Python 3.x 有没有一种方法可以在不将数据类型更改为对象的情况下，将NaT附加到带有时区的datetime？

我的数据框中有一列数据类型为：datetime64[ns，UTC]。当我在该列中附加一行None或NaT时，该列的数据类型将更改为“object”。这不会发生在数据类型为datetime64[ns]的列上

下面是一个演示：

# Test pandas with datetime columns
import pandas as pd
from datetime import datetime, timezone
df = pd.DataFrame([{'D': datetime.utcnow()}])
df_wtz = pd.DataFrame([{'D': datetime.now().astimezone(timezone.utc)}])
df_None = pd.DataFrame([{'D': None}])
# Note that the tz below is ignored even though specified
df_Nat = pd.DataFrame([{'D': pd.Timestamp(None,tz=timezone.utc)}])

print('df:\n', df['D'])
print('df_wtz:\n', df_wtz['D'])
print('df_None:\n', df_None['D'])
print('df_Nat:\n', df_Nat['D'])

print('df append df_None:\n', df.append(df_None, ignore_index=True, sort=False)['D'])
print('df append df_Nat:\n', df.append(df_Nat, ignore_index=True, sort=False)['D'])

print('df_wtz append df_None:\n', df_wtz.append(df_None, ignore_index=True, sort=False)['D'])
print('df_wtz append df_Nat:\n', df_wtz.append(df_Nat, ignore_index=True, sort=False)['D'])

以下是输出：

df:
 0   2019-08-13 19:58:18.811492
Name: D, dtype: datetime64[ns]
df_wtz:
 0   2019-08-13 19:58:18.811968+00:00
Name: D, **dtype: datetime64[ns, UTC]**
df_None:
 0    None
Name: D, dtype: object
df_Nat:
 0   NaT
Name: D, dtype: datetime64[ns]
df append df_None:
 0   2019-08-13 19:58:18.811492
1                          NaT
Name: D, dtype: datetime64[ns]
df append df_Nat:
 0   2019-08-13 19:58:18.811492
1                          NaT
Name: D, dtype: datetime64[ns]
df_wtz append df_None:
 0    2019-08-13 19:58:18.811968+00:00
1                                None
Name: D, dtype: object
df_wtz append df_Nat:
 0    2019-08-13 19:58:18.811968+00:00
1                                 NaT
Name: D, dtype: object

---

我原本希望在datetime64[ns，UTC]列追加None或NaT时保留列类型，但事实并非如此。这是预期的行为还是会被视为错误？

您可以通过以下方式将NaT放置在带有dtype

datetime64[ns，UTC]

的列中：

 In [380]: df_Nat = pd.DataFrame({'D': pd.to_datetime([None], utc=True)}); df_Nat
 Out[380]: 
     D
 0 NaT

 In [381]: df_Nat.info()
 <class 'pandas.core.frame.DataFrame'>
 RangeIndex: 1 entries, 0 to 0
 Data columns (total 1 columns):
 D    0 non-null datetime64[ns, UTC]
 dtypes: datetime64[ns, UTC](1)
 memory usage: 88.0 bytes

屈服

df_wtz append df_Nat:
 0   2019-08-13 20:28:15.928023+00:00
1                                NaT
Name: D, dtype: datetime64[ns, UTC]

---

NaT本身不知道时区：

In [383]: pd.Timestamp(None) is pd.Timestamp(None, tz=utc)
Out[383]: True

因此，

pd.DataFrame（[{'D'：pd.Timestamp（None，tz=utc）}]）

不会生成具有时区感知数据类型的列

由于无法使数据帧从NaT本身推断出时区感知数据类型， 我们需要构建一个容器（如Series或DatetimeIndex），它已经具有正确的时区感知数据类型。这就是pd.to_datetime（[None]，utc=True）所做的：

In [385]: pd.to_datetime([None], utc=True)
Out[385]: DatetimeIndex(['NaT'], dtype='datetime64[ns, UTC]', freq=None)

---

谢谢-这对我有用。我仍然很遗憾，添加None会修改列类型，尽管。。。

In [385]: pd.to_datetime([None], utc=True)
Out[385]: DatetimeIndex(['NaT'], dtype='datetime64[ns, UTC]', freq=None)