BUG: Series.mask incorrectly replaces positions of pd.NA in the cond argument

[kartoria]

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

series = pd.Series([None,1,2,None,3,4,None])

series.mask(series <= 2, -99)
"""
0     NaN
1   -99.0
2   -99.0
3     NaN
4     3.0
5     4.0
6     NaN
dtype: float64
"""

series = series.convert_dtypes()
series.mask(series <= 2, -99)
"""
0    -99
1    -99
2    -99
3    -99
4      3
5      4
6    -99
dtype: Int64
"""

series = series.convert_dtypes(dtype_backend='pyarrow')
series.mask(series <= 2, -99)
"""
0    -99
1    -99
2    -99
3    -99
4      3
5      4
6    -99
dtype: int64[pyarrow]
"""

Issue Description

When using Series.mask on a Series with a NumPy dtype, np.nan is not replaced. However, for Series with Pandas or PyArrow dtypes, pd.NA is replaced. This behavior is inconsistent and makes it difficult to predict the outcome.

Expected Behavior

import pandas as pd

series = pd.Series([None,1,2,None,3,4,None], dtype='int64[pyarrow]')
series.mask(series <= 2, -99)

"""
0    <NA>
1    -99
2    -99
3    <NA>
4      3
5      4
6    <NA>
dtype: int64[pyarrow]
"""

Installed Versions

INSTALLED VERSIONS

commit : 0691c5c
python : 3.9.18
python-bits : 64
OS : Linux
OS-release : 3.10.0-1160.15.2.el7.x86_64
Version : #1 SMP Wed Feb 3 15:06:38 UTC 2021
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 2.2.3
numpy : 1.26.4
pytz : 2024.1
dateutil : 2.9.0.post0
pip : 24.3.1
Cython : None
sphinx : None
IPython : None
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.12.3
blosc : None
bottleneck : None
dataframe-api-compat : None
fastparquet : None
fsspec : 2024.6.1
html5lib : None
hypothesis : None
gcsfs : None
jinja2 : None
lxml.etree : 5.2.2
matplotlib : 3.9.2
numba : None
numexpr : None
odfpy : None
openpyxl : 3.1.5
pandas_gbq : None
psycopg2 : 2.9.9
pymysql : None
pyarrow : 19.0.0
pyreadstat : 1.2.8
pytest : None
python-calamine : None
pyxlsb : None
s3fs : None
scipy : 1.13.0
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : 2.0.1
xlsxwriter : None
zstandard : None
tzdata : 2024.1
qtpy : None
pyqt5 : None

[sanggon6107]

Hi @kartoria ,
I think this is because :

1. 'series <= 2' returns pandas.NA for pandas.Int64Dtype and int64[pyarrow], whereas it returns False for float64.

series = pd.Series([None,1,2,None,3,4,None])
bool_1 = series < 2
print(bool_1)
"""
0    False
1     True
2    False
3    False
4    False
5    False
6    False
dtype: bool
"""
print(type(bool_1[0]))
"""
<class 'numpy.bool'>
"""

series = pd.Series([None,1,2,None,3,4,None])
series = series.convert_dtypes()
bool_2 = series < 2
print(bool_2)

"""
0     <NA>
1     True
2    False
3     <NA>
4    False
5    False
6     <NA>
dtype: boolean
"""
print(type(bool_2[0]))
"""
<class 'pandas._libs.missing.NAType'>
"""

series = pd.Series([None,1,2,None,3,4,None])
series = series.convert_dtypes(dtype_backend='pyarrow')
bool_3 = series < 2
print(bool_3)
"""
0     <NA>
1     True
2    False
3     <NA>
4    False
5    False
6     <NA>
dtype: bool[pyarrow]
"""
print(type(bool_3[0]))
"""
<class 'pandas._libs.missing.NAType'>
"""

2. NDFrame.fillna(inplace) replaces pandas.NA with the arg 'inplace'. You could actually see the issue doesn't appear when you call mask() with inplace=True. I think we need to investigate whether this behaviour is intentional as well as whether there would be any unexpected results if we make code changes here.

series = pd.Series([None,1,2,None,3,4,None])
series = series.convert_dtypes(dtype_backend='pyarrow')
series.mask(series <= 2, -99, inplace=True)
print(series)
"""
0    <NA>
1     -99
2     -99
3    <NA>
4       3
5       4
6    <NA>
dtype: int64[pyarrow]
"""

I'll do further investigation and make a PR if there's a nice way to fix this.

[sanggon6107]

take