How to properly handle datetime comparisons in an entire DataFrame with NaT values?

user_4333359 0
Viewed 2074

I stumbled upon this odd behavior when trying to check if a DataFrame has values above a certain date, while that DataFrame may also contain pd.NaT

Comparisons of values behaves as expected:

import pandas as pd

pd.NaT > pd.to_datetime('2018-10-15')
# False

Comparisons with a Series also behave as expected:

s = pd.Series([pd.NaT, pd.to_datetime('2018-10-16')])
s > pd.to_datetime('2018-10-15')

#0    False
#1     True
#dtype: bool

But the DataFrame comparison isn't correct:

s.to_frame() > pd.to_datetime('2018-10-15')
#      0
#0  True
#1  True

It seems to me the issue is that the comparison initially returns NaN which is (at some point?) coerced to True given the behavior of:

df = pd.DataFrame([[pd.NaT, pd.to_datetime('2018-10-16')],
                   [pd.to_datetime('2018-10-16'), pd.NaT]])

df >= pd.to_datetime('2018-10-15')
#      0     1
#0  True  True
#1  True  True

df.ge(pd.to_datetime('2018-10-15'))
#     0    1
#0  NaN  1.0
#1  1.0  NaN

So can we really not use the > < >= <= operators when comparing for a DataFrame and need to rely on .lt .gt .le .ge followed by a .fillna(0)? 

df.ge(pd.to_datetime('2018-10-15')).fillna(0)
#     0    1
#0  0.0  1.0
#1  1.0  0.0
pandas dataframe datetime python
1 Answers

This was a bug that will be fixed in the next release of pandas (0.24.0):

In [1]: import pandas as pd; pd.__version__
Out[1]: '0.24.0.dev0+1504.g9642fea9c'

In [2]: s = pd.Series([pd.NaT, pd.to_datetime('2018-10-16')])

In [3]: s > pd.to_datetime('2018-10-15')
Out[3]:
0    False
1     True
dtype: bool

In [4]: s.to_frame() > pd.to_datetime('2018-10-15')
Out[4]:
       0
0  False
1   True

In [5]: df = pd.DataFrame([[pd.NaT, pd.to_datetime('2018-10-16')],
   ...:                    [pd.to_datetime('2018-10-16'), pd.NaT]])
   ...:

In [6]: df >= pd.to_datetime('2018-10-15')
Out[6]:
       0      1
0  False   True
1   True  False

In [7]: df.ge(pd.to_datetime('2018-10-15'))
Out[7]:
       0      1
0  False   True
1   True  False

For the corresponding GitHub issue, see: https://github.com/pandas-dev/pandas/issues/22242

user_33399650
Related