Consistent dtype output for element wise operations on empty dataframes

[benmatwil]

Code Sample, a copy-pastable example if possible

import pandas as pd
s = pd.Series([], dtype='datetime64[ns, UTC]')
df = pd.concat([s, s], axis=1) # empty dataframe of two columns of datetime dtype
# want the minimum of the two datetime columns element wise
min_dates = df.min(axis=1) # output series is of dtype float64

# it's fine if dataframe is non-empty
import pandas as pd
s = pd.Series(pd.to_datetime([f'2020-01-0{i}' for i in range(1, 10)], utc=True))
df = pd.concat([s, s], axis=1) # non-empty dataframe of two columns of datetime dtype
min_dates = df.min(axis=1) # output series is of dtype datetime64[ns, UTC]

Problem description

When dataframe is empty, doing a .min(axis=1) outputs a series of dtype float64 even though all columns were dtype datetime64.
Should this be consitent on dtypes and output min_dates with dtype datetime64[ns, UTC].

This is only an issue when the dataframe is empty.
If the dataframe is not empty, the dtype is conserved and min_dates outputs with dtype datetime64[ns, UTC].

Output of pd.show_versions()

INSTALLED VERSIONS

commit : None
python : 3.6.3.final.0
python-bits : 64
OS : Linux
OS-release : 3.10.0-862.3.3.el7.x86_64
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.None
pandas : 1.0.2
numpy : 1.16.5
pytz : 2019.3
dateutil : 2.8.0
pip : 20.0.2
setuptools : 41.4.0
Cython : 0.29.13
pytest : 5.0.1
hypothesis : None
sphinx : 2.1.2
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : 2.8.3 (dt dec pq3 ext lo64)
jinja2 : 2.10.1
IPython : 7.7.0
pandas_datareader: None
bs4 : 4.8.1
bottleneck : None
fastparquet : None
gcsfs : 0.3.1
lxml.etree : None
matplotlib : 3.1.1
numexpr : 2.6.9
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : 0.15.1
pytables : None
pytest : 5.0.1
pyxlsb : None
s3fs : None
scipy : 1.3.1
sqlalchemy : 1.3.8
tables : 3.5.2
tabulate : None
xarray : 0.14.0
xlrd : 1.2.0
xlwt : None
xlsxwriter : None
numba : 0.46.0

[arw2019]

A slightly tweaked version of the original report:

import pandas as pd 
s1 = pd.Series([], dtype='datetime64[ns, UTC]') 
df1 = pd.concat([s1, s1], axis=1) # empty dataframe of two columns of datetime dtype 
# want the minimum of the two datetime columns element wise 
min_dates_1 = df1.min(axis=1) # output series is of dtype float64 
print(min_dates_1.dtype) 
 
s2 = pd.Series(pd.to_datetime([f'2020-01-0{i}' for i in range(1, 10)], utc=True)) 
df2 = pd.concat([s2, s2], axis=1) # non-empty dataframe of two columns of datetime dtype 
min_dates_2 = df2.min(axis=1) # output series is of dtype datetime64[ns, UTC] 
print(min_dates_2.dtype) 

Also - I still get this on the master version of pandas.

Output of pd.show_versions()

INSTALLED VERSIONS

commit : 9e1b95f
python : 3.8.2.final.0
python-bits : 64
OS : Linux
OS-release : 4.15.0-106-generic
Version : #107-Ubuntu SMP Thu Jun 4 11:27:52 UTC 2020
machine : x86_64
processor :
byteorder : little
LC_ALL : C.UTF-8
LANG : C.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.1.0.dev0+2003.g9e1b95f72.dirty
numpy : 1.17.5
pytz : 2020.1
dateutil : 2.8.1
pip : 20.1.1
setuptools : 46.4.0.post20200518
Cython : 0.29.19
pytest : 5.4.2
hypothesis : 5.15.1
sphinx : 3.0.4
blosc : None
feather : None
xlsxwriter : 1.2.8
lxml.etree : 4.5.1
html5lib : 1.0.1
pymysql : None
psycopg2 : None
jinja2 : 2.11.2
IPython : 7.14.0
pandas_datareader: None
bs4 : 4.9.1
bottleneck : 1.3.2
fastparquet : 0.4.0
gcsfs : None
matplotlib : 3.2.1
numexpr : 2.7.1
odfpy : None
openpyxl : 3.0.3
pandas_gbq : None
pyarrow : 0.17.1
pytables : None
pyxlsb : None
s3fs : 0.4.2
scipy : 1.4.1
sqlalchemy : 1.3.17
tables : 3.6.1
tabulate : 0.8.7
xarray : 0.15.1
xlrd : 1.2.0
xlwt : 1.3.0
numba : 0.49.1