Pandas时间序列：移动数据

import pandas as pd
import numpy as np

ts = pd.Series(np.random.randn(4),index=pd.date_range('1/1/2000',periods=4,freq='M'))
ts

2000-01-31    1.300295
2000-02-29   -1.472891
2000-03-31    1.590408
2000-04-30    0.600093
Freq: M, dtype: float64

一、移动数据

ts.shift(2) # 正向移动

2000-01-31         NaN
2000-02-29         NaN
2000-03-31    1.300295
2000-04-30   -1.472891
Freq: M, dtype: float64

ts.shift(-2) # 反向移动

2000-01-31    1.590408
2000-02-29    0.600093
2000-03-31         NaN
2000-04-30         NaN
Freq: M, dtype: float64

二、移动日期索引

通过参数freq指定频率来实现日期的移动

ts.shift(2,freq='M') #对索引进行移位

2000-03-31    1.300295
2000-04-30   -1.472891
2000-05-31    1.590408
2000-06-30    0.600093
Freq: M, dtype: float64

ts.shift(1,freq='3D')

2000-02-03    1.300295
2000-03-03   -1.472891
2000-04-03    1.590408
2000-05-03    0.600093
dtype: float64

三、通过日期偏移量对日期进行移动

from pandas.tseries.offsets import Day,MonthEnd
from datetime import datetime
now = datetime(2018,7,5)

now+3*Day()

Timestamp('2018-07-08 00:00:00')

MonthEnd是锚点偏移量，第1次增量后滚到符合频率规则的下一个日期

print(now+MonthEnd())
print(now+MonthEnd(2))

2018-07-31 00:00:00
2018-08-31 00:00:00

使用锚点偏移量进行前后滚动

print(MonthEnd().rollforward(now))
print(MonthEnd().rollback(now))

2018-07-31 00:00:00
2018-06-30 00:00:00

使用锚点偏移量进行groupby，使具有相同锚点的数据聚合

ts = pd.Series(np.random.randn(20),index=pd.date_range('1/15/2000',periods=20,freq='4d'))
ts

2000-01-15   -1.029260
2000-01-19    0.060567
2000-01-23   -0.846305
2000-01-27   -0.626401
2000-01-31   -1.070348
2000-02-04   -0.327185
2000-02-08    0.740582
2000-02-12   -0.725505
2000-02-16   -0.107928
2000-02-20    0.623038
2000-02-24    0.125203
2000-02-28    0.860681
2000-03-03   -0.160448
2000-03-07    0.021979
2000-03-11    0.873563
2000-03-15    0.091832
2000-03-19   -1.302709
2000-03-23   -0.854590
2000-03-27   -1.572235
2000-03-31   -0.536468
Freq: 4D, dtype: float64

ts.groupby(MonthEnd().rollforward).mean()  # 相同月份的数据求平均

2000-01-31   -0.702349
2000-02-29    0.169841
2000-03-31   -0.429885
dtype: float64