.dt访问器

Series has an accessor to succinctly return datetime like properties for the values of the Series, if it is a datetime/period like Series. This will return a Series, indexed like the existing Series.

  1. # datetime
  2. In [280]: s = pd.Series(pd.date_range('20130101 09:10:12', periods=4))
  4. In [281]: s
  5. Out[281]: 
  6. 0   2013-01-01 09:10:12
  7. 1   2013-01-02 09:10:12
  8. 2   2013-01-03 09:10:12
  9. 3   2013-01-04 09:10:12
  10. dtype: datetime64[ns]
  12. In [282]: s.dt.hour
  13. Out[282]: 
  14. 0    9
  15. 1    9
  16. 2    9
  17. 3    9
  18. dtype: int64
  20. In [283]: s.dt.second
  21. Out[283]: 
  22. 0    12
  23. 1    12
  24. 2    12
  25. 3    12
  26. dtype: int64
  28. In [284]: s.dt.day
  29. Out[284]: 
  30. 0    1
  31. 1    2
  32. 2    3
  33. 3    4
  34. dtype: int64

This enables nice expressions like this:

  1. In [285]: s[s.dt.day==2]
  2. Out[285]: 
  3. 1   2013-01-02 09:10:12
  4. dtype: datetime64[ns]

You can easily produces tz aware transformations:

  1. In [286]: stz = s.dt.tz_localize('US/Eastern')
  3. In [287]: stz
  4. Out[287]: 
  5. 0   2013-01-01 09:10:12-05:00
  6. 1   2013-01-02 09:10:12-05:00
  7. 2   2013-01-03 09:10:12-05:00
  8. 3   2013-01-04 09:10:12-05:00
  9. dtype: datetime64[ns, US/Eastern]
  11. In [288]: stz.dt.tz
  12. Out[288]: <DstTzInfo 'US/Eastern' LMT-1 day, 19:04:00 STD>

You can also chain these types of operations:

  1. In [289]: s.dt.tz_localize('UTC').dt.tz_convert('US/Eastern')
  2. Out[289]: 
  3. 0   2013-01-01 04:10:12-05:00
  4. 1   2013-01-02 04:10:12-05:00
  5. 2   2013-01-03 04:10:12-05:00
  6. 3   2013-01-04 04:10:12-05:00
  7. dtype: datetime64[ns, US/Eastern]

You can also format datetime values as strings with Series.dt.strftime() which supports the same format as the standard strftime().

  1. # DatetimeIndex
  2. In [290]: s = pd.Series(pd.date_range('20130101', periods=4))
  4. In [291]: s
  5. Out[291]: 
  6. 0   2013-01-01
  7. 1   2013-01-02
  8. 2   2013-01-03
  9. 3   2013-01-04
  10. dtype: datetime64[ns]
  12. In [292]: s.dt.strftime('%Y/%m/%d')
  13. Out[292]: 
  14. 0    2013/01/01
  15. 1    2013/01/02
  16. 2    2013/01/03
  17. 3    2013/01/04
  18. dtype: object
  1. # PeriodIndex
  2. In [293]: s = pd.Series(pd.period_range('20130101', periods=4))
  4. In [294]: s
  5. Out[294]: 
  6. 0   2013-01-01
  7. 1   2013-01-02
  8. 2   2013-01-03
  9. 3   2013-01-04
  10. dtype: object
  12. In [295]: s.dt.strftime('%Y/%m/%d')
  13. Out[295]: 
  14. 0    2013/01/01
  15. 1    2013/01/02
  16. 2    2013/01/03
  17. 3    2013/01/04
  18. dtype: object

The .dt accessor works for period and timedelta dtypes.

  1. # period
  2. In [296]: s = pd.Series(pd.period_range('20130101', periods=4, freq='D'))
  4. In [297]: s
  5. Out[297]: 
  6. 0   2013-01-01
  7. 1   2013-01-02
  8. 2   2013-01-03
  9. 3   2013-01-04
  10. dtype: object
  12. In [298]: s.dt.year
  13. Out[298]: 
  14. 0    2013
  15. 1    2013
  16. 2    2013
  17. 3    2013
  18. dtype: int64
  20. In [299]: s.dt.day
  21. Out[299]: 
  22. 0    1
  23. 1    2
  24. 2    3
  25. 3    4
  26. dtype: int64
  1. # timedelta
  2. In [300]: s = pd.Series(pd.timedelta_range('1 day 00:00:05', periods=4, freq='s'))
  4. In [301]: s
  5. Out[301]: 
  6. 0   1 days 00:00:05
  7. 1   1 days 00:00:06
  8. 2   1 days 00:00:07
  9. 3   1 days 00:00:08
  10. dtype: timedelta64[ns]
  12. In [302]: s.dt.days
  13. Out[302]: 
  14. 0    1
  15. 1    1
  16. 2    1
  17. 3    1
  18. dtype: int64
  20. In [303]: s.dt.seconds
  21. Out[303]: 
  22. 0    5
  23. 1    6
  24. 2    7
  25. 3    8
  26. dtype: int64
  28. In [304]: s.dt.components
  29. Out[304]: 
  30.    days  hours  minutes  seconds  milliseconds  microseconds  nanoseconds
  31. 0     1      0        0        5             0             0            0
  32. 1     1      0        0        6             0             0            0
  33. 2     1      0        0        7             0             0            0
  34. 3     1      0        0        8             0             0            0

Note: Series.dt will raise a TypeError if you access with a non-datetime-like values.