Unable to compare timedelta64 to timedelta

[cancan101]

In [426]:
s = pd.Series([timedelta(days=1), timedelta(days=2)])
s > timedelta(days=1)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-426-9094b7c93677> in <module>()
      1 s = pd.Series([timedelta(days=1), timedelta(days=2)])
----> 2 s > timedelta(days=1)

/usr/local/lib/python2.7/dist-packages/pandas-0.13.0rc1_78_g142ca62-py2.7-linux-x86_64.egg/pandas/core/ops.pyc in wrapper(self, other)
    561 
    562             # scalars
--> 563             res = na_op(values, other)
    564             if np.isscalar(res):
    565                 raise TypeError('Could not compare %s type with Series'

/usr/local/lib/python2.7/dist-packages/pandas-0.13.0rc1_78_g142ca62-py2.7-linux-x86_64.egg/pandas/core/ops.pyc in na_op(x, y)
    528 
    529             try:
--> 530                 result = getattr(x, name)(y)
    531                 if result is NotImplemented:
    532                     raise TypeError("invalid type comparison")

TypeError: can't compare datetime.timedelta to long

The comparison does work for timedelta64:

In [45]: s > np.timedelta64(timedelta(days=1))
Out[45]: 
0    False
1     True
dtype: bool

[cancan101]

This might be related: #5436

[jreback]

ok...this is legit...just need some coercion

can you update to show if timedelta64 works?

[dieterv77]

I ran into the same issue with pandas master, though i get a different exception (maybe a difference in numpy version?):

import pandas as pd
import numpy as np
from datetime import timedelta
print np.__version__
s = pd.Series([timedelta(days=1), timedelta(days=2)])
s > timedelta(days=1)

prints

1.6.1
Traceback (most recent call last):
  File "bug.py", line 9, in <module>
    s > timedelta(days=1)
  File "/home/dvandenbussche/.local/lib/python2.7/site-packages/pandas-0.13.0_472_g9109812-py2.7-linux-x86_64.egg/pandas/core/ops.py", line 563, in wrapper
    res = na_op(values, other)
  File "/home/dvandenbussche/.local/lib/python2.7/site-packages/pandas-0.13.0_472_g9109812-py2.7-linux-x86_64.egg/pandas/core/ops.py", line 530, in na_op
    result = getattr(x, name)(y)
TypeError: ufunc 'greater' not supported for the input types, and the inputs could not be safely coerced to any supported types according to the casting rule 'safe'

FWIW: this was working with 0.10.1.
@jreback, i'm not too well informed about all the different date related types, but if you think it is feasible/sensible to give me some instructions, i can try and take a look.
In the end, the code tries to call the comparison operator of a numpy array (with dtype timedelta64[ns]) with a datetime.timedelta type. When you say coercion is needed, are you talking about the numpy array or the datetime.timedelta?

Thanks!

[jreback]

@dieterv77

timedeltas have been steadily moving towards being kept as timedelta64[ns] < 10 I think were object arrays.

in any event if you want pretty much anything to work with timedeltas you need numpy >= 1.7

1.6.1 is barely supported (most tests are skipped) as it is so thoroughly broken (datetime64 in 1.6.1 at least work somewhat)

so welcome to take a look with 1.6.1 but its a nightmare (serieously)...

this issue is pretty straightforward in numpy >= 1.7

all of the timedelta conversions are done here: https://github.com/pydata/pandas/blob/master/pandas/core/ops.py#L243