一、时间序列常用方法

• 对时间做一些移动/滞后、频率转换、采样等相关操作

# 导包
import numpy as np
import pandas as pd

index = pd.date_range("2024-2-8",periods=365,freq="D")
ts = pd.Series(np.random.randint(0,500,len(index)),index=index)
ts
# 执行结果
2024-02-08    381
2024-02-09    120
2024-02-10    209
2024-02-11     33
2024-02-12    367
             ... 
2025-02-02     56
2025-02-03    395
2025-02-04    129
2025-02-05    129
2025-02-06     71
Freq: D, Length: 365, dtype: int32

• 移动

# 默认后移1位
ts.shift()
# 执行结果
2024-02-08      NaN
2024-02-09    381.0
2024-02-10    120.0
2024-02-11    209.0
2024-02-12     33.0
              ...  
2025-02-02     12.0
2025-02-03     56.0
2025-02-04    395.0
2025-02-05    129.0
2025-02-06    129.0
Freq: D, Length: 365, dtype: float64

# 后移2位
ts.shift(periods=2)
# 执行结果
2024-02-08      NaN
2024-02-09      NaN
2024-02-10    381.0
2024-02-11    120.0
2024-02-12    209.0
              ...  
2025-02-02    349.0
2025-02-03     12.0
2025-02-04     56.0
2025-02-05    395.0
2025-02-06    129.0
Freq: D, Length: 365, dtype: float64

ts.shift(periods=-2)
# 执行结果
2024-02-08    209.0
2024-02-09     33.0
2024-02-10    367.0
2024-02-11    154.0
2024-02-12     43.0
              ...  
2025-02-02    129.0
2025-02-03    129.0
2025-02-04     71.0
2025-02-05      NaN
2025-02-06      NaN
Freq: D, Length: 365, dtype: float64

• 频率转换

# 天 -> 星期
ts.asfreq(pd.tseries.offsets.Week())
# 执行结果
2024-02-08    381
2024-02-15     68
2024-02-22    432
2024-02-29    191
2024-03-07    490
2024-03-14    297
2024-03-21    244
2024-03-28    405
2024-04-04    129
2024-04-11    492
2024-04-18    351
2024-04-25    226
2024-05-02    460
2024-05-09    375
2024-05-16    296
2024-05-23    428
2024-05-30    379
2024-06-06    431
2024-06-13    296
2024-06-20    274
2024-06-27    321
2024-07-04    242
2024-07-11    110
2024-07-18    298
2024-07-25    236
2024-08-01    281
2024-08-08     81
2024-08-15    415
2024-08-22    489
2024-08-29     14
2024-09-05     80
2024-09-12    197
2024-09-19    245
2024-09-26    261
2024-10-03    240
2024-10-10    324
2024-10-17    474
2024-10-24    283
2024-10-31     44
2024-11-07     98
2024-11-14    380
2024-11-21     13
2024-11-28     42
2024-12-05    292
2024-12-12     95
2024-12-19    147
2024-12-26    348
2025-01-02    379
2025-01-09     54
2025-01-16    235
2025-01-23      8
2025-01-30    479
2025-02-06     71
Freq: W, dtype: int32

# 天 -> 月
ts.asfreq(pd.tseries.offsets.MonthEnd())
# 执行结果
2024-02-29    191
2024-03-31    237
2024-04-30    344
2024-05-31    440
2024-06-30    425
2024-07-31    268
2024-08-31    325
2024-09-30    307
2024-10-31     44
2024-11-30     91
2024-12-31    289
2025-01-31    349
Freq: M, dtype: int32

# 天 -> 小时，数据由少变多：fill_value填充
ts.asfreq(pd.tseries.offsets.Hour(),fill_value=0)
# 执行结果
2024-02-08 00:00:00    381
2024-02-08 01:00:00      0
2024-02-08 02:00:00      0
2024-02-08 03:00:00      0
2024-02-08 04:00:00      0
                      ... 
2025-02-05 20:00:00      0
2025-02-05 21:00:00      0
2025-02-05 22:00:00      0
2025-02-05 23:00:00      0
2025-02-06 00:00:00     71
Freq: H, Length: 8737, dtype: int32

二、resample：根据日期维度进行数据聚合

• 按照分钟（T）、小时（H）、日（D）、周（W）、月（M）、年（Y）等来作为日期维度

# 导包
import numpy as np
import pandas as pd

index = pd.date_range("2024-2-8",periods=365,freq="D")
ts = pd.Series(np.random.randint(0,500,len(index)),index=index)
ts
# 执行结果
2024-02-08    484
2024-02-09     23
2024-02-10    135
2024-02-11    407
2024-02-12    169
             ... 
2025-02-02    358
2025-02-03     52
2025-02-04    140
2025-02-05    305
2025-02-06    227
Freq: D, Length: 365, dtype: int32

# 重采样 resample
ts.resample("D").sum()   # 以1天为单位进行汇总，求和
# 执行结果
2024-02-08    484
2024-02-09     23
2024-02-10    135
2024-02-11    407
2024-02-12    169
             ... 
2025-02-02    358
2025-02-03     52
2025-02-04    140
2025-02-05    305
2025-02-06    227
Freq: D, Length: 365, dtype: int32

ts.resample("2D").sum()   # 以2天为单位进行汇总，求和
# 执行结果
2024-02-08    507
2024-02-10    542
2024-02-12    397
2024-02-14    696
2024-02-16    670
             ... 
2025-01-29    696
2025-01-31    911
2025-02-02    410
2025-02-04    445
2025-02-06    227
Freq: 2D, Length: 183, dtype: int32

ts.resample("2W").sum()   # 以2周为单位进行汇总，求和
# 执行结果
2024-02-11    1049
2024-02-25    3436
2024-03-10    3220
2024-03-24    3876
2024-04-07    3139
2024-04-21    3295
2024-05-05    3379
2024-05-19    3297
2024-06-02    3369
2024-06-16    2960
2024-06-30    4095
2024-07-14    2866
2024-07-28    2571
2024-08-11    3603
2024-08-25    3739
2024-09-08    3119
2024-09-22    2961
2024-10-06    2966
2024-10-20    3809
2024-11-03    3701
2024-11-17    3496
2024-12-01    4087
2024-12-15    3275
2024-12-29    3301
2025-01-12    3018
2025-01-26    3032
2025-02-09    3260
Freq: 2W-SUN, dtype: int32

ts.resample("3M").sum()   # 以3个月（季度）为单位进行汇总，求和
# 执行结果
2024-02-29     5418
2024-05-31    22268
2024-08-31    21430
2024-11-30    22536
2025-02-28    16267
Freq: 3M, dtype: int32

ts.resample("3M").sum().cumsum()   # 以3个月（季度）为单位进行汇总，求和，累加
# 执行结果
2024-02-29     5418
2024-05-31    27686
2024-08-31    49116
2024-11-30    71652
2025-02-28    87919
Freq: 3M, dtype: int32

ts.resample("H").sum()   # 以1小时为单位进行汇总，求和
# 执行结果
2024-02-08 00:00:00    484
2024-02-08 01:00:00      0
2024-02-08 02:00:00      0
2024-02-08 03:00:00      0
2024-02-08 04:00:00      0
                      ... 
2025-02-05 20:00:00      0
2025-02-05 21:00:00      0
2025-02-05 22:00:00      0
2025-02-05 23:00:00      0
2025-02-06 00:00:00    227
Freq: H, Length: 8737, dtype: int32

ts.resample("T").sum()   # 以1分钟为单位进行汇总，求和
# 执行结果
2024-02-08 00:00:00    484
2024-02-08 00:01:00      0
2024-02-08 00:02:00      0
2024-02-08 00:03:00      0
2024-02-08 00:04:00      0
                      ... 
2025-02-05 23:56:00      0
2025-02-05 23:57:00      0
2025-02-05 23:58:00      0
2025-02-05 23:59:00      0
2025-02-06 00:00:00    227
Freq: T, Length: 524161, dtype: int32

ts.resample("S").sum()   # 以1秒为单位进行汇总，求和
# 执行结果
2024-02-08 00:00:00    484
2024-02-08 00:00:01      0
2024-02-08 00:00:02      0
2024-02-08 00:00:03      0
2024-02-08 00:00:04      0
                      ... 
2025-02-05 23:59:56      0
2025-02-05 23:59:57      0
2025-02-05 23:59:58      0
2025-02-05 23:59:59      0
2025-02-06 00:00:00    227
Freq: S, Length: 31449601, dtype: int32

• DataFrame重采样

# DataFrame重采样
d = {
    "price":[10,11,2,44,33,44,55,66],
    "score":[40,30,20,50,60,70,80,10],
    "week":pd.date_range("2024-2-8",periods=8,freq="W")
}
df = pd.DataFrame(d)
df

# 对week列进行按月汇总求和
df.resample("M",on="week").sum()

df.resample("M",on="week").apply(np.sum)

# 对week列进行按月汇总：price求平均值，score求和
df.resample("M",on="week").agg({"price":np.mean,"score":np.sum})

三、时区

# 导包
import numpy as np
import pandas as pd

index = pd.date_range("2024-2-8 00:00",periods=3,freq="D")
ts = pd.Series(np.random.randn(len(index)),index=index)
ts
# 执行结果
2024-02-08    1.866786
2024-02-09   -0.475396
2024-02-10   -0.723146
Freq: D, dtype: float64

# tz：timezone 时区
import pytz

# 常用的时区
pytz.common_timezones

# 时区表明
ts = ts.tz_localize(tz="UTC")
ts
# 执行结果
2024-02-08 00:00:00+00:00    1.866786
2024-02-09 00:00:00+00:00   -0.475396
2024-02-10 00:00:00+00:00   -0.723146
Freq: D, dtype: float64

# 时区转换
ts.tz_convert(tz="Asia/Shanghai")
# 执行结果
2024-02-08 08:00:00+08:00    1.866786
2024-02-09 08:00:00+08:00   -0.475396
2024-02-10 08:00:00+08:00   -0.723146
Freq: D, dtype: float64