BUG: max and min returns incorrect result

[YarShev]

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

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

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Note: Please read this guide detailing how to provide the necessary information for us to reproduce your bug.

Code Sample, a copy-pastable example

# Your code here
import pandas
import numpy as np
df = pandas.DataFrame(
    {
        "int_col": [-5, 2, 7, 16],
        "float_col": [np.NaN, -9.4, 10.1, np.NaN],
        "str_col": ["a", np.NaN, "c", "d"],
        "bool_col": [False, True, True, False],
    }
)
df
   int_col  float_col  bool_col
0       -5        NaN     False
1        2       -9.4      True
2        7       10.1      True
3       16        NaN     False
# pandas 1.1.5 behavior:
df.max(axis=0, numeric_only=True)
int_col      16.0
float_col    10.1
bool_col      1.0
dtype: float64
# pandas 1.2.1 behavior:
df.max(axis=0, numeric_only=True)
int_col        16
float_col    10.1
bool_col     True
dtype: object

# the difference is in last two lines
bool_col      1.0
dtype: float64
# and
bool_col     True
dtype: object

Problem description

The same relates to min.

Expected Output

Output of pd.show_versions()

pandas : 1.2.1
numpy : 1.19.5
pytz : 2020.5
dateutil : 2.8.1
pip : 20.3.3
setuptools : 52.0.0.post20210125
Cython : None
pytest : 6.2.1
hypothesis : None
sphinx : 3.4.3
blosc : None
feather : 0.4.1
xlsxwriter : None
lxml.etree : 4.6.2
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.11.2
IPython : 7.19.0
pandas_datareader: None
bs4 : 4.9.3
bottleneck : None
fsspec : 0.8.3
fastparquet : None
gcsfs : None
matplotlib : 3.2.2
numexpr : 2.7.2
odfpy : None
openpyxl : 3.0.6
pandas_gbq : None
pyarrow : 2.0.0
pyxlsb : None
s3fs : 0.5.2
scipy : 1.6.0
sqlalchemy : 1.3.22
tables : 3.6.1
tabulate : None
xarray : 0.16.2
xlrd : 2.0.1
xlwt : None
numba : None

[jreback]

are you sure? the prior was coerced to float

[YarShev]

To me, 1.2.1 behavior looks more correct. I just wanted to make sure of this. If you think so too, the issue should be closed.

[jbrockmendel]

i think this is the result of the change to operate block-wise instead of on df.values

[jorisvandenbossche]

The 1.2 result preserves more information about the actual dtype of the original columns, but on the other hand it also results in a less-useful object dtype Series (and this object dtype can then propagate in following operations).

See also #39817, as this is also related to the question how to combine bool + numeric dtypes.

[jbrockmendel]

This is correct as-is, closing.