python - pandas.merge 在使用 tzinfo 合并时间戳列时失败

Question

我需要对时间戳列进行合并，但行为取决于是否设置了时区。

下面的代码可以正常工作

import pandas as pd, datetime
now = datetime.datetime.now()
df1 = pd.DataFrame({'ts': pd.to_datetime([now])})
df2 = pd.DataFrame({'ts': pd.to_datetime([now])})
pd.merge(df1, df2, on='ts')

相比之下，这不

import pandas as pd, datetime
now = datetime.datetime.now().replace(tzinfo=pytz.utc)
df3 = pd.DataFrame({'ts': pd.to_datetime([now])})
df4 = pd.DataFrame({'ts': pd.to_datetime([now])})
pd.merge(df3, df4, on='ts')

我有以下错误

/path/to/env3.4/lib/python3.4/site-packages/pandas/tools/merge.py in merge(left, right, how, on, left_on, right_on, left_index, right_index, sort, suffixes, copy, indicator)
     33                          right_index=right_index, sort=sort, suffixes=suffixes,
     34                          copy=copy, indicator=indicator)
---> 35     return op.get_result()
     36 if __debug__:
     37     merge.__doc__ = _merge_doc % '\nleft : DataFrame'

/path/to/env3.4/lib/python3.4/site-packages/pandas/tools/merge.py in get_result(self)
    194             self.left, self.right = self._indicator_pre_merge(self.left, self.right)
    195 
--> 196         join_index, left_indexer, right_indexer = self._get_join_info()
    197 
    198         ldata, rdata = self.left._data, self.right._data

/path/to/env3.4/lib/python3.4/site-packages/pandas/tools/merge.py in _get_join_info(self)
    323              right_indexer) = _get_join_indexers(self.left_join_keys,
    324                                                  self.right_join_keys,
--> 325                                                  sort=self.sort, how=self.how)
    326 
    327             if self.right_index:

/path/to/env3.4/lib/python3.4/site-packages/pandas/tools/merge.py in _get_join_indexers(left_keys, right_keys, sort, how)
    514 
    515     # get left & right join labels and num. of levels at each location
--> 516     llab, rlab, shape = map(list, zip( * map(fkeys, left_keys, right_keys)))
    517 
    518     # get flat i8 keys from label lists

TypeError: type object argument after * must be a sequence, not map

我的环境:

1. python 3.4
2. Pandas 0.17.0
3. numpy 1.10.1

dtype 不同:

In [4]: df1.dtypes
Out[4]: 
ts    datetime64[ns]
dtype: object

In [5]: df3.dtypes
Out[5]: 
ts    datetime64[ns, UTC]
dtype: object

pandas 中日期的处理有点神秘。你必须知道:

• 如果您操纵 datetime.datetime、numpy.datetime64 或 pandas.Timestamp
• 有或没有时区
• 秒/毫秒/微秒/纳秒精度

我在这里错过了什么？

Answer 1

这是 0.17.0 中的错误，已在 master 中修复 here ，并将在即将发布的 0.17.1 中发布。

不包括在内

In [13]: now = datetime.datetime.now()

In [14]: df1 = pd.DataFrame({'ts': pd.to_datetime([now])})

In [15]: df2 = pd.DataFrame({'ts': pd.to_datetime([now])})

In [16]: pd.merge(df1, df2, on='ts')
Out[16]: 
                          ts
0 2015-11-01 18:33:59.771962

与兹

In [8]: now = datetime.datetime.now().replace(tzinfo=pytz.utc)

In [9]: df3 = pd.DataFrame({'ts': pd.to_datetime([now])})

In [10]: df4 = pd.DataFrame({'ts': pd.to_datetime([now])})

In [11]: pd.merge(df3, df4, on='ts')
Out[11]: 
                                ts
0 2015-11-01 18:32:46.801009+00:00