BUG: `np.where` called with `ps.Series` returns `np.array` instead of `pd.Series`

[domsmrz]

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
s = pd.Series([1,2,3], name='a')
res = np.where(s > 2, s, -s)
print(res)
# res is np.array([-1, -2, 3])

Issue Description

res in the example is np.array. Such behavior is inconsistent with other numpy functions such as np.floor or np.arctan2.

Expected Behavior

res should be Series instead (particularly pd.Series([-1, -2, 3], name='a') in this case.

As for the cases where x and y (i.e., the two last params of np.where) are of mixed types or names, we should probably keep consistency with np.arctan2, meaning:

• if only one of x and y is pd.Series and the other is np.array we should return pd.Series with the same name as input Series
• if both x and y are pd.Series of the same name we should return a pd.Series with that name
• if both x and y are pd.Series but with different names we should return unnamed pd.Series

Installed Versions

INSTALLED VERSIONS

commit : bdc79c1
python : 3.12.3.final.0
python-bits : 64
OS : Windows
OS-release : 11
Version : 10.0.22631
machine : AMD64
processor : AMD64 Family 25 Model 33 Stepping 2, AuthenticAMD
byteorder : little
LC_ALL : None
LANG : None
LOCALE : English_Germany.1252

pandas : 2.2.1
numpy : 1.26.4
pytz : 2024.1
dateutil : 2.9.0.post0
setuptools : None
pip : 23.3.2
Cython : None
pytest : 8.1.1
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : None
IPython : 8.23.0
pandas_datareader : None
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : None
bottleneck : None
dataframe-api-compat : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : None
numba : None
numexpr : 2.10.0
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyreadstat : None
python-calamine : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
zstandard : None
tzdata : 2024.1
qtpy : None
pyqt5 : None

[AkisPanagiotopoulos]

take

[rhshadrach]

Is there is anything pandas can do here? You're calling a NumPy function, and I don't think NumPy is deferring to pandas code to define its behavior.

[domsmrz]

IIUC that is what __array_ufunc__ method is for. At least that is the method that gets invoked for the other methods (e.g., np.floor) and handles the trasformation from array to Series. That being said, I don't know why it isn't invoked in case of np.where and whether there is a reason on the numpy's side or pandas' side.

[rhshadrach]

I think this is numpy/numpy#8994; closing as an upstream issue. @domsmrz - let me know you think I'm missing something.

[domsmrz]

Thanks for linking the issue @rhshadrach . That explains why __array_ufunc__ is not invoked in this case. However we may be able to fix this within pandas by implementing __array_function__ (as per numpy/numpy#5095 (comment) )?