PERF: shift() of boolean series gives drawdown by an order of magnitude with default filling np.NaN comparing with filling by bool like False

[TsiVit]

Pandas version checks

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

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

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

Reproducible Example

import pandas as pd
import numpy as np

df = pd.DataFrame(np.random.randn(10**6))<0
%timeit -n5 -r20 df.shift(1)

41.3 ms ± 6.21 ms per loop (mean ± std. dev. of 20 runs, 5 loops each)

Installed Versions

INSTALLED VERSIONS

commit : d9cdd2e
python : 3.11.4.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.19045
machine : AMD64
processor : Intel64 Family 6 Model 142 Stepping 9, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : English_United States.1251

pandas : 2.2.2
numpy : 1.25.1
pytz : 2023.3
dateutil : 2.8.2
setuptools : 65.5.0
pip : 24.0
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.9.3
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.1.2
IPython : 8.14.0
pandas_datareader : None
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.12.2
bottleneck : None
dataframe-api-compat : None
fastparquet : None
fsspec : 2023.10.0
gcsfs : None
matplotlib : 3.7.2
numba : 0.58.1
numexpr : None
odfpy : None
openpyxl : 3.1.2
pandas_gbq : None
pyarrow : 13.0.0
pyreadstat : None
python-calamine : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
zstandard : None
tzdata : 2023.3
qtpy : None
pyqt5 : None

Prior Performance

import pandas as pd
import numpy as np

df = pd.DataFrame(np.random.randn(10**6))<0
%timeit -n5 -r20 df.shift(1, fill_value=False)

389 µs ± 110 µs per loop (mean ± std. dev. of 20 runs, 5 loops each)

[rhshadrach]

I think this is expected since df.shift(1) needs to coerce the dtype to object in order to hold True, False, and np.nan.

df = pd.DataFrame(np.random.randn(10**6)) < 0
print(df.shift(1).dtypes)
# 0    object
# dtype: object
print(df.shift(1, fill_value=False).dtypes)
# 0    bool
# dtype: object

[TsiVit]

I have expected boolean filling for boolean series by default as according the docs:

the default depends on the dtype of self. For numeric data, np.nan is used. For datetime, timedelta, or period data, etc. NaT is used. For extension dtypes, self.dtype.na_value is used.

[rhshadrach]

Do you see something that disagrees with the docs? If that is the case, please provide an example, the result you get, and the result you expect.

[TsiVit]

Yes, according the docs if I have a boolean series, instead of NaN filling of missed values have to be boolean by False (as False is default boolean state)

the default depends on the dtype of self.

[TsiVit]

but anyway, if no one met that before - it's not a problem and only my interpretation, and fill_value make a deal for somebody like me in such case.

[rhshadrach]

Yes, according the docs if I have a boolean series, instead of NaN filling of missed values have to be boolean by False

I do not believe the docs state this. Saying the default depends on the dtype of self does not mean that False is used for Boolean data, and in fact I do not think that should be the behavior. The default value should represent "missing".

In any case, I do think it would improve the docs to read

For Boolean and numeric NumPy data types, np.nan is used.

A PR would be welcome to improve the documentation!

[SongYeongEng]

Hi @rhshadrach, I tried to create a PR to update the documentation.