1. Python和Pandas日期工具的区别

  1. #  引入datetime模块，创建date、time和datetime对象
  2.  In[2]: import datetime
  4.         date = datetime.date(year=2013, month=6, day=7)
  5.         time = datetime.time(hour=12, minute=30, second=19, microsecond=463198)
  6.         dt = datetime.datetime(year=2013, month=6, day=7, 
  7.                                hour=12, minute=30, second=19, microsecond=463198)
  9.         print("date is ", date)
  10.         print("time is", time)
  11.         print("datetime is", dt)
  13. date is  2013-06-07
  14. time is 12:30:19.463198
  15. datetime is 2013-06-07 12:30:19.463198
  1. #  创建并打印一个timedelta对象
  2.  In[3]: td = datetime.timedelta(weeks=2, days=5, hours=10, minutes=20, 
  3.                                 seconds=6.73, milliseconds=99, microseconds=8)
  4.         print(td)
  6. 19 days, 10:20:06.829008
  1. #  将date和datetime，与timedelta做加减
  2.  In[4]: print('new date is', date + td)
  3.         print('new datetime is', dt + td)
  5. new date is 2013-06-26
  6. new datetime is 2013-06-26 22:50:26.292206
  1. #  time和timedelta不能做加法
  2.  In[5]: time + td
  3. ---------------------------------------------------------------------------
  4. TypeError                                 Traceback (most recent call last)
  5. <ipython-input-5-bd4e11db43bd> in <module>()
  6. ----> 1 time + td
  8. TypeError: unsupported operand type(s) for +: 'datetime.time' and 'datetime.timedelta'
  1. #  再来看一下pandas的Timestamp对象。Timestamp构造器比较灵活，可以处理多种输入
  2.  In[6]: pd.Timestamp(year=2012, month=12, day=21, hour=5, minute=10, second=8, microsecond=99)
  3. Out[6]: Timestamp('2012-12-21 05:10:08.000099')
  5.  In[7]: pd.Timestamp('2016/1/10')
  6. Out[7]: Timestamp('2016-01-10 00:00:00')
  8.  In[8]: pd.Timestamp('2014-5/10')
  9. Out[8]: Timestamp('2014-05-10 00:00:00')
  11.  In[9]: pd.Timestamp('Jan 3, 2019 20:45.56')
  12. Out[9]: Timestamp('2019-01-03 20:45:33')
  14.  In[10]: pd.Timestamp('2016-01-05T05:34:43.123456789')
  15. Out[10]: Timestamp('2016-01-05 05:34:43.123456789')
  1. #  也可以传入一个整数或浮点数，表示距离1970年1月1日的时间
  2.  In[11]: pd.Timestamp(500)
  3. Out[11]: Timestamp('1970-01-01 00:00:00.000000500')
  5.  In[12]: pd.Timestamp(5000, unit='D')
  6. Out[12]: Timestamp('1983-09-10 00:00:00')
  1. #  pandas的to_datetime函数与Timestamp类似，但有些参数不同
  2.  In[13]: pd.to_datetime('2015-5-13')
  3. Out[13]: Timestamp('2015-05-13 00:00:00')
  5.  In[14]: pd.to_datetime('2015-13-5', dayfirst=True)
  6. Out[14]: Timestamp('2015-05-13 00:00:00')
  8.  In[15]: pd.Timestamp('Saturday September 30th, 2017')
  9. Out[15]: Timestamp('2017-09-30 00:00:00')
  11.  In[16]: pd.to_datetime('Start Date: Sep 30, 2017 Start Time: 1:30 pm', format='Start Date: %b %d, %Y Start Time: %I:%M %p')
  12. Out[16]: Timestamp('2017-09-30 13:30:00')
  14.  In[17]: pd.to_datetime(100, unit='D', origin='2013-1-1')
  15. Out[17]: Timestamp('2013-04-11 00:00:00')
  1. #  to_datetime可以将一个字符串或整数列表或Series转换为时间戳
  2.  In[18]: s = pd.Series([10, 100, 1000, 10000])
  3.          pd.to_datetime(s, unit='D')
  4. Out[18]: 0   1970-01-11
  5.          1   1970-04-11
  6.          2   1972-09-27
  7.          3   1997-05-19
  8.          dtype: datetime64[ns]
  10.  In[19]: s = pd.Series(['12-5-2015', '14-1-2013', '20/12/2017', '40/23/2017'])
  11.          pd.to_datetime(s, dayfirst=True, errors='coerce')
  12. Out[19]: 0   2015-05-12
  13.          1   2013-01-14
  14.          2   2017-12-20
  15.          3          NaT
  16.          dtype: datetime64[ns]
  18.  In[20]: pd.to_datetime(['Aug 3 1999 3:45:56', '10/31/2017'])
  19. Out[20]: DatetimeIndex(['1999-08-03 03:45:56', '2017-10-31 00:00:00'], dtype='datetime64[ns]', freq=None)
  1. #  Pandas的Timedelta和to_timedelta也可以用来表示一定的时间量。
  2. #  to_timedelta函数可以产生一个Timedelta对象。
  3. #  与to_datetime类似，to_timedelta也可以转换列表或Series变成Timedelta对象。
  4.  In[21]: pd.Timedelta('12 days 5 hours 3 minutes 123456789 nanoseconds')
  5. Out[21]: Timedelta('12 days 05:03:00.123456')
  7.  In[22]: pd.Timedelta(days=5, minutes=7.34)
  8. Out[22]: Timedelta('5 days 00:07:20.400000')
  10.  In[23]: pd.Timedelta(100, unit='W')
  11. Out[23]: Timedelta('700 days 00:00:00')
  13.  In[24]: pd.to_timedelta('5 dayz', errors='ignore')
  14. Out[24]: '5 dayz'
  16.  In[25]: pd.to_timedelta('67:15:45.454')
  17. Out[25]: Timedelta('2 days 19:15:45.454000')
  19.  In[26]: s = pd.Series([10, 100])
  20.          pd.to_timedelta(s, unit='s')
  21. Out[26]: 0   00:00:10
  22.          1   00:01:40
  23.          dtype: timedelta64[ns]
  25.  In[27]: time_strings = ['2 days 24 minutes 89.67 seconds', '00:45:23.6']
  26.          pd.to_timedelta(time_strings)
  27. Out[27]: TimedeltaIndex(['2 days 00:25:29.670000', '0 days 00:45:23.600000'], dtype='timedelta64[ns]', freq=None)
  1. #  Timedeltas对象可以和Timestamps互相加减，甚至可以相除返回一个浮点数
  2.  In[28]: pd.Timedelta('12 days 5 hours 3 minutes') * 2
  3. Out[28]: Timedelta('24 days 10:06:00')
  5.  In[29]: pd.Timestamp('1/1/2017') + pd.Timedelta('12 days 5 hours 3 minutes') * 2
  6. Out[29]: Timestamp('2017-01-25 10:06:00')
  8.  In[30]: td1 = pd.to_timedelta([10, 100], unit='s')
  9.          td2 = pd.to_timedelta(['3 hours', '4 hours'])
  10.          td1 + td2
  11. Out[30]: TimedeltaIndex(['03:00:10', '04:01:40'], dtype='timedelta64[ns]', freq=None)
  13.  In[31]: pd.Timedelta('12 days') / pd.Timedelta('3 days')
  14. Out[31]: 4.0
  1. #  Timestamps 和 Timedeltas有许多可用的属性和方法，下面列举了一些：
  2.  In[32]: ts = pd.Timestamp('2016-10-1 4:23:23.9')
  3.  In[33]: ts.ceil('h')
  4. Out[33]: Timestamp('2016-10-01 05:00:00')
  6.  In[34]: ts.year, ts.month, ts.day, ts.hour, ts.minute, ts.second
  7. Out[34]: (2016, 10, 1, 4, 23, 23)
  9.  In[35]: ts.dayofweek, ts.dayofyear, ts.daysinmonth
  10. Out[35]: (5, 275, 31)
  12.  In[36]: ts.to_pydatetime()
  13. Out[36]: datetime.datetime(2016, 10, 1, 4, 23, 23, 900000)
  15.  In[37]: td = pd.Timedelta(125.8723, unit='h')
  16.          td
  17. Out[37]: Timedelta('5 days 05:52:20.280000')
  19.  In[38]: td.round('min')
  20. Out[38]: Timedelta('5 days 05:52:00')
  22.  In[39]: td.components
  23. Out[39]: Components(days=5, hours=5, minutes=52, seconds=20, milliseconds=280, microseconds=0, nanoseconds=0)
  25.  In[40]: td.total_seconds()
  26. Out[40]: 453140.28

更多

  1. #  对比一下，在使用和没使用格式指令的条件下，将字符串转换为Timestamps对象的速度
  2.  In[41]: date_string_list = ['Sep 30 1984'] * 10000
  4.  In[42]: %timeit pd.to_datetime(date_string_list, format='%b %d %Y')
  5. 37.8 ms ± 556 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
  7.  In[43]: %timeit pd.to_datetime(date_string_list)
  8. 1.33 s ± 57.9 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)