我有这两个数据框:
main_df:
value feed_id created_at
0 0.0 1010077.0 2019-03-06 07:38:18-05:00
1 1.0 1010077.0 2019-03-06 07:39:26-05:00
2 1.0 1010077.0 2019-03-06 07:40:33-05:00
3 1.0 1010077.0 2019-03-06 07:41:41-05:00
4 1.0 1010077.0 2019-03-06 07:42:49-05:00
5 1.0 1010077.0 2019-03-06 07:43:56-05:00
aux_df:
value feed_id created_at
0 20.298492 1009408.0 2019-03-06 07:35:33-05:00
1 20.315002 1009408.0 2019-03-06 07:36:34-05:00
2 20.315002 1009408.0 2019-03-06 07:37:36-05:00
3 20.359650 1009408.0 2019-03-06 07:38:36-05:00
4 20.359650 1009408.0 2019-03-06 07:39:37-05:00
5 20.383179 1009408.0 2019-03-06 07:40:38-05:00
6 20.383179 1009408.0 2019-03-06 07:41:38-05:00
7 20.449524 1009408.0 2019-03-06 07:42:39-05:00
8 20.449524 1009408.0 2019-03-06 07:43:40-05:00
9 20.521912 1009408.0 2019-03-06 07:44:41-05:00
在这种情况下,我需要以下 (final_df):我希望 aux_df 的“created_at”列中描述的“时间线”完全合并到 main_df 中,无论它在两列中是否具有公共(public)值。对于常见的,我采用整个时间戳并忽略以秒为单位的部分(注意所有值如何按相同的日期、小时和分钟而不是秒对齐)。
value feed_id created_at
0 nan nan 2019-03-06 07:35:33-05:00
1 nan nan 2019-03-06 07:36:34-05:00
2 nan nan 2019-03-06 07:37:36-05:00
3 0.0 1010077.0 2019-03-06 07:38:36-05:00
4 1.0 1010077.0 2019-03-06 07:39:37-05:00
5 1.0 1010077.0 2019-03-06 07:40:38-05:00
6 1.0 1010077.0 2019-03-06 07:41:38-05:00
7 1.0 1010077.0 2019-03-06 07:42:39-05:00
8 1.0 1010077.0 2019-03-06 07:43:40-05:00
9 nan nan 2019-03-06 07:44:41-05:00
我尝试过但没有成功的策略:
- 使用以下命令在名为“created_at_2”的两个数据框中创建一个新列 在每个时间戳上按分钟“舍入”,所以我可以丢弃 在我进行合并之前,时间戳的秒部分。
使用合并。
main_df['created_at_2'] = main_df.created_at.dt.round('min') aux_df['created_at_2'] = aux_df.created_at.dt.round('min') final_df = pd.merge(main_df, aux_df, on=['created_at_2'], how='inner')
但如本例所示,此方法并不稳健。当您四舍五入像 2019-03-06 07:40:33-05:00 这样的时间戳时,您将得到 41 分钟而不是 40 分钟。我需要一个连续的按分钟列。
我可以使用这个重新格式化时间戳时间轴:
main_df.created_at.map(lambda t: t.strftime('%Y-%m-%d %H:%M'))
aux_df.created_at.map(lambda t: t.strftime('%Y-%m-%d %H:%M'))
final_df = pd.merge(main_df, aux_df, on=['created_at_2'], how='inner')
但不确定此方法是否可靠,我仍然需要索引“created_at”列中不常见的值。那么,有没有更合适的方法来实现这一点?
提前致谢!
请您参考如下方法:
一个想法是使用 merge_asof ,但最后一行不同:
main_df['created_at'] = pd.to_datetime(main_df['created_at'])
aux_df['created_at'] = pd.to_datetime(aux_df['created_at'])
df = pd.merge_asof(aux_df[['created_at']], main_df, on=['created_at'])
print (df)
created_at value feed_id
0 2019-03-06 07:35:33-05:00 NaN NaN
1 2019-03-06 07:36:34-05:00 NaN NaN
2 2019-03-06 07:37:36-05:00 NaN NaN
3 2019-03-06 07:38:36-05:00 0.0 1010077.0
4 2019-03-06 07:39:37-05:00 1.0 1010077.0
5 2019-03-06 07:40:38-05:00 1.0 1010077.0
6 2019-03-06 07:41:38-05:00 1.0 1010077.0
7 2019-03-06 07:42:39-05:00 1.0 1010077.0
8 2019-03-06 07:43:40-05:00 1.0 1010077.0
9 2019-03-06 07:44:41-05:00 1.0 1010077.0
另一个是使用Series.dt.floor而不是 round:
main_df['created_at'] = pd.to_datetime(main_df['created_at'])
aux_df['created_at'] = pd.to_datetime(aux_df['created_at'])
main_df['created_at_2'] = main_df.created_at.dt.floor('min')
aux_df['created_at_2'] = aux_df.created_at.dt.floor('min')
df = pd.merge(aux_df[['created_at_2']], main_df, on=['created_at_2'], how='left')
print (df)
created_at_2 value feed_id created_at
0 2019-03-06 07:35:00-05:00 NaN NaN NaT
1 2019-03-06 07:36:00-05:00 NaN NaN NaT
2 2019-03-06 07:37:00-05:00 NaN NaN NaT
3 2019-03-06 07:38:00-05:00 0.0 1010077.0 2019-03-06 07:38:18-05:00
4 2019-03-06 07:39:00-05:00 1.0 1010077.0 2019-03-06 07:39:26-05:00
5 2019-03-06 07:40:00-05:00 1.0 1010077.0 2019-03-06 07:40:33-05:00
6 2019-03-06 07:41:00-05:00 1.0 1010077.0 2019-03-06 07:41:41-05:00
7 2019-03-06 07:42:00-05:00 1.0 1010077.0 2019-03-06 07:42:49-05:00
8 2019-03-06 07:43:00-05:00 1.0 1010077.0 2019-03-06 07:43:56-05:00
9 2019-03-06 07:44:00-05:00 NaN NaN NaT
