df[col].dtype==object reverts back to df[col].dtype==datetime upon df.replace({np.nan: None})

[rdmtinez]

I explicitly coerce the dtype==datetime of a column containing NaT and datetimes (call it: DateTimeCol) to be of type object so that I can convert NaT to None (because Django complains about it). Then later during the workflow, when I have a need to replace all other NaN to None (because Django also compalins about it) using df.replace({np.nan: None}), the DateTimeCol is retransformed back into a type: datetime automatically without it explicit input and the NaT are present again which forces django to complain. Surely this is not a wanted behaviour.

`

pre replace

0 None
1 None
2 None
3 None
4 None
...
163 2013-12-19 16:30:00+01:00
164 2013-12-19 16:30:00+01:00
165 2013-12-19 16:30:00+01:00
166 2013-12-19 16:30:00+01:00
167 2013-12-19 16:30:00+01:00
Name: Tempering DateTime, Length: 168, dtype: object

df.replace({np.nan: None})

post replace

0 NaT
1 NaT
2 NaT
3 NaT
4 NaT
...
163 2013-12-19 16:30:00+01:00
164 2013-12-19 16:30:00+01:00
165 2013-12-19 16:30:00+01:00
166 2013-12-19 16:30:00+01:00
167 2013-12-19 16:30:00+01:00
Name: Tempering DateTime, Length: 168, dtype: datetime64[ns, Europe/Berlin]
`

[mroeschke]

We would need a minimal, copy-pastable example: https://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports

[jbrockmendel]

I think the OP's issue is this:

import numpy as np
import pandas as pd

ser = pd.Series([None, None, pd.Timestamp.now()], dtype=object)
res = ser.replace({np.nan: None})  # <- would expect this to be a no-op, but it casts to dt64

assert res.dtype == object  # raises in master