重塑和透视表#

pandas 提供了用于操作 SeriesDataFrame 的方法,以改变数据的表示形式,用于进一步的数据处理或数据汇总。

pivot()pivot_table()#

../_images/reshaping_pivot.png

pivot()#

数据通常以所谓的“堆叠”(stacked)或“记录”(record)格式存储。在“记录”或“宽”(wide)格式中,通常每个主体对应一行。在“堆叠”或“长”(long)格式中,在适用情况下,每个主体对应多行。

In [1]: data = {
   ...:    "value": range(12),
   ...:    "variable": ["A"] * 3 + ["B"] * 3 + ["C"] * 3 + ["D"] * 3,
   ...:    "date": pd.to_datetime(["2020-01-03", "2020-01-04", "2020-01-05"] * 4)
   ...: }
   ...: 

In [2]: df = pd.DataFrame(data)

为了对每个唯一变量执行时间序列操作,更好的表示形式是 columns 为唯一变量,且日期的 index 标识单个观测值。要将数据重塑为这种形式,我们使用 DataFrame.pivot() 方法(也作为顶层函数 pivot() 实现)。

In [3]: pivoted = df.pivot(index="date", columns="variable", values="value")

In [4]: pivoted
Out[4]: 
variable    A  B  C   D
date                   
2020-01-03  0  3  6   9
2020-01-04  1  4  7  10
2020-01-05  2  5  8  11

如果省略 values 参数,并且输入的 DataFrame 有多个值列未用作 pivot() 的列或索引输入,则结果“透视”后的 DataFrame 将具有 层级列,其最顶层指示相应的值列。

In [5]: df["value2"] = df["value"] * 2

In [6]: pivoted = df.pivot(index="date", columns="variable")

In [7]: pivoted
Out[7]: 
           value           value2            
variable       A  B  C   D      A   B   C   D
date                                         
2020-01-03     0  3  6   9      0   6  12  18
2020-01-04     1  4  7  10      2   8  14  20
2020-01-05     2  5  8  11      4  10  16  22

然后,您可以从透视后的 DataFrame 中选择子集。

In [8]: pivoted["value2"]
Out[8]: 
variable    A   B   C   D
date                     
2020-01-03  0   6  12  18
2020-01-04  2   8  14  20
2020-01-05  4  10  16  22

请注意,在数据类型同质的情况下,这将返回底层数据的视图。

注意

pivot() 只能处理由 indexcolumns 指定的唯一行。如果您的数据包含重复项,请使用 pivot_table()

pivot_table()#

尽管 pivot() 提供了各种数据类型的通用透视功能,pandas 还提供了 pivot_table()pivot_table(),用于对数值数据进行聚合透视。

函数 pivot_table() 可用于创建电子表格风格的透视表。请参阅秘籍了解一些高级策略。

In [9]: import datetime

In [10]: df = pd.DataFrame(
   ....:     {
   ....:         "A": ["one", "one", "two", "three"] * 6,
   ....:         "B": ["A", "B", "C"] * 8,
   ....:         "C": ["foo", "foo", "foo", "bar", "bar", "bar"] * 4,
   ....:         "D": np.random.randn(24),
   ....:         "E": np.random.randn(24),
   ....:         "F": [datetime.datetime(2013, i, 1) for i in range(1, 13)]
   ....:         + [datetime.datetime(2013, i, 15) for i in range(1, 13)],
   ....:     }
   ....: )
   ....: 

In [11]: df
Out[11]: 
        A  B    C         D         E          F
0     one  A  foo  0.469112  0.404705 2013-01-01
1     one  B  foo -0.282863  0.577046 2013-02-01
2     two  C  foo -1.509059 -1.715002 2013-03-01
3   three  A  bar -1.135632 -1.039268 2013-04-01
4     one  B  bar  1.212112 -0.370647 2013-05-01
..    ... ..  ...       ...       ...        ...
19  three  B  foo -1.087401 -0.472035 2013-08-15
20    one  C  foo -0.673690 -0.013960 2013-09-15
21    one  A  bar  0.113648 -0.362543 2013-10-15
22    two  B  bar -1.478427 -0.006154 2013-11-15
23  three  C  bar  0.524988 -0.923061 2013-12-15

[24 rows x 6 columns]

In [12]: pd.pivot_table(df, values="D", index=["A", "B"], columns=["C"])
Out[12]: 
C             bar       foo
A     B                    
one   A -0.995460  0.595334
      B  0.393570 -0.494817
      C  0.196903 -0.767769
three A -0.431886       NaN
      B       NaN -1.065818
      C  0.798396       NaN
two   A       NaN  0.197720
      B -0.986678       NaN
      C       NaN -1.274317

In [13]: pd.pivot_table(
   ....:     df, values=["D", "E"],
   ....:     index=["B"],
   ....:     columns=["A", "C"],
   ....:     aggfunc="sum",
   ....: )
   ....: 
Out[13]: 
          D                      ...         E                   
A       one               three  ...     three      two          
C       bar       foo       bar  ...       foo      bar       foo
B                                ...                             
A -1.990921  1.190667 -0.863772  ...       NaN      NaN -1.067650
B  0.787140 -0.989634       NaN  ...  0.372851  1.63741       NaN
C  0.393806 -1.535539  1.596791  ...       NaN      NaN -3.491906

[3 rows x 12 columns]

In [14]: pd.pivot_table(
   ....:     df, values="E",
   ....:     index=["B", "C"],
   ....:     columns=["A"],
   ....:     aggfunc=["sum", "mean"],
   ....: )
   ....: 
Out[14]: 
            sum                          mean                    
A           one     three       two       one     three       two
B C                                                              
A bar -0.471593 -2.008182       NaN -0.235796 -1.004091       NaN
  foo  0.761726       NaN -1.067650  0.380863       NaN -0.533825
B bar -1.665170       NaN  1.637410 -0.832585       NaN  0.818705
  foo -0.097554  0.372851       NaN -0.048777  0.186425       NaN
C bar -0.744154 -2.392449       NaN -0.372077 -1.196224       NaN
  foo  1.061810       NaN -3.491906  0.530905       NaN -1.745953

结果是一个 DataFrame,其索引或列上可能带有 MultiIndex。如果未提供 values 列名,透视表将在列中包含所有数据,并在额外的层级上显示层级结构。

In [15]: pd.pivot_table(df[["A", "B", "C", "D", "E"]], index=["A", "B"], columns=["C"])
Out[15]: 
                D                   E          
C             bar       foo       bar       foo
A     B                                        
one   A -0.995460  0.595334 -0.235796  0.380863
      B  0.393570 -0.494817 -0.832585 -0.048777
      C  0.196903 -0.767769 -0.372077  0.530905
three A -0.431886       NaN -1.004091       NaN
      B       NaN -1.065818       NaN  0.186425
      C  0.798396       NaN -1.196224       NaN
two   A       NaN  0.197720       NaN -0.533825
      B -0.986678       NaN  0.818705       NaN
      C       NaN -1.274317       NaN -1.745953

此外,您可以将 Grouper 用于 indexcolumns 关键字。有关 Grouper 的详细信息,请参阅使用 Grouper 规范进行分组

In [16]: pd.pivot_table(df, values="D", index=pd.Grouper(freq="ME", key="F"), columns="C")
Out[16]: 
C                bar       foo
F                             
2013-01-31       NaN  0.595334
2013-02-28       NaN -0.494817
2013-03-31       NaN -1.274317
2013-04-30 -0.431886       NaN
2013-05-31  0.393570       NaN
2013-06-30  0.196903       NaN
2013-07-31       NaN  0.197720
2013-08-31       NaN -1.065818
2013-09-30       NaN -0.767769
2013-10-31 -0.995460       NaN
2013-11-30 -0.986678       NaN
2013-12-31  0.798396       NaN

添加边距#

margins=True 传递给 pivot_table() 将添加带有 All 标签的行和列,其中包含跨行和列类别的部分组聚合结果。

In [17]: table = df.pivot_table(
   ....:     index=["A", "B"],
   ....:     columns="C",
   ....:     values=["D", "E"],
   ....:     margins=True,
   ....:     aggfunc="std"
   ....: )
   ....: 

In [18]: table
Out[18]: 
                D                             E                    
C             bar       foo       All       bar       foo       All
A     B                                                            
one   A  1.568517  0.178504  1.293926  0.179247  0.033718  0.371275
      B  1.157593  0.299748  0.860059  0.653280  0.885047  0.779837
      C  0.523425  0.133049  0.638297  1.111310  0.770555  0.938819
three A  0.995247       NaN  0.995247  0.049748       NaN  0.049748
      B       NaN  0.030522  0.030522       NaN  0.931203  0.931203
      C  0.386657       NaN  0.386657  0.386312       NaN  0.386312
two   A       NaN  0.111032  0.111032       NaN  1.146201  1.146201
      B  0.695438       NaN  0.695438  1.166526       NaN  1.166526
      C       NaN  0.331975  0.331975       NaN  0.043771  0.043771
All      1.014073  0.713941  0.871016  0.881376  0.984017  0.923568

此外,您可以调用 DataFrame.stack() 将透视后的 DataFrame 显示为具有多级索引的形式。

In [19]: table.stack(future_stack=True)
Out[19]: 
                  D         E
A   B C                      
one A bar  1.568517  0.179247
      foo  0.178504  0.033718
      All  1.293926  0.371275
    B bar  1.157593  0.653280
      foo  0.299748  0.885047
...             ...       ...
two C foo  0.331975  0.043771
      All  0.331975  0.043771
All   bar  1.014073  0.881376
      foo  0.713941  0.984017
      All  0.871016  0.923568

[30 rows x 2 columns]

stack()unstack()#

../_images/reshaping_stack.png

pivot() 方法密切相关的是 SeriesDataFrame 上可用的 stack()unstack() 方法。这些方法旨在与 MultiIndex 对象一起使用(参见 层级索引 部分)。

  • stack(): 将(可能具有层级的)列标签的某个层级“透视”(pivot)到行轴,返回一个 DataFrame,其索引包含新的最内层行标签。

  • unstack(): (stack() 的逆操作)将(可能具有层级的)行索引的某个层级“透视”(pivot)到列轴,生成一个重塑后的 DataFrame,其列标签包含新的最内层。

../_images/reshaping_unstack.png
In [20]: tuples = [
   ....:    ["bar", "bar", "baz", "baz", "foo", "foo", "qux", "qux"],
   ....:    ["one", "two", "one", "two", "one", "two", "one", "two"],
   ....: ]
   ....: 

In [21]: index = pd.MultiIndex.from_arrays(tuples, names=["first", "second"])

In [22]: df = pd.DataFrame(np.random.randn(8, 2), index=index, columns=["A", "B"])

In [23]: df2 = df[:4]

In [24]: df2
Out[24]: 
                     A         B
first second                    
bar   one     0.895717  0.805244
      two    -1.206412  2.565646
baz   one     1.431256  1.340309
      two    -1.170299 -0.226169

函数 stack()DataFrame 列中的某个层级“压缩”,生成以下两者之一:

如果列包含 MultiIndex,您可以选择要堆叠(stack)哪个层级。堆叠的层级将成为列上新的最低层级 MultiIndex

In [25]: stacked = df2.stack(future_stack=True)

In [26]: stacked
Out[26]: 
first  second   
bar    one     A    0.895717
               B    0.805244
       two     A   -1.206412
               B    2.565646
baz    one     A    1.431256
               B    1.340309
       two     A   -1.170299
               B   -0.226169
dtype: float64

对于一个“堆叠”(stacked)的 DataFrameSeries(其 index 包含 MultiIndex),stack() 的逆操作是 unstack(),它默认取消堆叠(unstack)**最后一个层级**。

In [27]: stacked.unstack()
Out[27]: 
                     A         B
first second                    
bar   one     0.895717  0.805244
      two    -1.206412  2.565646
baz   one     1.431256  1.340309
      two    -1.170299 -0.226169

In [28]: stacked.unstack(1)
Out[28]: 
second        one       two
first                      
bar   A  0.895717 -1.206412
      B  0.805244  2.565646
baz   A  1.431256 -1.170299
      B  1.340309 -0.226169

In [29]: stacked.unstack(0)
Out[29]: 
first          bar       baz
second                      
one    A  0.895717  1.431256
       B  0.805244  1.340309
two    A -1.206412 -1.170299
       B  2.565646 -0.226169
../_images/reshaping_unstack_1.png

如果索引有名称,您可以使用层级名称而不是指定层级编号。

In [30]: stacked.unstack("second")
Out[30]: 
second        one       two
first                      
bar   A  0.895717 -1.206412
      B  0.805244  2.565646
baz   A  1.431256 -1.170299
      B  1.340309 -0.226169
../_images/reshaping_unstack_0.png

请注意,stack()unstack() 方法会隐式地对涉及的索引层级进行排序。因此,先调用 stack() 再调用 unstack(),或者反过来,都将导致原始 DataFrameSeries 的一个**已排序**副本。

In [31]: index = pd.MultiIndex.from_product([[2, 1], ["a", "b"]])

In [32]: df = pd.DataFrame(np.random.randn(4), index=index, columns=["A"])

In [33]: df
Out[33]: 
            A
2 a -1.413681
  b  1.607920
1 a  1.024180
  b  0.569605

In [34]: all(df.unstack().stack(future_stack=True) == df.sort_index())
Out[34]: True

多个层级#

您也可以通过传递一个层级列表来一次性堆叠(stack)或取消堆叠(unstack)多个层级,在这种情况下,最终结果就如同列表中的每个层级都被单独处理一样。

In [35]: columns = pd.MultiIndex.from_tuples(
   ....:     [
   ....:         ("A", "cat", "long"),
   ....:         ("B", "cat", "long"),
   ....:         ("A", "dog", "short"),
   ....:         ("B", "dog", "short"),
   ....:     ],
   ....:     names=["exp", "animal", "hair_length"],
   ....: )
   ....: 

In [36]: df = pd.DataFrame(np.random.randn(4, 4), columns=columns)

In [37]: df
Out[37]: 
exp                 A         B         A         B
animal            cat       cat       dog       dog
hair_length      long      long     short     short
0            0.875906 -2.211372  0.974466 -2.006747
1           -0.410001 -0.078638  0.545952 -1.219217
2           -1.226825  0.769804 -1.281247 -0.727707
3           -0.121306 -0.097883  0.695775  0.341734

In [38]: df.stack(level=["animal", "hair_length"], future_stack=True)
Out[38]: 
exp                          A         B
  animal hair_length                    
0 cat    long         0.875906 -2.211372
  dog    short        0.974466 -2.006747
1 cat    long        -0.410001 -0.078638
  dog    short        0.545952 -1.219217
2 cat    long        -1.226825  0.769804
  dog    short       -1.281247 -0.727707
3 cat    long        -0.121306 -0.097883
  dog    short        0.695775  0.341734

层级列表可以包含层级名称或层级编号,但不能混合使用。

# df.stack(level=['animal', 'hair_length'], future_stack=True)
# from above is equivalent to:
In [39]: df.stack(level=[1, 2], future_stack=True)
Out[39]: 
exp                          A         B
  animal hair_length                    
0 cat    long         0.875906 -2.211372
  dog    short        0.974466 -2.006747
1 cat    long        -0.410001 -0.078638
  dog    short        0.545952 -1.219217
2 cat    long        -1.226825  0.769804
  dog    short       -1.281247 -0.727707
3 cat    long        -0.121306 -0.097883
  dog    short        0.695775  0.341734

缺失数据#

如果子组没有相同的标签集,取消堆叠(unstacking)可能会导致缺失值。默认情况下,缺失值将替换为该数据类型的默认填充值。

In [40]: columns = pd.MultiIndex.from_tuples(
   ....:     [
   ....:         ("A", "cat"),
   ....:         ("B", "dog"),
   ....:         ("B", "cat"),
   ....:         ("A", "dog"),
   ....:     ],
   ....:     names=["exp", "animal"],
   ....: )
   ....: 

In [41]: index = pd.MultiIndex.from_product(
   ....:     [("bar", "baz", "foo", "qux"), ("one", "two")], names=["first", "second"]
   ....: )
   ....: 

In [42]: df = pd.DataFrame(np.random.randn(8, 4), index=index, columns=columns)

In [43]: df3 = df.iloc[[0, 1, 4, 7], [1, 2]]

In [44]: df3
Out[44]: 
exp                  B          
animal             dog       cat
first second                    
bar   one    -1.110336 -0.619976
      two     0.687738  0.176444
foo   one     1.314232  0.690579
qux   two     0.380396  0.084844

In [45]: df3.unstack()
Out[45]: 
exp            B                              
animal       dog                 cat          
second       one       two       one       two
first                                         
bar    -1.110336  0.687738 -0.619976  0.176444
foo     1.314232       NaN  0.690579       NaN
qux          NaN  0.380396       NaN  0.084844

缺失值可以通过 fill_value 参数用特定值填充。

In [46]: df3.unstack(fill_value=-1e9)
Out[46]: 
exp                B                                          
animal           dog                         cat              
second           one           two           one           two
first                                                         
bar    -1.110336e+00  6.877384e-01 -6.199759e-01  1.764443e-01
foo     1.314232e+00 -1.000000e+09  6.905793e-01 -1.000000e+09
qux    -1.000000e+09  3.803956e-01 -1.000000e+09  8.484421e-02

melt()wide_to_long()#

../_images/reshaping_melt.png

顶层函数 melt() 和对应的 DataFrame.melt() 函数非常有用,可以将 DataFrame 整理成这样一种格式:其中一个或多个列是*标识符变量*,而所有其他列(被视为*测量变量*)则被“取消透视”(unpivoted)到行轴,只剩下两个非标识符列:“variable”和“value”。这些列的名称可以通过提供 var_namevalue_name 参数进行自定义。

In [47]: cheese = pd.DataFrame(
   ....:     {
   ....:         "first": ["John", "Mary"],
   ....:         "last": ["Doe", "Bo"],
   ....:         "height": [5.5, 6.0],
   ....:         "weight": [130, 150],
   ....:     }
   ....: )
   ....: 

In [48]: cheese
Out[48]: 
  first last  height  weight
0  John  Doe     5.5     130
1  Mary   Bo     6.0     150

In [49]: cheese.melt(id_vars=["first", "last"])
Out[49]: 
  first last variable  value
0  John  Doe   height    5.5
1  Mary   Bo   height    6.0
2  John  Doe   weight  130.0
3  Mary   Bo   weight  150.0

In [50]: cheese.melt(id_vars=["first", "last"], var_name="quantity")
Out[50]: 
  first last quantity  value
0  John  Doe   height    5.5
1  Mary   Bo   height    6.0
2  John  Doe   weight  130.0
3  Mary   Bo   weight  150.0

使用 melt() 转换 DataFrame 时,索引将被忽略。可以通过将 ignore_index=False 参数设置为 False 来保留原始索引值(默认值为 True)。但是,ignore_index=False 会导致索引值重复。

In [51]: index = pd.MultiIndex.from_tuples([("person", "A"), ("person", "B")])

In [52]: cheese = pd.DataFrame(
   ....:     {
   ....:         "first": ["John", "Mary"],
   ....:         "last": ["Doe", "Bo"],
   ....:         "height": [5.5, 6.0],
   ....:         "weight": [130, 150],
   ....:     },
   ....:     index=index,
   ....: )
   ....: 

In [53]: cheese
Out[53]: 
         first last  height  weight
person A  John  Doe     5.5     130
       B  Mary   Bo     6.0     150

In [54]: cheese.melt(id_vars=["first", "last"])
Out[54]: 
  first last variable  value
0  John  Doe   height    5.5
1  Mary   Bo   height    6.0
2  John  Doe   weight  130.0
3  Mary   Bo   weight  150.0

In [55]: cheese.melt(id_vars=["first", "last"], ignore_index=False)
Out[55]: 
         first last variable  value
person A  John  Doe   height    5.5
       B  Mary   Bo   height    6.0
       A  John  Doe   weight  130.0
       B  Mary   Bo   weight  150.0

wide_to_long()melt() 类似,但在列匹配方面提供更多自定义选项。

In [56]: dft = pd.DataFrame(
   ....:     {
   ....:         "A1970": {0: "a", 1: "b", 2: "c"},
   ....:         "A1980": {0: "d", 1: "e", 2: "f"},
   ....:         "B1970": {0: 2.5, 1: 1.2, 2: 0.7},
   ....:         "B1980": {0: 3.2, 1: 1.3, 2: 0.1},
   ....:         "X": dict(zip(range(3), np.random.randn(3))),
   ....:     }
   ....: )
   ....: 

In [57]: dft["id"] = dft.index

In [58]: dft
Out[58]: 
  A1970 A1980  B1970  B1980         X  id
0     a     d    2.5    3.2  1.519970   0
1     b     e    1.2    1.3 -0.493662   1
2     c     f    0.7    0.1  0.600178   2

In [59]: pd.wide_to_long(dft, ["A", "B"], i="id", j="year")
Out[59]: 
                X  A    B
id year                  
0  1970  1.519970  a  2.5
1  1970 -0.493662  b  1.2
2  1970  0.600178  c  0.7
0  1980  1.519970  d  3.2
1  1980 -0.493662  e  1.3
2  1980  0.600178  f  0.1

get_dummies()from_dummies()#

要将 Series 中的分类变量转换为“虚拟变量”(dummy)或“指标变量”(indicator),get_dummies() 会创建一个新的 DataFrame,其中列是唯一变量,值表示每行是否存在这些变量。

In [60]: df = pd.DataFrame({"key": list("bbacab"), "data1": range(6)})

In [61]: pd.get_dummies(df["key"])
Out[61]: 
       a      b      c
0  False   True  False
1  False   True  False
2   True  False  False
3  False  False   True
4   True  False  False
5  False   True  False

In [62]: df["key"].str.get_dummies()
Out[62]: 
   a  b  c
0  0  1  0
1  0  1  0
2  1  0  0
3  0  0  1
4  1  0  0
5  0  1  0

prefix 会为列名添加前缀,这对于将结果与原始 DataFrame 合并非常有用。

In [63]: dummies = pd.get_dummies(df["key"], prefix="key")

In [64]: dummies
Out[64]: 
   key_a  key_b  key_c
0  False   True  False
1  False   True  False
2   True  False  False
3  False  False   True
4   True  False  False
5  False   True  False

In [65]: df[["data1"]].join(dummies)
Out[65]: 
   data1  key_a  key_b  key_c
0      0  False   True  False
1      1  False   True  False
2      2   True  False  False
3      3  False  False   True
4      4   True  False  False
5      5  False   True  False

此函数通常与离散化函数(如 cut())一起使用。

In [66]: values = np.random.randn(10)

In [67]: values
Out[67]: 
array([ 0.2742,  0.1329, -0.0237,  2.4102,  1.4505,  0.2061, -0.2519,
       -2.2136,  1.0633,  1.2661])

In [68]: bins = [0, 0.2, 0.4, 0.6, 0.8, 1]

In [69]: pd.get_dummies(pd.cut(values, bins))
Out[69]: 
   (0.0, 0.2]  (0.2, 0.4]  (0.4, 0.6]  (0.6, 0.8]  (0.8, 1.0]
0       False        True       False       False       False
1        True       False       False       False       False
2       False       False       False       False       False
3       False       False       False       False       False
4       False       False       False       False       False
5       False        True       False       False       False
6       False       False       False       False       False
7       False       False       False       False       False
8       False       False       False       False       False
9       False       False       False       False       False

get_dummies() 也接受 DataFrame。默认情况下,objectstringcategorical 类型的列被编码为虚拟变量,而其他列保持不变。

In [70]: df = pd.DataFrame({"A": ["a", "b", "a"], "B": ["c", "c", "b"], "C": [1, 2, 3]})

In [71]: pd.get_dummies(df)
Out[71]: 
   C    A_a    A_b    B_b    B_c
0  1   True  False  False   True
1  2  False   True  False   True
2  3   True  False   True  False

指定 columns 关键字将对任何类型的列进行编码。

In [72]: pd.get_dummies(df, columns=["A"])
Out[72]: 
   B  C    A_a    A_b
0  c  1   True  False
1  c  2  False   True
2  b  3   True  False

Series 版本类似,您可以为 prefixprefix_sep 传递值。默认情况下,列名用作前缀,_ 用作前缀分隔符。您可以通过 3 种方式指定 prefixprefix_sep

  • 字符串 (string): 为要编码的每列使用相同的 prefixprefix_sep 值。

  • 列表 (list): 长度必须与正在编码的列数相同。

  • 字典 (dict): 将列名映射到前缀。

In [73]: simple = pd.get_dummies(df, prefix="new_prefix")

In [74]: simple
Out[74]: 
   C  new_prefix_a  new_prefix_b  new_prefix_b  new_prefix_c
0  1          True         False         False          True
1  2         False          True         False          True
2  3          True         False          True         False

In [75]: from_list = pd.get_dummies(df, prefix=["from_A", "from_B"])

In [76]: from_list
Out[76]: 
   C  from_A_a  from_A_b  from_B_b  from_B_c
0  1      True     False     False      True
1  2     False      True     False      True
2  3      True     False      True     False

In [77]: from_dict = pd.get_dummies(df, prefix={"B": "from_B", "A": "from_A"})

In [78]: from_dict
Out[78]: 
   C  from_A_a  from_A_b  from_B_b  from_B_c
0  1      True     False     False      True
1  2     False      True     False      True
2  3      True     False      True     False

为了避免将结果输入统计模型时出现共线性,请指定 drop_first=True

In [79]: s = pd.Series(list("abcaa"))

In [80]: pd.get_dummies(s)
Out[80]: 
       a      b      c
0   True  False  False
1  False   True  False
2  False  False   True
3   True  False  False
4   True  False  False

In [81]: pd.get_dummies(s, drop_first=True)
Out[81]: 
       b      c
0  False  False
1   True  False
2  False   True
3  False  False
4  False  False

当列只包含一个层级时,结果中将省略该列。

In [82]: df = pd.DataFrame({"A": list("aaaaa"), "B": list("ababc")})

In [83]: pd.get_dummies(df)
Out[83]: 
    A_a    B_a    B_b    B_c
0  True   True  False  False
1  True  False   True  False
2  True   True  False  False
3  True  False   True  False
4  True  False  False   True

In [84]: pd.get_dummies(df, drop_first=True)
Out[84]: 
     B_b    B_c
0  False  False
1   True  False
2  False  False
3   True  False
4  False   True

可以使用 dtype 参数将值转换为不同的类型。

In [85]: df = pd.DataFrame({"A": list("abc"), "B": [1.1, 2.2, 3.3]})

In [86]: pd.get_dummies(df, dtype=np.float32).dtypes
Out[86]: 
B      float64
A_a    float32
A_b    float32
A_c    float32
dtype: object

在 1.5.0 版本中添加。

from_dummies()get_dummies() 的输出从指标值转换回 Series 的分类值。

In [87]: df = pd.DataFrame({"prefix_a": [0, 1, 0], "prefix_b": [1, 0, 1]})

In [88]: df
Out[88]: 
   prefix_a  prefix_b
0         0         1
1         1         0
2         0         1

In [89]: pd.from_dummies(df, sep="_")
Out[89]: 
  prefix
0      b
1      a
2      b

虚拟编码数据只需要包含 k - 1 个类别,在这种情况下,最后一个类别是默认类别。默认类别可以使用 default_category 进行修改。

In [90]: df = pd.DataFrame({"prefix_a": [0, 1, 0]})

In [91]: df
Out[91]: 
   prefix_a
0         0
1         1
2         0

In [92]: pd.from_dummies(df, sep="_", default_category="b")
Out[92]: 
  prefix
0      b
1      a
2      b

explode()#

对于一个包含嵌套的列表状值的 DataFrame 列,explode() 会将每个列表状值转换为单独的行。结果 Index 将根据原始行的索引标签进行复制。

In [93]: keys = ["panda1", "panda2", "panda3"]

In [94]: values = [["eats", "shoots"], ["shoots", "leaves"], ["eats", "leaves"]]

In [95]: df = pd.DataFrame({"keys": keys, "values": values})

In [96]: df
Out[96]: 
     keys            values
0  panda1    [eats, shoots]
1  panda2  [shoots, leaves]
2  panda3    [eats, leaves]

In [97]: df["values"].explode()
Out[97]: 
0      eats
0    shoots
1    shoots
1    leaves
2      eats
2    leaves
Name: values, dtype: object

DataFrame.explode 也可以展开(explode)DataFrame 中的列。

In [98]: df.explode("values")
Out[98]: 
     keys  values
0  panda1    eats
0  panda1  shoots
1  panda2  shoots
1  panda2  leaves
2  panda3    eats
2  panda3  leaves

Series.explode() 将用缺失值指示符替换空列表,并保留标量条目。

In [99]: s = pd.Series([[1, 2, 3], "foo", [], ["a", "b"]])

In [100]: s
Out[100]: 
0    [1, 2, 3]
1          foo
2           []
3       [a, b]
dtype: object

In [101]: s.explode()
Out[101]: 
0      1
0      2
0      3
1    foo
2    NaN
3      a
3      b
dtype: object

逗号分隔的字符串值可以拆分为列表中的单个值,然后展开(explode)到新行。

In [102]: df = pd.DataFrame([{"var1": "a,b,c", "var2": 1}, {"var1": "d,e,f", "var2": 2}])

In [103]: df.assign(var1=df.var1.str.split(",")).explode("var1")
Out[103]: 
  var1  var2
0    a     1
0    b     1
0    c     1
1    d     2
1    e     2
1    f     2

crosstab()#

使用 crosstab() 计算两个(或更多)因子的交叉表。默认情况下,除非传递一组值和一个聚合函数,否则 crosstab() 会计算因子的频率表。

任何传递的 Series 将使用其 name 属性,除非为交叉表指定了行或列名称。

In [104]: a = np.array(["foo", "foo", "bar", "bar", "foo", "foo"], dtype=object)

In [105]: b = np.array(["one", "one", "two", "one", "two", "one"], dtype=object)

In [106]: c = np.array(["dull", "dull", "shiny", "dull", "dull", "shiny"], dtype=object)

In [107]: pd.crosstab(a, [b, c], rownames=["a"], colnames=["b", "c"])
Out[107]: 
b    one        two      
c   dull shiny dull shiny
a                        
bar    1     0    0     1
foo    2     1    1     0

如果 crosstab() 只接收两个 Series,它将提供一个频率表。

In [108]: df = pd.DataFrame(
   .....:     {"A": [1, 2, 2, 2, 2], "B": [3, 3, 4, 4, 4], "C": [1, 1, np.nan, 1, 1]}
   .....: )
   .....: 

In [109]: df
Out[109]: 
   A  B    C
0  1  3  1.0
1  2  3  1.0
2  2  4  NaN
3  2  4  1.0
4  2  4  1.0

In [110]: pd.crosstab(df["A"], df["B"])
Out[110]: 
B  3  4
A      
1  1  0
2  1  3

crosstab() 也可以汇总为 Categorical 数据。

In [111]: foo = pd.Categorical(["a", "b"], categories=["a", "b", "c"])

In [112]: bar = pd.Categorical(["d", "e"], categories=["d", "e", "f"])

In [113]: pd.crosstab(foo, bar)
Out[113]: 
col_0  d  e
row_0      
a      1  0
b      0  1

对于 Categorical 数据,即使实际数据不包含特定类别的任何实例,若要包含数据的**所有**类别,请使用 dropna=False

In [114]: pd.crosstab(foo, bar, dropna=False)
Out[114]: 
col_0  d  e  f
row_0         
a      1  0  0
b      0  1  0
c      0  0  0

规范化#

频率表也可以使用 normalize 参数进行规范化,以显示百分比而不是计数。

In [115]: pd.crosstab(df["A"], df["B"], normalize=True)
Out[115]: 
B    3    4
A          
1  0.2  0.0
2  0.2  0.6

normalize 还可以规范化每行内或每列内的值。

In [116]: pd.crosstab(df["A"], df["B"], normalize="columns")
Out[116]: 
B    3    4
A          
1  0.5  0.0
2  0.5  1.0

crosstab() 还可以接受第三个 Series 和一个聚合函数 (aggfunc),该函数将应用于由前两个 Series 定义的每个组中第三个 Series 的值。

In [117]: pd.crosstab(df["A"], df["B"], values=df["C"], aggfunc="sum")
Out[117]: 
B    3    4
A          
1  1.0  NaN
2  1.0  2.0

添加边距#

margins=True 将添加带有 All 标签的行和列,其中包含跨行和列类别的部分组聚合结果。

In [118]: pd.crosstab(
   .....:     df["A"], df["B"], values=df["C"], aggfunc="sum", normalize=True, margins=True
   .....: )
   .....: 
Out[118]: 
B       3    4   All
A                   
1    0.25  0.0  0.25
2    0.25  0.5  0.75
All  0.50  0.5  1.00

cut()#

cut() 函数计算输入数组值的分组,常用于将连续变量转换为离散或分类变量。

整数值 bins 将形成等宽箱。

In [119]: ages = np.array([10, 15, 13, 12, 23, 25, 28, 59, 60])

In [120]: pd.cut(ages, bins=3)
Out[120]: 
[(9.95, 26.667], (9.95, 26.667], (9.95, 26.667], (9.95, 26.667], (9.95, 26.667], (9.95, 26.667], (26.667, 43.333], (43.333, 60.0], (43.333, 60.0]]
Categories (3, interval[float64, right]): [(9.95, 26.667] < (26.667, 43.333] < (43.333, 60.0]]

有序箱边界列表将为每个变量分配一个区间。

In [121]: pd.cut(ages, bins=[0, 18, 35, 70])
Out[121]: 
[(0, 18], (0, 18], (0, 18], (0, 18], (18, 35], (18, 35], (18, 35], (35, 70], (35, 70]]
Categories (3, interval[int64, right]): [(0, 18] < (18, 35] < (35, 70]]

如果 bins 关键字是一个 IntervalIndex,则这些将被用于对传递的数据进行分箱。

In [122]: pd.cut(ages, bins=pd.IntervalIndex.from_breaks([0, 40, 70]))
Out[122]: 
[(0, 40], (0, 40], (0, 40], (0, 40], (0, 40], (0, 40], (0, 40], (40, 70], (40, 70]]
Categories (2, interval[int64, right]): [(0, 40] < (40, 70]]

factorize()#

factorize() 将 1 维值编码为整数标签。缺失值被编码为 -1

In [123]: x = pd.Series(["A", "A", np.nan, "B", 3.14, np.inf])

In [124]: x
Out[124]: 
0       A
1       A
2     NaN
3       B
4    3.14
5     inf
dtype: object

In [125]: labels, uniques = pd.factorize(x)

In [126]: labels
Out[126]: array([ 0,  0, -1,  1,  2,  3])

In [127]: uniques
Out[127]: Index(['A', 'B', 3.14, inf], dtype='object')

Categorical 将类似地编码 1 维值,以便进行进一步的分类操作。

In [128]: pd.Categorical(x)
Out[128]: 
['A', 'A', NaN, 'B', 3.14, inf]
Categories (4, object): [3.14, inf, 'A', 'B']