Python 使用pandas.merge\u asof进行完全外部连接
您好,我需要将一些时间序列数据与最近的时间戳对齐,因此我认为Python 使用pandas.merge\u asof进行完全外部连接,python,pandas,dataframe,merge,outer-join,Python,Pandas,Dataframe,Merge,Outer Join,您好,我需要将一些时间序列数据与最近的时间戳对齐,因此我认为pandas.merge\u asof可能是一个很好的候选者。但是,它没有在标准的merge方法中设置how='outer'的选项 例如: df1: df2: 然后,例如,执行以下操作: pd.merge_asof(df1, df2, left_index=True, right_index=True, direction='nearest', tolerance=pd.Timedelta('0.3s')) 结果将是:
pandas.merge\u asof
可能是一个很好的候选者。但是,它没有在标准的merge
方法中设置how='outer'
的选项
例如:
df1:
df2:
然后,例如,执行以下操作:
pd.merge_asof(df1, df2, left_index=True, right_index=True, direction='nearest', tolerance=pd.Timedelta('0.3s'))
结果将是:
Value1 Value2
Time
2020-07-17 14:25:03.535906075 108 222.0
2020-07-17 14:25:05.457247019 110 NaN
2020-07-17 14:25:07.467777014 126 60.0
但我想要的是:
Value1 Value2
Time
2020-07-17 14:25:03.535906075 108 222.0
2020-07-17 14:25:04.545104980 NaN 150.0 <---- this is the difference
2020-07-17 14:25:05.457247019 110 NaN
2020-07-17 14:25:07.467777014 126 60.0
Value1值2
时间
2020-07-17 14:25:03.535906075 108 222.0
2020-07-17 14:25:04.545104980南150.0
不幸的是,pd.merge\u asof
中没有与pd.merge
类似的how
参数,否则您只需传递how='outer'
作为一种解决方法,您可以手动添加另一个数据帧中不匹配的值
然后,使用.sort\u index()
这似乎很简单,但没有直接的解决办法。有一个选项可以再次合并,以引入缺少的行:
# enumerate the rows of `df2` to later identify which are missing
df2 = df2.reset_index().assign(idx=np.arange(df2.shape[0]))
(pd.merge_asof(df1.reset_index(),
df2[['Time','idx']],
on='Time',
direction='nearest',
tolerance=pd.Timedelta('0.3s'))
.merge(df2, on='idx', how='outer') # merge back on row number
.assign(Time=lambda x: x['Time_x'].fillna(x['Time_y'])) # fill the time
.set_index(['Time']) # set index back
.drop(['Time_x','Time_y','idx'], axis=1)
.sort_index()
)
Value1 Value2
Time
2020-07-17 14:25:03.535906075 108.0 222.0
2020-07-17 14:25:04.545104980 NaN 150.0
2020-07-17 14:25:05.457247019 110.0 NaN
2020-07-17 14:25:07.467777014 126.0 60.0
嗨,谢谢!您认为什么是合并2个以上数据帧的好方法?请查看更新的问题。@circle999需要其他解决方案。你能创建一个新问题并引用回这个问题吗?您可以复制和粘贴所有数据,并添加多个示例数据帧(如3而不是2)。一般不赞成像这样更新问题。嗨,谢谢!您认为什么是合并2个以上数据帧的好方法?请查看更新的问题。
Value1 Value2
Time
2020-07-17 14:25:03.535906075 108 222.0
2020-07-17 14:25:04.545104980 NaN 150.0 <---- this is the difference
2020-07-17 14:25:05.457247019 110 NaN
2020-07-17 14:25:07.467777014 126 60.0
df3 = pd.merge_asof(df1, df2, left_index=True, right_index=True, direction='nearest', tolerance=pd.Timedelta('0.3s'))
df4 = pd.merge_asof(df2, df1, left_index=True, right_index=True, direction='nearest', tolerance=pd.Timedelta('0.3s'))
df5 = df3.append(df4[df4['Value1'].isnull()]).sort_index()
df5
Out[1]:
Value1 Value2
Time
2020-07-17 14:25:03.535906075 108.0 222.0
2020-07-17 14:25:04.545104980 NaN 150.0
2020-07-17 14:25:05.457247019 110.0 NaN
2020-07-17 14:25:07.467777014 126.0 60.0
# enumerate the rows of `df2` to later identify which are missing
df2 = df2.reset_index().assign(idx=np.arange(df2.shape[0]))
(pd.merge_asof(df1.reset_index(),
df2[['Time','idx']],
on='Time',
direction='nearest',
tolerance=pd.Timedelta('0.3s'))
.merge(df2, on='idx', how='outer') # merge back on row number
.assign(Time=lambda x: x['Time_x'].fillna(x['Time_y'])) # fill the time
.set_index(['Time']) # set index back
.drop(['Time_x','Time_y','idx'], axis=1)
.sort_index()
)
Value1 Value2
Time
2020-07-17 14:25:03.535906075 108.0 222.0
2020-07-17 14:25:04.545104980 NaN 150.0
2020-07-17 14:25:05.457247019 110.0 NaN
2020-07-17 14:25:07.467777014 126.0 60.0