pandas 处理日期时间变量
创建Timestamp类和Period类
Timestamp对象
import pandas as pd
from datetime import datetime
print(datetime(2019,10,21))
print(pd.Timestamp(datetime(2019,10,21)))
print(pd.Timestamp(datetime(2019,10,21,1,2,3)))
print(pd.Timestamp('2012-09-09 1:3:3'))
print(pd.Timestamp(2012,4,4))
转换为Timestamp对象
df1 = pd.read_csv('Beijing_2008.csv',header=2)
print(pd.Timestamp(df1['Date (LST)'][0]))
print(df1['Date (LST)'].apply(pd.Timestamp))
利用to_datetime进行批量转换
‘’’
pd.to_datetime(
arg,errors,dayfirst =‘False’,yeatfirst = ‘False’,box,format
)
‘’’
print(pd.to_datetime(datetime(2019,10,21,1,2,3)))
print(pd.to_datetime('2019-10-21 1:2:3'))
print(pd.to_datetime(['2015/2/2','2010.3.3']))
print(pd.to_datetime(df1['Date (LST)'],format = '%Y-%m-%d %H:%M'))
多列数据合成Timestamp类
print(pd.to_datetime(df1[['Year','Month','Day','Hour']]))
DatetimeIndex类
创建DatetimeIndex对象
创建索引时自动转换
bj08_datetimeindex = df1.set_index(pd.to_datetime(df1['Date (LST)']))
print(bj08_datetimeindex)
使用date_range创建DatetimeIndex(主要和reindex配合使用以快速完成对时间序列中缺失值的填充)
print(pd.date_range('2019/1/1',periods=12,freq = '2M'))
print(bj08_datetimeindex['2008-11'])
print(bj08_datetimeindex['2008-11-1':'2008-11-6 9:00:00'])
对时间做分组汇总
序列的分组汇总
直接取出索引的相应层级
print(bj08_datetimeindex.index.hour)
print(bj08_datetimeindex.index.quarter)
print(bj08_datetimeindex.index.week)
直接利用groupby方法进行汇总
print(bj08_datetimeindex.groupby(bj08_datetimeindex.index.month).Value.max())
使用更强大的resample函数
print(bj08_datetimeindex.resample('3D').mean())
序列的缺失值处理
‘’’
时间序列要求记录的时间点连贯无缺失 ，因此需要：
首先建立针对整个时间范围的完整序列框架
随后针对无数据的时间点进行缺失值处理、
df.reindex()是完成该任务的强有力工具
‘’’
bjpm09 = pd.read_csv('Beijing_2009.csv',encoding='utf-8',header = 0)
bjpm09idx = bjpm09.set_index(pd.to_datetime(bjpm09['Date (LST)']))
bjpm09idx = bjpm09idx[bjpm09idx['Value'] > 0]
print(bjpm09idx)
idx = pd.date_range(start = '2009-02-01 00:00',
end = '2009-12-31 23:00',freq = 'H')
print(idx)
bj09new = bjpm09idx[~bjpm09idx.index.duplicated()].reindex(idx,method = 'bfill')
print(bj09new)
序列数值平移
print(bj09new.shift(3))