Maximum of time series with NaT returns a dtype float64 with only NaN's

[MaximeWeyl]

Code Sample, a copy-pastable example if possible

import pandas as pd

show_bug = True

a = pd.date_range("2019-1-1", periods=10, freq="1D", tz="Europe/Paris")
b = pd.date_range("2019-1-3", periods=10, freq="1H", tz="Europe/Paris")

if show_bug:
    a = pd.Index([pd.NaT] + a.to_list()[1:])



multi = pd.MultiIndex.from_arrays([a, b])
m = multi.to_frame(index=False).max(axis=1)
print(m)

Problem description

When show_bug is False, no NaT's are present. This gives this output :

0   2019-01-03 00:00:00+01:00
1   2019-01-03 01:00:00+01:00
2   2019-01-03 02:00:00+01:00
3   2019-01-04 00:00:00+01:00
4   2019-01-05 00:00:00+01:00
5   2019-01-06 00:00:00+01:00
6   2019-01-07 00:00:00+01:00
7   2019-01-08 00:00:00+01:00
8   2019-01-09 00:00:00+01:00
9   2019-01-10 00:00:00+01:00
dtype: datetime64[ns, Europe/Paris]

When show_bug is True, all it does is replacing the first value in a with a NaT.
I expect the output to be the same, except for the first row being a NaT.

Here is what I get for show_bug=True :

0   NaN
1   NaN
2   NaN
3   NaN
4   NaN
5   NaN
6   NaN
7   NaN
8   NaN
9   NaN
dtype: float64

Expected Output

0   NaT
1   2019-01-03 01:00:00+01:00
2   2019-01-03 02:00:00+01:00
3   2019-01-04 00:00:00+01:00
4   2019-01-05 00:00:00+01:00
5   2019-01-06 00:00:00+01:00
6   2019-01-07 00:00:00+01:00
7   2019-01-08 00:00:00+01:00
8   2019-01-09 00:00:00+01:00
9   2019-01-10 00:00:00+01:00
dtype: datetime64[ns, Europe/Paris]

Output of pd.show_versions()

INSTALLED VERSIONS

commit : None
python : 3.7.3.final.0
python-bits : 64
OS : Windows
OS-release : 10
machine : AMD64
processor : Intel64 Family 6 Model 142 Stepping 9, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.None

pandas : 0.25.1
numpy : 1.17.0
pytz : 2019.2
dateutil : 2.8.0
pip : 19.0.3
setuptools : 40.8.0
Cython : None
pytest : 4.6.5
hypothesis : 4.33.1
sphinx : 2.2.0
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.4.1
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.10.1
IPython : 7.7.0
pandas_datareader: None
bs4 : 4.8.0
bottleneck : None
fastparquet : None
gcsfs : None
lxml.etree : 4.4.1
matplotlib : 3.1.1
numexpr : None
odfpy : None
openpyxl : 2.6.3
pandas_gbq : None
pyarrow : None
pytables : None
s3fs : None
scipy : 1.3.1
sqlalchemy : None
tables : None
xarray : None
xlrd : 1.2.0
xlwt : None
xlsxwriter : None

[MaximeWeyl]

Here is a workaround I intend to use while this bug exists :

import pandas as pd

insert_nat = True

def column_max_without_nans(df):
    return df[df.notnull().all(axis=1)].max(axis=1).reindex(df.index)


a = pd.date_range("2019-1-1", periods=10, freq="1D", tz="Europe/Paris")
b = pd.date_range("2019-1-3", periods=10, freq="1H", tz="Europe/Paris")

if insert_nat:
    a = pd.Index([pd.NaT] + a.to_list()[1:])



multi = pd.MultiIndex.from_arrays([a, b])
m = column_max_without_nans(multi.to_frame(index=False))
print(m)

Wich gives me what I want :

0                         NaT
1   2019-01-03 01:00:00+01:00
2   2019-01-03 02:00:00+01:00
3   2019-01-04 00:00:00+01:00
4   2019-01-05 00:00:00+01:00
5   2019-01-06 00:00:00+01:00
6   2019-01-07 00:00:00+01:00
7   2019-01-08 00:00:00+01:00
8   2019-01-09 00:00:00+01:00
9   2019-01-10 00:00:00+01:00
dtype: datetime64[ns, Europe/Paris]

[mroeschke]

Duplicate of #27794. Investigations and PR's welcome!