10.10.final.0
python-bits : 64
OS : Linux
OS-release : 6.1.26-1-MANJARO
Version : #1 SMP PREEMPT_DYNAMIC Wed Apr 26 22:07:35 UTC 2023
machine : x86_64
processor :
byteorder : little
LC_ALL : None
LANG : en_GB.UTF-8
LOCALE : en_GB.UTF-8

pandas : 1.5.3
numpy : 1.23.5
pytz : 2022.7.1
dateutil : 2.8.2
setuptools : 65.5.0
pip : 22.3.1
Cython : None
pytest : 7.2.1
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.1.2
IPython : 8.8.0
pandas_datareader: None
bs4 : 4.11.1
bottleneck : None
brotli : 1.0.9
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : 3.6.3
numba : 0.56.4
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : 10.0.1
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : 1.10.0
snappy : None
sqlalchemy : None
tables : None
tabulate : 0.8.10
xarray : None
xlrd : None
xlwt : None
zstandard : None
tzdata : 2023.3

Footnotes

1. Martin Hirzel, Scott Schneider, and Kanat Tangwongsan. 2017. Sliding-Window Aggregation Algorithms: Tutorial. In Proceedings of the 11th ACM International Conference on Distributed and Event-based Systems (DEBS '17). Association for Computing Machinery, New York, NY, USA, 11–14. https://doi.org/10.1145/3093742.3095107 ↩

[lithomas1]

Hi @daviddavo,
Thanks for the report. I can reproduce this on main.

The issue here is that our Cython code that handles rolling can only take in float64 inputs, so the input is cast beforehand to float64 here.

pandas/pandas/core/window/rolling.py

Line 369 in 935244a

values = ensure_float64(values)

Your code is called from within the Cython function, so that's why it fails with a TypeError, since the input is already a float by then.

I believe #46619 has the same root cause as this one.

I will close that one and relabel this as an enhancement.

[lithomas1]

For reference, the MWE for the complex issue is (notice the warning thrown):

Complex issue

import pandas as pd
pd.DataFrame([1j,1+2j]).rolling(2).apply(lambda x: print(x) is None)