利用Python进行数据分析(7)-时区处理

image

时区处理

很多时间用户选择世界协调时间或者UTC，它是格林治时间的后继者，目前的国家标准。时区通常表示为UTC的偏置。

Python语言中，时区信息通常是来自于第三库pytz。pandas中封装了pytz的功能

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

# 获取时区名称
import pytz
pytz.common_timezones[-5:]

['US/Eastern', 'US/Hawaii', 'US/Mountain', 'US/Pacific', 'UTC']

# 获取pytz对象，使用pytz-timezone
tz = pytz.timezone('America/New_York')
tz

<DstTzInfo 'America/New_York' LMT-1 day, 19:04:00 STD>

时区集合生成

rng = pd.date_range('5/10/2020 11:30'
                    ,periods=6
                   ,freq='D')
ts = pd.Series(np.random.randn(len(rng))
              ,index=rng)
ts

2020-05-10 11:30:00    1.072220
2020-05-11 11:30:00    2.088327
2020-05-12 11:30:00   -0.795575
2020-05-13 11:30:00    1.230427
2020-05-14 11:30:00   -0.012184
2020-05-15 11:30:00   -0.786641
Freq: D, dtype: float64

ts.index

DatetimeIndex(['2020-05-10 11:30:00', '2020-05-11 11:30:00',
               '2020-05-12 11:30:00', '2020-05-13 11:30:00',
               '2020-05-14 11:30:00', '2020-05-15 11:30:00'],
              dtype='datetime64[ns]', freq='D')

print(ts.index.tz)  # tz属性为None

None

# 时区集合生成
pd.date_range('5/10/2020',periods=10,freq='D',tz='UTC')

DatetimeIndex(['2020-05-10 00:00:00+00:00', '2020-05-11 00:00:00+00:00',
               '2020-05-12 00:00:00+00:00', '2020-05-13 00:00:00+00:00',
               '2020-05-14 00:00:00+00:00', '2020-05-15 00:00:00+00:00',
               '2020-05-16 00:00:00+00:00', '2020-05-17 00:00:00+00:00',
               '2020-05-18 00:00:00+00:00', '2020-05-19 00:00:00+00:00'],
              dtype='datetime64[ns, UTC]', freq='D')

简单时区转换到本地化：tz_localize

ts

2020-05-10 11:30:00    1.072220
2020-05-11 11:30:00    2.088327
2020-05-12 11:30:00   -0.795575
2020-05-13 11:30:00    1.230427
2020-05-14 11:30:00   -0.012184
2020-05-15 11:30:00   -0.786641
Freq: D, dtype: float64

ts_utc=ts.tz_localize('UTC')
ts_utc

2020-05-10 11:30:00+00:00    1.072220
2020-05-11 11:30:00+00:00    2.088327
2020-05-12 11:30:00+00:00   -0.795575
2020-05-13 11:30:00+00:00    1.230427
2020-05-14 11:30:00+00:00   -0.012184
2020-05-15 11:30:00+00:00   -0.786641
Freq: D, dtype: float64

ts_utc.index

DatetimeIndex(['2020-05-10 11:30:00+00:00', '2020-05-11 11:30:00+00:00',
               '2020-05-12 11:30:00+00:00', '2020-05-13 11:30:00+00:00',
               '2020-05-14 11:30:00+00:00', '2020-05-15 11:30:00+00:00'],
              dtype='datetime64[ns, UTC]', freq='D')

转换到其他时区：tz_convert

ts_utc.tz_convert("America/New_York")   # 转到纽约时区

2020-05-10 07:30:00-04:00    1.072220
2020-05-11 07:30:00-04:00    2.088327
2020-05-12 07:30:00-04:00   -0.795575
2020-05-13 07:30:00-04:00    1.230427
2020-05-14 07:30:00-04:00   -0.012184
2020-05-15 07:30:00-04:00   -0.786641
Freq: D, dtype: float64

ts_utc.tz_convert("Asia/Shanghai")     # 转到上海时区

2020-05-10 19:30:00+08:00    1.072220
2020-05-11 19:30:00+08:00    2.088327
2020-05-12 19:30:00+08:00   -0.795575
2020-05-13 19:30:00+08:00    1.230427
2020-05-14 19:30:00+08:00   -0.012184
2020-05-15 19:30:00+08:00   -0.786641
Freq: D, dtype: float64

实例化方法

tz_localzie、tz_convert是DatetimeIndex的实例化方法

ts.index.tz_localize('Asia/Shanghai')

DatetimeIndex(['2020-05-10 11:30:00+08:00', '2020-05-11 11:30:00+08:00',
               '2020-05-12 11:30:00+08:00', '2020-05-13 11:30:00+08:00',
               '2020-05-14 11:30:00+08:00', '2020-05-15 11:30:00+08:00'],
              dtype='datetime64[ns, Asia/Shanghai]', freq='D')

时区感知时间戳对象的操作

单独的Timestamp对象也可以从简单时间戳本地为时区感知时间戳

Timestamp对象的转化

stamp = pd.Timestamp('2020-05-10 23:49')
stamp

Timestamp('2020-05-10 23:49:00')

stamp_utc = stamp.tz_localize('utc')  # 本地化
stamp_utc

Timestamp('2020-05-10 23:49:00+0000', tz='UTC')

stamp_utc.tz_convert("Asia/Shanghai")  # 时区转化

Timestamp('2020-05-11 07:49:00+0800', tz='Asia/Shanghai')

创建的时候直接传递时区

stamp_shanghai = pd.Timestamp("2020-05-10 23:58"
                     ,tz="Asia/Shanghai")   # 直接传递时区
stamp_shanghai

Timestamp('2020-05-10 23:58:00+0800', tz='Asia/Shanghai')

时间戳数值不变性

时区感知的Timestamp对象内部存储的一个UNix到现在的时间戳数值，保持不变

stamp_shanghai.value

1589126280000000000

# 结果同上
stamp_shanghai.tz_convert("America/New_York").value

1589126280000000000

dateOffset

from pandas.tseries.offsets import Hour

data = pd.Timestamp("2020-05-10 01:30"   # 创建一个Timestamp对象
                    ,tz="Asia/Shanghai")
data

Timestamp('2020-05-10 01:30:00+0800', tz='Asia/Shanghai')

data + Hour(2)  # 加上2个小时
# data +2 * Hour()

Timestamp('2020-05-10 03:30:00+0800', tz='Asia/Shanghai')

不同时区的操作

如果两个不同时区的时间序列需要联合，结果将是UTC时间的。时间戳按照UTC格式存储

rng = pd.date_range("2020-05-10 23:43"
                   ,periods=10
                   ,freq="B")
rng

DatetimeIndex(['2020-05-11 23:43:00', '2020-05-12 23:43:00',
               '2020-05-13 23:43:00', '2020-05-14 23:43:00',
               '2020-05-15 23:43:00', '2020-05-18 23:43:00',
               '2020-05-19 23:43:00', '2020-05-20 23:43:00',
               '2020-05-21 23:43:00', '2020-05-22 23:43:00'],
              dtype='datetime64[ns]', freq='B')

ts = pd.Series(np.random.randn(len(rng))   # 随机取值
              ,index=rng)  # 行index
ts

2020-05-11 23:43:00    0.258933
2020-05-12 23:43:00    0.416673
2020-05-13 23:43:00    0.089695
2020-05-14 23:43:00   -0.347376
2020-05-15 23:43:00   -0.304925
2020-05-18 23:43:00   -0.891367
2020-05-19 23:43:00   -0.960866
2020-05-20 23:43:00   -0.420829
2020-05-21 23:43:00    0.591673
2020-05-22 23:43:00    1.431417
Freq: B, dtype: float64

ts1 = ts[:7].tz_localize('Asia/Shanghai')
ts2 = ts1[2:].tz_convert('Europe/Moscow')

ts1

2020-05-11 23:43:00+08:00    0.258933
2020-05-12 23:43:00+08:00    0.416673
2020-05-13 23:43:00+08:00    0.089695
2020-05-14 23:43:00+08:00   -0.347376
2020-05-15 23:43:00+08:00   -0.304925
2020-05-18 23:43:00+08:00   -0.891367
2020-05-19 23:43:00+08:00   -0.960866
Freq: B, dtype: float64

ts2

2020-05-13 18:43:00+03:00    0.089695
2020-05-14 18:43:00+03:00   -0.347376
2020-05-15 18:43:00+03:00   -0.304925
2020-05-18 18:43:00+03:00   -0.891367
2020-05-19 18:43:00+03:00   -0.960866
Freq: B, dtype: float64

result = ts1 + ts2
result

2020-05-11 15:43:00+00:00         NaN
2020-05-12 15:43:00+00:00         NaN
2020-05-13 15:43:00+00:00    0.179391
2020-05-14 15:43:00+00:00   -0.694751
2020-05-15 15:43:00+00:00   -0.609850
2020-05-18 15:43:00+00:00   -1.782735
2020-05-19 15:43:00+00:00   -1.921732
Freq: B, dtype: float64

result.index

DatetimeIndex(['2020-05-11 15:43:00+00:00', '2020-05-12 15:43:00+00:00',
               '2020-05-13 15:43:00+00:00', '2020-05-14 15:43:00+00:00',
               '2020-05-15 15:43:00+00:00', '2020-05-18 15:43:00+00:00',
               '2020-05-19 15:43:00+00:00'],
              dtype='datetime64[ns, UTC]', freq='B')