BUG: The behavior for DataFrame.corrwith is changed in pandas 1.5.0 ??

[itholic]

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
import numpy as np

df1 = pd.DataFrame(
    {"A": [1, np.nan, 7, 8], "B": [False, True, True, False], "C": [10, 4, 9, 3]}
)
df2 = df1[["B", "C"]]
(df1 + 1).corrwith(df2.B, method="spearman")

Issue Description

Seems like the behavior of pandas.DataFrame.corrwith is changed from pandas 1.5.0:

The result from the reproducible example in pandas 1.4.0:

>>> (df1 + 1).corrwith(df2.B, method="spearman")
A    0.0
B    1.0
C    0.0
dtype: float64

However, the result from the reproducible example in pandas 1.5.0:

>>> (df1 + 1).corrwith(df2.B, method="spearman")
A    0.5
B    1.0
C   -0.2
dtype: float64

I see the #46174 that improve the performance, but maybe the behavior also changed?

Expected Behavior

The expected behavior in the previous pandas was:

>>> (df1 + 1).corrwith(df2.B, method="spearman")
A    0.0
B    1.0
C    0.0
dtype: float64

Installed Versions

INSTALLED VERSIONS

commit : 87cfe4e
python : 3.9.11.final.0
python-bits : 64
OS : Darwin
OS-release : 21.6.0
Version : Darwin Kernel Version 21.6.0: Wed Aug 10 14:25:27 PDT 2022; root:xnu-8020.141.5~2/RELEASE_X86_64
machine : x86_64
processor : i386
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.UTF-8

pandas : 1.5.0
numpy : 1.22.3
pytz : 2021.3
dateutil : 2.8.2
setuptools : 61.2.0
pip : 22.2.2
Cython : 0.29.24
pytest : 6.2.5
hypothesis : None
sphinx : 3.0.4
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.0.3
IPython : 7.29.0
pandas_datareader: None
bs4 : 4.10.0
bottleneck : 1.3.4
brotli : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : 3.5.1
numba : None
numexpr : 2.8.1
odfpy : None
openpyxl : 3.0.10
pandas_gbq : None
pyarrow : 9.0.0
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : 1.7.2
snappy : None
sqlalchemy : 1.4.26
tables : None
tabulate : 0.8.9
xarray : None
xlrd : None
xlwt : None
zstandard : None
tzdata : None

[MarcoGorelli]

Hey, thanks for the report

I see the #46174 that improve the performance, but maybe the behavior also changed?

Yup, git bisect confirms

[MarcoGorelli]

cc @fractionalhare (no blame, just in case you know how to fix it)

[fractionalhare]

Thanks for flagging. I can look later today after work. This is strange, my guess is an edge case not covered by tests before. I think it involves the NaN not being ranked by argsort for the Spearman calculation. Will confirm later.

[phofl]

Reverting would be an option, if we can not figure out what caused this

[MarcoGorelli]

Hi @itholic

Thanks for having reported this - there's now a release candidate for pandas 2.0.0, which you can install with

mamba install -c conda-forge/label/pandas_rc pandas==2.0.0rc0

or

python -m pip install --upgrade --pre pandas==2.0.0rc0

If you try it out and report bugs, then we can fix them before the final 2.0.0 release