python - 在 Pandas 中重采样时的时间增量平均值

Question

给定这个数据框:

df = pd.DataFrame(pd.to_timedelta(['00:00:02','00:00:05','00:00:10','00:00:15','00:00:05']))
df.index = pd.to_datetime(['20/02/2017 12:42:10','20/02/2017 12:43:10','20/02/2017 12:45:10','20/02/2017 12:45:10','20/02/2017 12:45:10'])
df.columns = ['time']

df
Out[232]: 
                        time
2017-02-20 12:42:10  00:00:02
2017-02-20 12:43:10  00:00:05
2017-02-20 12:45:10  00:00:10
2017-02-20 12:45:10  00:00:15
2017-02-20 12:45:10  00:00:05

我尝试重新采样，获取每分钟的平均时间。P.e 它在对他们进行总结时起作用:

df.resample('min').sum()
Out[245]: 
                        time
2017-02-20 12:42:00 00:00:02
2017-02-20 12:43:00 00:00:05
2017-02-20 12:44:00 00:00:00
2017-02-20 12:45:00 00:00:30

有什么方法可以使这项工作变得有意义吗？

类似于:

df.resample('min').mean()

Answer 1

您可以先将 timedeltas 转换为 total_seconds ( float )，resample 并使用 fillna .上次转换 to_timedelta返回:

df = pd.to_timedelta(df.time.dt.total_seconds().resample('min').mean().fillna(0), unit='s')
print (df)
2017-02-20 12:42:00   00:00:02
2017-02-20 12:43:00   00:00:05
2017-02-20 12:44:00   00:00:00
2017-02-20 12:45:00   00:00:10
Freq: T, Name: time, dtype: timedelta64[ns]

通过转换为纳秒可以提高精度:

print (pd.Series(df.time.values.astype(np.int64), index=df.index))
2017-02-20 12:42:10     2000000000
2017-02-20 12:43:10     5000000000
2017-02-20 12:45:10    10000000000
2017-02-20 12:45:10    15000000000
2017-02-20 12:45:10     5000000000
dtype: int64

df = pd.to_timedelta(pd.Series(df.time.values.astype(np.int64), index=df.index)
                                 .resample('min').mean().fillna(0))
print (df)
2017-02-20 12:42:00   00:00:02
2017-02-20 12:43:00   00:00:05
2017-02-20 12:44:00   00:00:00
2017-02-20 12:45:00   00:00:10
Freq: T, dtype: timedelta64[ns]