Repository navigation
nanosecond precsion datetime converts to int in stead of datetime #18363
Description
When converting nanosecond precision np.datetime64 to datetime.datetime, an int is returned, whereas a datetime.datetime would be desireable/expected.
Reproducing code example:
from this SO thread. I also explained it in this xarray issue
import numpy as np
print('second precision')
t = np.datetime64('2000-01-01 00:00:00')
print(t)
print(t.astype(datetime.datetime))
print(t.item())
print('microsecond precision')
t = np.datetime64('2000-01-01 00:00:00.0000')
print(t)
print(t.astype(datetime.datetime))
print(t.item())
print('nanosecond precision')
t = np.datetime64('2000-01-01 00:00:00.0000000')
print(t)
print(t.astype(datetime.datetime))
print(t.item())
import pandas as pd
print(pd.to_datetime(str(t)))
second precision
2000-01-01T00:00:00
2000-01-01 00:00:00
2000-01-01 00:00:00
microsecond precision
2000-01-01T00:00:00.000000
2000-01-01 00:00:00
2000-01-01 00:00:00
nanosecond precision
2000-01-01T00:00:00.000000000
946684800000000000
946684800000000000
2000-01-01 00:00:00Error message:
none,
NumPy/Python version information:
numpy 1.20.0
sys:
3.7.9 | packaged by conda-forge | (default, Dec 9 2020, 21:08:20)
[GCC 9.3.0]
Reacted by drf5n
Activity
feefladder commentedon Feb 8, 2021
In the specifications of datetime:
it says that conversion to datetime should lose precision or overflow as needed. Not convert to int
Also note that explicit conversion. i.e.
t.astype('datetime[us]').astype(datetime.datetime)works as expected.
gerritholl commentedon Nov 11, 2021
Duplicate of #7619
drf5n commentedon Oct 20, 2023
Maybe this instead:
or
Either way it's casting the ns precision into a datetime-compatible us precision.
InessaPawson commentedon Oct 17, 2026
Still reproduces on 2.5.3. The workaround from this thread,
t.astype('datetime64[us]').item(), remains the way to go in the meantime.Closing as a duplicate of #7619, which tracks the same underlying problem and has the discussion on what the right fix is.