I want to make array of datetime given by interval of months. It easy if I use days as interval like this
xyz = np.arange(np.datetime64('2020-03-24'), 3)
xyz
OUTPUT
array(['2020-03-24', '2020-03-25', '2020-03-26'], dtype='datetime64[D]')
It only incremental by 3 dyas. How about 3 months? I have tried this way and ERROR
np.arange(datetime('2020-03-28'), np.timedelta64(3,'M'))
I have tried this and Giving wrong result
np.arange(np.datetime64("2020-03-24"), np.datetime64("2020-06-24"),
np.timedelta64(1, 'M'),
dtype='datetime64[M]').astype('datetime64[D]')
OUTPUT
array(['2020-03-01', '2020-04-01', '2020-05-01'], dtype='datetime64[D]')
Your arange without the dtype raises an error:
In [91]: x = np.arange(np.datetime64("2020-03-24"), np.datetime64("2020-06-24"),
...: np.timedelta64(1, 'M'))
...
TypeError: Cannot get a common metadata divisor for NumPy datetime metadata [M] and [D] because they have incompatible nonlinear base time units
Stepping by one month is not the same as stepping by n days.
With the dtype:
In [85]: x = np.arange(np.datetime64("2020-03-24"), np.datetime64("2020-06-24"),
...: np.timedelta64(1, 'M'),
...: dtype='datetime64[M]')
In [86]: x
Out[86]: array(['2020-03', '2020-04', '2020-05'], dtype='datetime64[M]')
The end points have been converted to month (without any implied date).
Note that the differences are the expected 1 month:
In [87]: np.diff(x)
Out[87]: array([1, 1], dtype='timedelta64[M]')
If I convert the dates to D dtype, it chooses the start of the month:
In [89]: x.astype('datetime64[D]')
Out[89]: array(['2020-03-01', '2020-04-01', '2020-05-01'], dtype='datetime64[D]')
The date time delta is no longer uniform:
In [90]: np.diff(x.astype('datetime64[D]'))
Out[90]: array([31, 30], dtype='timedelta64[D]')
===
Instead of astype, you could add the appropriate timedelta:
In [96]: x + np.array(3, 'timedelta64[D]')
Out[96]: array(['2020-03-04', '2020-04-04', '2020-05-04'], dtype='datetime64[D]')