根据dtype选择列

The select_dtypes() method implements subsetting of columns based on their dtype.

First, let’s create a DataFrame with a slew of different dtypes:

  1. In [435]: df = pd.DataFrame({'string': list('abc'),
  2.    .....:                    'int64': list(range(1, 4)),
  3.    .....:                    'uint8': np.arange(3, 6).astype('u1'),
  4.    .....:                    'float64': np.arange(4.0, 7.0),
  5.    .....:                    'bool1': [True, False, True],
  6.    .....:                    'bool2': [False, True, False],
  7.    .....:                    'dates': pd.date_range('now', periods=3).values,
  8.    .....:                    'category': pd.Series(list("ABC")).astype('category')})
  9.    .....: 
  11. In [436]: df['tdeltas'] = df.dates.diff()
  13. In [437]: df['uint64'] = np.arange(3, 6).astype('u8')
  15. In [438]: df['other_dates'] = pd.date_range('20130101', periods=3).values
  17. In [439]: df['tz_aware_dates'] = pd.date_range('20130101', periods=3, tz='US/Eastern')
  19. In [440]: df
  20. Out[440]: 
  21.   string  int64  uint8  float64  bool1  bool2                      dates category tdeltas  uint64 other_dates            tz_aware_dates
  22. 0      a      1      3      4.0   True  False 2018-08-05 11:57:39.507525        A     NaT       3  2013-01-01 2013-01-01 00:00:00-05:00
  23. 1      b      2      4      5.0  False   True 2018-08-06 11:57:39.507525        B  1 days       4  2013-01-02 2013-01-02 00:00:00-05:00
  24. 2      c      3      5      6.0   True  False 2018-08-07 11:57:39.507525        C  1 days       5  2013-01-03 2013-01-03 00:00:00-05:00

And the dtypes:

  1. In [441]: df.dtypes
  2. Out[441]: 
  3. string                                object
  4. int64                                  int64
  5. uint8                                  uint8
  6. float64                              float64
  7. bool1                                   bool
  8. bool2                                   bool
  9. dates                         datetime64[ns]
  10. category                            category
  11. tdeltas                      timedelta64[ns]
  12. uint64                                uint64
  13. other_dates                   datetime64[ns]
  14. tz_aware_dates    datetime64[ns, US/Eastern]
  15. dtype: object

select_dtypes() has two parameters include and exclude that allow you to say “give me the columns with these dtypes” (include) and/or “give the columns without these dtypes” (exclude).

For example, to select bool columns:

  1. In [442]: df.select_dtypes(include=[bool])
  2. Out[442]: 
  3.    bool1  bool2
  4. 0   True  False
  5. 1  False   True
  6. 2   True  False

You can also pass the name of a dtype in the NumPy dtype hierarchy:

  1. In [443]: df.select_dtypes(include=['bool'])
  2. Out[443]: 
  3.    bool1  bool2
  4. 0   True  False
  5. 1  False   True
  6. 2   True  False

select_dtypes() also works with generic dtypes as well.

For example, to select all numeric and boolean columns while excluding unsigned integers:

  1. In [444]: df.select_dtypes(include=['number', 'bool'], exclude=['unsignedinteger'])
  2. Out[444]: 
  3.    int64  float64  bool1  bool2 tdeltas
  4. 0      1      4.0   True  False     NaT
  5. 1      2      5.0  False   True  1 days
  6. 2      3      6.0   True  False  1 days

To select string columns you must use the object dtype:

  1. In [445]: df.select_dtypes(include=['object'])
  2. Out[445]: 
  3.   string
  4. 0      a
  5. 1      b
  6. 2      c

To see all the child dtypes of a generic dtype like numpy.number you can define a function that returns a tree of child dtypes:

  1. In [446]: def subdtypes(dtype):
  2.    .....:     subs = dtype.__subclasses__()
  3.    .....:     if not subs:
  4.    .....:         return dtype
  5.    .....:     return [dtype, [subdtypes(dt) for dt in subs]]
  6.    .....:

All NumPy dtypes are subclasses of numpy.generic:

  1. In [447]: subdtypes(np.generic)
  2. Out[447]: 
  3. [numpy.generic,
  4.  [[numpy.number,
  5.    [[numpy.integer,
  6.      [[numpy.signedinteger,
  7.        [numpy.int8,
  8.         numpy.int16,
  9.         numpy.int32,
  10.         numpy.int64,
  11.         numpy.int64,
  12.         numpy.timedelta64]],
  13.       [numpy.unsignedinteger,
  14.        [numpy.uint8,
  15.         numpy.uint16,
  16.         numpy.uint32,
  17.         numpy.uint64,
  18.         numpy.uint64]]]],
  19.     [numpy.inexact,
  20.      [[numpy.floating,
  21.        [numpy.float16, numpy.float32, numpy.float64, numpy.float128]],
  22.       [numpy.complexfloating,
  23.        [numpy.complex64, numpy.complex128, numpy.complex256]]]]]],
  24.   [numpy.flexible,
  25.    [[numpy.character, [numpy.bytes_, numpy.str_]],
  26.     [numpy.void, [numpy.record]]]],
  27.   numpy.bool_,
  28.   numpy.datetime64,
  29.   numpy.object_]]

Note: Pandas also defines the types category, and datetime64[ns, tz], which are not integrated into the normal NumPy hierarchy and won’t show up with the above function.