Numpy check if time delta is negative-CodePudding

I have an array of values that can be any type. When they are time-deltas, it causes a problem:

>>> d = np.diff(np.array([
...         '2022-10-13 10:00:00', '2022-10-13 10:30:00',
...         '2022-10-13 11:00:00', '2022-10-14 10:00:00',
...         '2022-10-13 23:00:00', '2022-10-14 00:00:00'
...     ], dtype='datetime64[ns]'))
>>> d
array([  1800000000000,   1800000000000,  82800000000000, -39600000000000,
         3600000000000], dtype='timedelta64[ns]')

I want get a mask for the elements that are less than zero:

>>> d < 0
UFuncTypeError: Cannot cast ufunc 'less' input 0 from dtype('<m8[ns]') to dtype('<m8') with casting rule 'same_kind'

Since I want a general purpose solution that works for any dtype, I tried using d[0] as a template, also without luck:

>>> d < d.dtype(0)
TypeError: 'numpy.dtype[timedelta64]' object is not callable

The only way I've found to bypass this by constructing a special zero:

>>> if d.dtype.type == np.timedelta64: print(d < np.timedelta64(0, 'ns'))
[False False False  True False]

Is there a general purpose approach that will let me check when an array is negative that will work with time-deltas?

CodePudding user response：

I think you can use np.sign.

np.sign(d)

Output:

array([ 1,  1,  1, -1,  1], dtype='timedelta64[ns]')

CodePudding user response：

I realized that I almost had it:

>>> d < d.dtype.type(0)
array([False, False, False,  True, False])