您的位置：首页 > 编程语言 > Python开发

利用Python进行数据分析——数据规整化：清理、转换、合并、重塑(七)(4) .

2014-07-03 13:48 971 查看
1、数据转换

目前为止介绍的都是数据的重排。另一类重要操作则是过滤、清理以及其他的转换工作。

2、移除重复数据

DataFrame中常常会出现重复行。下面就是一个例子：

[python]
view plaincopyprint?





In [4]: data = pd.DataFrame({'k1':['one'] * 3 + ['two'] * 4,
'k2':[1, 1, 2, 3, 3, 4, 4]})

In [5]: data
Out[5]:
k1 k2
0 one 1
1 one 1
2 one 2
3 two 3
4 two 3
5 two 4
6 two 4

[7 rows x 2 columns]

In [4]: data = pd.DataFrame({'k1':['one'] * 3 + ['two'] * 4,
'k2':[1, 1, 2, 3, 3, 4, 4]})

In [5]: data
Out[5]:
k1  k2
0  one   1
1  one   1
2  one   2
3  two   3
4  two   3
5  two   4
6  two   4

[7 rows x 2 columns]
DataFrame的duplicated方法返回一个布尔型Series，表示各行是否是重复行：

[python]
view plaincopyprint?





In [6]: data.duplicated() Out[6]: 0 False 1 True 2 False 3 False 4 True 5 False 6 True dtype: bool
In [6]: data.duplicated()
Out[6]:
0    False
1     True
2    False
3    False
4     True
5    False
6     True
dtype: bool
还有一个与此相关的drop_duplicates方法，它用于返回一个移除了重复行的DataFrame：

[python]
view plaincopyprint?





In [7]: data.drop_duplicates() Out[7]: k1 k2 0 one 1 2 one 2 3 two 3 5 two 4 [4 rows x 2 columns]
In [7]: data.drop_duplicates()
Out[7]:
k1  k2
0  one   1
2  one   2
3  two   3
5  two   4

[4 rows x 2 columns]
这两个方法默认会判断全部列，你也可以指定部分列进行重复项判断。假设你还有一列值，且只希望根据k1列过滤重复项：

[python]
view plaincopyprint?





In [8]: data['v1'] = range(7)

In [9]: data
Out[9]:
k1 k2 v1
0 one 1 0
1 one 1 1
2 one 2 2
3 two 3 3
4 two 3 4
5 two 4 5
6 two 4 6

[7 rows x 3 columns]

In [10]: data.drop_duplicates(['k1'])
Out[10]:
k1 k2 v1
0 one 1 0
3 two 3 3

[2 rows x 3 columns]

In [8]: data['v1'] = range(7)

In [9]: data
Out[9]:
k1  k2  v1
0  one   1   0
1  one   1   1
2  one   2   2
3  two   3   3
4  two   3   4
5  two   4   5
6  two   4   6

[7 rows x 3 columns]

In [10]: data.drop_duplicates(['k1'])
Out[10]:
k1  k2  v1
0  one   1   0
3  two   3   3

[2 rows x 3 columns]
duplicated和drop_duplicates默认保留的是第一个出现的值组合。传入take_last=True则保留最后一个：

[python]
view plaincopyprint?





In [11]: data.drop_duplicates(['k1', 'k2'], take_last=True)
Out[11]:
k1 k2 v1
1 one 1 1
2 one 2 2
4 two 3 4
6 two 4 6

[4 rows x 3 columns]

In [11]: data.drop_duplicates(['k1', 'k2'], take_last=True)
Out[11]:
k1  k2  v1
1  one   1   1
2  one   2   2
4  two   3   4
6  two   4   6

[4 rows x 3 columns]


3、利用函数或映射进行数据转换

在对数据集进行转换时，你可能希望根据数组、Series或DataFrame列中的值来实现该转换工作。我们来看看下面这组有关肉类的数据：

[python]
view plaincopyprint?





In [12]: data = pd.DataFrame({'food':['bacon', 'pulled pork', 'bacon', 'Pastrami', 'corned beef', 'Bacon',
'pastrami', 'honey ham', 'nova lox'],
....: 'ounces':[4, 3, 12, 6, 7.5, 8, 3, 5, 6]})

In [13]: data
Out[13]:
food ounces
0 bacon 4.0
1 pulled pork 3.0
2 bacon 12.0
3 Pastrami 6.0
4 corned beef 7.5
5 Bacon 8.0
6 pastrami 3.0
7 honey ham 5.0
8 nova lox 6.0

[9 rows x 2 columns]

In [12]: data = pd.DataFrame({'food':['bacon', 'pulled pork', 'bacon', 'Pastrami', 'corned beef', 'Bacon',
                                      'pastrami', 'honey ham', 'nova lox'],
....:                     'ounces':[4, 3, 12, 6, 7.5, 8, 3, 5, 6]})

In [13]: data
Out[13]:
food  ounces
0        bacon     4.0
1  pulled pork     3.0
2        bacon    12.0
3     Pastrami     6.0
4  corned beef     7.5
5        Bacon     8.0
6     pastrami     3.0
7    honey ham     5.0
8     nova lox     6.0

[9 rows x 2 columns]
假设你想要添加一列表示该肉类食物来源的动物类型。我们先编写一个肉类到动物的映射：

[python]
view plaincopyprint?





In [14]: meat_to_animal = {
....: 'bacon': 'pig',
....: 'pulled pork': 'pig',
....: 'pastrami': 'cow',
....: 'corned beef': 'cow',
....: 'honey ham': 'pig',
....: 'nova lox': 'salmon'
....: }

In [14]: meat_to_animal = {
....:        'bacon': 'pig',
....:        'pulled pork': 'pig',
....:        'pastrami': 'cow',
....:        'corned beef': 'cow',
....:        'honey ham': 'pig',
....:        'nova lox': 'salmon'
....: }
Series的map方法可以接受一个函数或含有映射关系的字典型对象，但是这里有一个小问题，即有些肉类的首字母大写了，而另一些则没有。因此，我们还需要将各个值转换为小写：

[python]
view plaincopyprint?





In [15]: data['animal'] = data['food'].map(str.lower).map(meat_to_animal)

In [16]: data
Out[16]:
food ounces animal
0 bacon 4.0 pig
1 pulled pork 3.0 pig
2 bacon 12.0 pig
3 Pastrami 6.0 cow
4 corned beef 7.5 cow
5 Bacon 8.0 pig
6 pastrami 3.0 cow
7 honey ham 5.0 pig
8 nova lox 6.0 salmon

[9 rows x 3 columns]

In [15]: data['animal'] = data['food'].map(str.lower).map(meat_to_animal)

In [16]: data
Out[16]:
food  ounces  animal
0        bacon     4.0     pig
1  pulled pork     3.0     pig
2        bacon    12.0     pig
3     Pastrami     6.0     cow
4  corned beef     7.5     cow
5        Bacon     8.0     pig
6     pastrami     3.0     cow
7    honey ham     5.0     pig
8     nova lox     6.0  salmon

[9 rows x 3 columns]
我们也可以传入一个能够完成全部这些工作的函数：

[python]
view plaincopyprint?





In [17]: data['food'].map(lambda x: meat_to_animal[x.lower()])
Out[17]:
0 pig
1 pig
2 pig
3 cow
4 cow
5 pig
6 cow
7 pig
8 salmon
Name: food, dtype: object

In [17]: data['food'].map(lambda x: meat_to_animal[x.lower()])
Out[17]:
0       pig
1       pig
2       pig
3       cow
4       cow
5       pig
6       cow
7       pig
8    salmon
Name: food, dtype: object

说明：

使用map是一种实现元素级转换以及其他数据清理工作的便捷方式。

4、替换值

利用fillna方法填充缺失数据可以看做值替换的一种特殊情况。虽然前面提到的mao可用于修改对象的数据子集，而replace则提供了一种实现该功能的更简单、更灵活的方式。我们来看看下面这个Series：

[python]
view plaincopyprint?





In [18]: data = pd.Series([1., -999, 2., -999, -1000., 3.]) In [19]: data Out[19]: 0 1 1 -999 2 2 3 -999 4 -1000 5 3 dtype: float64
In [18]: data = pd.Series([1., -999, 2., -999, -1000., 3.])

In [19]: data
Out[19]:
0       1
1    -999
2       2
3    -999
4   -1000
5       3
dtype: float64
-999这个值可能是一个表示缺失数据的标记值。要将其替换为pandas能够理解的NA值，我们可以利用replace来产生一个新的Series：

[python]
view plaincopyprint?





In [20]: data.replace(-999, np.nan) Out[20]: 0 1 1 NaN 2 2 3 NaN 4 -1000 5 3 dtype: float64
In [20]: data.replace(-999, np.nan)
Out[20]:
0       1
1     NaN
2       2
3     NaN
4   -1000
5       3
dtype: float64
如果你希望一次性替换多个值，可以传入一个由待替换值组成的列表以及一个替换值：

[python]
view plaincopyprint?





In [21]: data.replace([-999, -1000], np.nan) Out[21]: 0 1 1 NaN 2 2 3 NaN 4 NaN 5 3 dtype: float64
In [21]: data.replace([-999, -1000], np.nan)
Out[21]:
0     1
1   NaN
2     2
3   NaN
4   NaN
5     3
dtype: float64
如果希望对不同的值进行不同的替换，则传入一个由替换关系组成的列表即可：

[python]
view plaincopyprint?





In [22]: data.replace([-999, -1000], [np.nan, 0]) Out[22]: 0 1 1 NaN 2 2 3 NaN 4 0 5 3 dtype: float64
In [22]: data.replace([-999, -1000], [np.nan, 0])
Out[22]:
0     1
1   NaN
2     2
3   NaN
4     0
5     3
dtype: float64
传入的参数也可以是字典：

[python]
view plaincopyprint?





In [23]: data.replace({-999: np.nan, -1000: 0}) Out[23]: 0 1 1 NaN 2 2 3 NaN 4 0 5 3 dtype: float64
In [23]: data.replace({-999: np.nan, -1000: 0})
Out[23]:
0     1
1   NaN
2     2
3   NaN
4     0
5     3
dtype: float64


5、重命名轴索引

跟Series中的值一样，轴标签也可以通过函数或映射进行转换，从而得到一个新对象。轴还可以被就地修改，而无需新建一个数据结构。接下来看看下面这个简单的例子：

[python]
view plaincopyprint?





In [24]: data = pd.DataFrame(np.arange(12).reshape((3, 4)),
....: index=['Ohio', 'Colorado', 'New York'],
....: columns=['one', 'two', 'three', 'four'])

In [24]: data = pd.DataFrame(np.arange(12).reshape((3, 4)),
....:                     index=['Ohio', 'Colorado', 'New York'],
....:                     columns=['one', 'two', 'three', 'four'])
跟Series一样，轴标签也有一个map方法：

[python]
view plaincopyprint?





In [25]: data.index.map(str.upper)
Out[25]: array(['OHIO', 'COLORADO', 'NEW YORK'], dtype=object)

In [25]: data.index.map(str.upper)
Out[25]: array(['OHIO', 'COLORADO', 'NEW YORK'], dtype=object)
你可以将其赋值给index，这样就可以对DataFrame进行就地修改了：

[python]
view plaincopyprint?





In [26]: data.index = data.index.map(str.upper) In [27]: data Out[27]: one two three four OHIO 0 1 2 3 COLORADO 4 5 6 7 NEW YORK 8 9 10 11 [3 rows x 4 columns]
In [26]: data.index = data.index.map(str.upper)

In [27]: data
Out[27]:
one  two  three  four
OHIO        0    1      2     3
COLORADO    4    5      6     7
NEW YORK    8    9     10    11

[3 rows x 4 columns]
如果想要创建数据集的转换版（而不是修改原始数据），比较使用的方式是rename：

[python]
view plaincopyprint?





In [28]: data.rename(index=str.title, columns=str.upper) Out[28]: ONE TWO THREE FOUR Ohio 0 1 2 3 Colorado 4 5 6 7 New York 8 9 10 11 [3 rows x 4 columns]
In [28]: data.rename(index=str.title, columns=str.upper)
Out[28]:
ONE  TWO  THREE  FOUR
Ohio        0    1      2     3
Colorado    4    5      6     7
New York    8    9     10    11

[3 rows x 4 columns]
特别说明一下，rename可以结合字典型对象实现对部分轴标签的更新：

[python]
view plaincopyprint?





In [31]: data.rename(index={'OHIO': 'INDIANA'},
columns={'three': 'peekaboo'})
Out[31]:
one two peekaboo four
INDIANA 0 1 2 3
COLORADO 4 5 6 7
NEW YORK 8 9 10 11

[3 rows x 4 columns]

In [31]: data.rename(index={'OHIO': 'INDIANA'},
columns={'three': 'peekaboo'})
Out[31]:
one  two  peekaboo  four
INDIANA     0    1         2     3
COLORADO    4    5         6     7
NEW YORK    8    9        10    11

[3 rows x 4 columns]
rename帮我们实现了：复制DataFrame并对其索引和列标签进行赋值。如果希望就地修改某个数据集，传入inplace=True即可：

[python]
view plaincopyprint?





In [32]: # 总是返回DataFrame的引用

In [33]: _ = data.rename(index={'OHIO': 'INDIANA'}, inplace=True)

In [34]: data
Out[34]:
one two three four
INDIANA 0 1 2 3
COLORADO 4 5 6 7
NEW YORK 8 9 10 11

[3 rows x 4 columns]

In [32]: # 总是返回DataFrame的引用

In [33]: _ = data.rename(index={'OHIO': 'INDIANA'}, inplace=True)

In [34]: data
Out[34]:
one  two  three  four
INDIANA     0    1      2     3
COLORADO    4    5      6     7
NEW YORK    8    9     10    11

[3 rows x 4 columns]


6、离散化和面元划分

为了便于分析，连续数据常常被离散化或拆分为“面元”（bin）。假设有一组人员数据，而你希望将它们划分为不同的年龄组：

[python]
view plaincopyprint?





In [35]: ages = [20, 22, 25, 27, 21, 23, 37, 31, 61, 45, 41, 32]
In [35]: ages = [20, 22, 25, 27, 21, 23, 37, 31, 61, 45, 41, 32]
接下来将这些数据划分为“18到25”、“26到35”、“35到60”以及“60以上”几个面元。要实现该功能，你需要使用pandas的cut函数：

[python]
view plaincopyprint?





In [36]: bins = [18, 25, 35, 60, 100]

In [37]: cats = pd.cut(ages, bins)

In [38]: cats
Out[38]:
(18, 25]
(18, 25]
(18, 25]
(25, 35]
(18, 25]
(18, 25]
(35, 60]
(25, 35]
(60, 100]
(35, 60]
(35, 60]
(25, 35]
Levels (4): Index(['(18, 25]', '(25, 35]', '(35, 60]', '(60, 100]'], dtype=object)

In [36]: bins = [18, 25, 35, 60, 100]

In [37]: cats = pd.cut(ages, bins)

In [38]: cats
Out[38]:
(18, 25]
(18, 25]
(18, 25]
(25, 35]
(18, 25]
(18, 25]
(35, 60]
(25, 35]
(60, 100]
(35, 60]
(35, 60]
(25, 35]
Levels (4): Index(['(18, 25]', '(25, 35]', '(35, 60]', '(60, 100]'], dtype=object)
pandas返回的是一个特殊的Categorical对象。你可以将其看做一组表示面元名称的字符串。实际上，它含有一个表示不同分类名称的levels数组以及一个为年龄数据进行标号的labels属性：

[python]
view plaincopyprint?





In [39]: cats.labels
Out[39]: array([0, 0, 0, 1, 0, 0, 2, 1, 3, 2, 2, 1])

In [40]: cats.levels
Out[40]: Index([u'(18, 25]', u'(25, 35]', u'(35, 60]', u'(60, 100]'], dtype='object')

In [41]: pd.value_counts(cats)
Out[41]:
(18, 25] 5
(35, 60] 3
(25, 35] 3
(60, 100] 1
dtype: int64

In [39]: cats.labels
Out[39]: array([0, 0, 0, 1, 0, 0, 2, 1, 3, 2, 2, 1])

In [40]: cats.levels
Out[40]: Index([u'(18, 25]', u'(25, 35]', u'(35, 60]', u'(60, 100]'], dtype='object')

In [41]: pd.value_counts(cats)
Out[41]:
(18, 25]     5
(35, 60]     3
(25, 35]     3
(60, 100]    1
dtype: int64
跟“区间”的数学符号一样，园括号表示开端，而方括号则表示闭断（包括）。哪边是闭端可以通过right=False进行修改：

[python]
view plaincopyprint?





In [42]: pd.cut(ages, [18, 26, 36, 61, 100], right=False)
Out[42]:
[18, 26)
[18, 26)
[18, 26)
[26, 36)
[18, 26)
[18, 26)
[36, 61)
[26, 36)
[61, 100)
[36, 61)
[36, 61)
[26, 36)
Levels (4): Index(['[18, 26)', '[26, 36)', '[36, 61)', '[61, 100)'], dtype=object)

In [42]: pd.cut(ages, [18, 26, 36, 61, 100], right=False)
Out[42]:
[18, 26)
[18, 26)
[18, 26)
[26, 36)
[18, 26)
[18, 26)
[36, 61)
[26, 36)
[61, 100)
[36, 61)
[36, 61)
[26, 36)
Levels (4): Index(['[18, 26)', '[26, 36)', '[36, 61)', '[61, 100)'], dtype=object)
你也可以设置自己的面元名称，将labels选项设置为一个列表或数组即可：

[python]
view plaincopyprint?





In [43]: group_names = ['Youth', 'YoungAdult', 'MiddleAged', 'Senior']

In [44]: pd.cut(ages, bins, labels=group_names)
Out[44]:
Youth
Youth
Youth
YoungAdult
Youth
Youth
MiddleAged
YoungAdult
Senior
MiddleAged
MiddleAged
YoungAdult
Levels (4): Index(['Youth', 'YoungAdult', 'MiddleAged', 'Senior'], dtype=object)

In [43]: group_names = ['Youth', 'YoungAdult', 'MiddleAged', 'Senior']

In [44]: pd.cut(ages, bins, labels=group_names)
Out[44]:
Youth
Youth
Youth
YoungAdult
Youth
Youth
MiddleAged
YoungAdult
Senior
MiddleAged
MiddleAged
YoungAdult
Levels (4): Index(['Youth', 'YoungAdult', 'MiddleAged', 'Senior'], dtype=object)
如果向cut传入的是面元的数量而不是确切的面元边界，则它会根据数据的最小值和最大值计算等长面元。下面这个例子中，我们将一些均匀分布的数据分成四组：

[python]
view plaincopyprint?





In [45]: data = np.random.rand(20)

In [46]: pd.cut(data, 4, precision=2)
Out[46]:
(0.037, 0.26]
(0.037, 0.26]
(0.48, 0.7]
(0.7, 0.92]
(0.037, 0.26]
(0.037, 0.26]
(0.7, 0.92]
(0.7, 0.92]
(0.037, 0.26]
(0.26, 0.48]
(0.26, 0.48]
(0.26, 0.48]
(0.037, 0.26]
(0.26, 0.48]
(0.48, 0.7]
(0.7, 0.92]
(0.037, 0.26]
(0.7, 0.92]
(0.037, 0.26]
(0.037, 0.26]
Levels (4): Index(['(0.037, 0.26]', '(0.26, 0.48]', '(0.48, 0.7]',
'(0.7, 0.92]'], dtype=object)

In [45]: data = np.random.rand(20)

In [46]: pd.cut(data, 4, precision=2)
Out[46]:
(0.037, 0.26]
(0.037, 0.26]
(0.48, 0.7]
(0.7, 0.92]
(0.037, 0.26]
(0.037, 0.26]
(0.7, 0.92]
(0.7, 0.92]
(0.037, 0.26]
(0.26, 0.48]
(0.26, 0.48]
(0.26, 0.48]
(0.037, 0.26]
(0.26, 0.48]
(0.48, 0.7]
(0.7, 0.92]
(0.037, 0.26]
(0.7, 0.92]
(0.037, 0.26]
(0.037, 0.26]
Levels (4): Index(['(0.037, 0.26]', '(0.26, 0.48]', '(0.48, 0.7]',
'(0.7, 0.92]'], dtype=object)
qcut是一个非常类似于cut的函数，它可以根据样本分位数对数据进行面元划分。根据数据的分布情况，cut可能无法使各个面元中含有相同数量的数据点。而qcut由于使用的是样本分位数，因此可以得到大小基本相等的面元：

[python]
view plaincopyprint?





In [48]: data = np.random.randn(1000) # 正态分布

In [49]: cats = pd.qcut(data, 4) # 按四分位数进行分隔

In [50]: cats
Out[50]:
[-3.636, -0.717]
(0.647, 3.531]
[-3.636, -0.717]
[-3.636, -0.717]
[-3.636, -0.717]
(0.647, 3.531]
[-3.636, -0.717]
(-0.717, -0.0323]
(-0.717, -0.0323]
(0.647, 3.531]
[-3.636, -0.717]
(-0.717, -0.0323]
(0.647, 3.531]
...
[-3.636, -0.717]
[-3.636, -0.717]
(0.647, 3.531]
(-0.717, -0.0323]
(0.647, 3.531]
[-3.636, -0.717]
[-3.636, -0.717]
(-0.0323, 0.647]
[-3.636, -0.717]
(-0.717, -0.0323]
(-0.717, -0.0323]
(-0.0323, 0.647]
(0.647, 3.531]
Levels (4): Index(['[-3.636, -0.717]', '(-0.717, -0.0323]',
'(-0.0323, 0.647]', '(0.647, 3.531]'], dtype=object)
Length: 1000

In [51]: pd.value_counts(cats)
Out[51]:
(-0.717, -0.0323] 250
(-0.0323, 0.647] 250
(0.647, 3.531] 250
[-3.636, -0.717] 250
dtype: int64

In [48]: data = np.random.randn(1000) # 正态分布

In [49]: cats = pd.qcut(data, 4) # 按四分位数进行分隔

In [50]: cats
Out[50]:
[-3.636, -0.717]
(0.647, 3.531]
[-3.636, -0.717]
[-3.636, -0.717]
[-3.636, -0.717]
(0.647, 3.531]
[-3.636, -0.717]
(-0.717, -0.0323]
(-0.717, -0.0323]
(0.647, 3.531]
[-3.636, -0.717]
(-0.717, -0.0323]
(0.647, 3.531]
...
[-3.636, -0.717]
[-3.636, -0.717]
(0.647, 3.531]
(-0.717, -0.0323]
(0.647, 3.531]
[-3.636, -0.717]
[-3.636, -0.717]
(-0.0323, 0.647]
[-3.636, -0.717]
(-0.717, -0.0323]
(-0.717, -0.0323]
(-0.0323, 0.647]
(0.647, 3.531]
Levels (4): Index(['[-3.636, -0.717]', '(-0.717, -0.0323]',
'(-0.0323, 0.647]', '(0.647, 3.531]'], dtype=object)
Length: 1000

In [51]: pd.value_counts(cats)
Out[51]:
(-0.717, -0.0323]    250
(-0.0323, 0.647]     250
(0.647, 3.531]       250
[-3.636, -0.717]     250
dtype: int64
跟cut一样，也可以设置自定义的分位数（0到1之间的数值，包含端点）：

[python]
view plaincopyprint?





In [52]: pd.qcut(data, [0, 0.1, 0.5, 0.9, 1.])
Out[52]:
(-1.323, -0.0323]
(-0.0323, 1.234]
(-1.323, -0.0323]
[-3.636, -1.323]
[-3.636, -1.323]
(-0.0323, 1.234]
(-1.323, -0.0323]
(-1.323, -0.0323]
(-1.323, -0.0323]
(1.234, 3.531]
(-1.323, -0.0323]
(-1.323, -0.0323]
(-0.0323, 1.234]
...
[-3.636, -1.323]
(-1.323, -0.0323]
(-0.0323, 1.234]
(-1.323, -0.0323]
(-0.0323, 1.234]
[-3.636, -1.323]
(-1.323, -0.0323]
(-0.0323, 1.234]
(-1.323, -0.0323]
(-1.323, -0.0323]
(-1.323, -0.0323]
(-0.0323, 1.234]
(-0.0323, 1.234]
Levels (4): Index(['[-3.636, -1.323]', '(-1.323, -0.0323]',
'(-0.0323, 1.234]', '(1.234, 3.531]'], dtype=object)
Length: 1000

In [52]: pd.qcut(data, [0, 0.1, 0.5, 0.9, 1.])
Out[52]:
(-1.323, -0.0323]
(-0.0323, 1.234]
(-1.323, -0.0323]
[-3.636, -1.323]
[-3.636, -1.323]
(-0.0323, 1.234]
(-1.323, -0.0323]
(-1.323, -0.0323]
(-1.323, -0.0323]
(1.234, 3.531]
(-1.323, -0.0323]
(-1.323, -0.0323]
(-0.0323, 1.234]
...
[-3.636, -1.323]
(-1.323, -0.0323]
(-0.0323, 1.234]
(-1.323, -0.0323]
(-0.0323, 1.234]
[-3.636, -1.323]
(-1.323, -0.0323]
(-0.0323, 1.234]
(-1.323, -0.0323]
(-1.323, -0.0323]
(-1.323, -0.0323]
(-0.0323, 1.234]
(-0.0323, 1.234]
Levels (4): Index(['[-3.636, -1.323]', '(-1.323, -0.0323]',
'(-0.0323, 1.234]', '(1.234, 3.531]'], dtype=object)
Length: 1000

说明：

稍后在讲解聚合和分组运算时会再次用到cut和qcut，因为这两个离散化函数对分量和分组分析非常重要。

7、检测和过滤异常值

异常值（outlier）的过滤或变换运算在很大程度上其实就是数组运算。来看一个含有正态分布数据的DataFrame：

[python]
view plaincopyprint?





In [53]: np.random.seed(12345) In [54]: data = pd.DataFrame(np.random.randn(1000, 4)) In [55]: data.describe() Out[55]: 0 1 2 3 count 1000.000000 1000.000000 1000.000000 1000.000000 mean -0.067684 0.067924 0.025598 -0.002298 std 0.998035 0.992106 1.006835 0.996794 min -3.428254 -3.548824 -3.184377 -3.745356 25% -0.774890 -0.591841 -0.641675 -0.644144 50% -0.116401 0.101143 0.002073 -0.013611 75% 0.616366 0.780282 0.680391 0.654328 max 3.366626 2.653656 3.260383 3.927528 [8 rows x 4 columns]
In [53]: np.random.seed(12345)

In [54]: data = pd.DataFrame(np.random.randn(1000, 4))

In [55]: data.describe()
Out[55]:
0            1            2            3
count  1000.000000  1000.000000  1000.000000  1000.000000
mean     -0.067684     0.067924     0.025598    -0.002298
std       0.998035     0.992106     1.006835     0.996794
min      -3.428254    -3.548824    -3.184377    -3.745356
25%      -0.774890    -0.591841    -0.641675    -0.644144
50%      -0.116401     0.101143     0.002073    -0.013611
75%       0.616366     0.780282     0.680391     0.654328
max       3.366626     2.653656     3.260383     3.927528

[8 rows x 4 columns]
假设你想要找出某列中绝对值大小超过3的值：

[python]
view plaincopyprint?





In [56]: col = data[3] In [57]: col[np.abs(col) > 3] Out[57]: 97 3.927528 305 -3.399312 400 -3.745356 Name: 3, dtype: float64
In [56]: col = data[3]

In [57]: col[np.abs(col) > 3]
Out[57]:
97     3.927528
305   -3.399312
400   -3.745356
Name: 3, dtype: float64
要选出全部含有“超过3或-3的值”的行，你可以利用布尔型DataFrame以及any方法：

[python]
view plaincopyprint?





In [58]: data[(np.abs(data) > 3).any(1)] Out[58]: 0 1 2 3 5 -0.539741 0.476985 3.248944 -1.021228 97 -0.774363 0.552936 0.106061 3.927528 102 -0.655054 -0.565230 3.176873 0.959533 305 -2.315555 0.457246 -0.025907 -3.399312 324 0.050188 1.951312 3.260383 0.963301 400 0.146326 0.508391 -0.196713 -3.745356 499 -0.293333 -0.242459 -3.056990 1.918403 523 -3.428254 -0.296336 -0.439938 -0.867165 586 0.275144 1.179227 -3.184377 1.369891 808 -0.362528 -3.548824 1.553205 -2.186301 900 3.366626 -2.372214 0.851010 1.332846 [11 rows x 4 columns]
In [58]: data[(np.abs(data) > 3).any(1)]
Out[58]:
0         1         2         3
5   -0.539741  0.476985  3.248944 -1.021228
97  -0.774363  0.552936  0.106061  3.927528
102 -0.655054 -0.565230  3.176873  0.959533
305 -2.315555  0.457246 -0.025907 -3.399312
324  0.050188  1.951312  3.260383  0.963301
400  0.146326  0.508391 -0.196713 -3.745356
499 -0.293333 -0.242459 -3.056990  1.918403
523 -3.428254 -0.296336 -0.439938 -0.867165
586  0.275144  1.179227 -3.184377  1.369891
808 -0.362528 -3.548824  1.553205 -2.186301
900  3.366626 -2.372214  0.851010  1.332846

[11 rows x 4 columns]
根据这些条件，即可轻松地对值进行设置。下面的代码可以将值限制在区间-3到3以内：

[python]
view plaincopyprint?





In [59]: data[np.abs(data) > 3] = np.sign(data) * 3 In [60]: data.describe() Out[60]: 0 1 2 3 count 1000.000000 1000.000000 1000.000000 1000.000000 mean -0.067623 0.068473 0.025153 -0.002081 std 0.995485 0.990253 1.003977 0.989736 min -3.000000 -3.000000 -3.000000 -3.000000 25% -0.774890 -0.591841 -0.641675 -0.644144 50% -0.116401 0.101143 0.002073 -0.013611 75% 0.616366 0.780282 0.680391 0.654328 max 3.000000 2.653656 3.000000 3.000000 [8 rows x 4 columns]
In [59]: data[np.abs(data) > 3] = np.sign(data) * 3

In [60]: data.describe()
Out[60]:
0            1            2            3
count  1000.000000  1000.000000  1000.000000  1000.000000
mean     -0.067623     0.068473     0.025153    -0.002081
std       0.995485     0.990253     1.003977     0.989736
min      -3.000000    -3.000000    -3.000000    -3.000000
25%      -0.774890    -0.591841    -0.641675    -0.644144
50%      -0.116401     0.101143     0.002073    -0.013611
75%       0.616366     0.780282     0.680391     0.654328
max       3.000000     2.653656     3.000000     3.000000

[8 rows x 4 columns]

说明：

np.sign这个ufunc返回的是一个由1和-1组成的数组，表示原始值的符号。

8、排列和随机采样

利用numpy.random.permutation函数可以轻松实现对Series或DataFrame的列的排列工作（permuting，随机重排序）。通过需要排列的轴的长度调用permutation，可产生一个表示新顺序的整数数组：

[python]
view plaincopyprint?





In [61]: df = pd.DataFrame(np.arange(5 * 4).reshape(5, 4)) In [62]: sampler = np.random.permutation(5) In [63]: sampler Out[63]: array([1, 0, 2, 3, 4])
In [61]: df = pd.DataFrame(np.arange(5 * 4).reshape(5, 4))

In [62]: sampler = np.random.permutation(5)

In [63]: sampler
Out[63]: array([1, 0, 2, 3, 4])
然后就可以在基于ix的索引操作或take函数中使用该数组了：

[python]
view plaincopyprint?





In [64]: df Out[64]: 0 1 2 3 0 0 1 2 3 1 4 5 6 7 2 8 9 10 11 3 12 13 14 15 4 16 17 18 19 [5 rows x 4 columns] In [65]: df.take(sampler) Out[65]: 0 1 2 3 1 4 5 6 7 0 0 1 2 3 2 8 9 10 11 3 12 13 14 15 4 16 17 18 19 [5 rows x 4 columns]
In [64]: df
Out[64]:
0   1   2   3
0   0   1   2   3
1   4   5   6   7
2   8   9  10  11
3  12  13  14  15
4  16  17  18  19

[5 rows x 4 columns]

In [65]: df.take(sampler)
Out[65]:
0   1   2   3
1   4   5   6   7
0   0   1   2   3
2   8   9  10  11
3  12  13  14  15
4  16  17  18  19

[5 rows x 4 columns]
如果不想用替换的方式选取随机子集，则可以使用permutation：从permutation返回的数组中切下前k个元素，其中k为期望的子集大小。虽然有很多高效的算法可以实现非替换式采样，但是手边就有的工具为什么不用呢？

[python]
view plaincopyprint?





In [66]: df.take(np.random.permutation(len(df))[:3]) Out[66]: 0 1 2 3 1 4 5 6 7 3 12 13 14 15 4 16 17 18 19 [3 rows x 4 columns]
In [66]: df.take(np.random.permutation(len(df))[:3])
Out[66]:
0   1   2   3
1   4   5   6   7
3  12  13  14  15
4  16  17  18  19

[3 rows x 4 columns]
要通过替换的方式产生样本，最快的方式是通过np.random.randint得到一组随机整数：

[python]
view plaincopyprint?





In [67]: bag = np.array([5, 7, -1, 6, 4]) In [68]: sampler = np.random.randint(0, len(bag), size=10) In [69]: sampler Out[69]: array([4, 4, 2, 2, 2, 0, 3, 0, 4, 1]) In [70]: draws = bag.take(sampler) In [71]: draws Out[71]: array([ 4, 4, -1, -1, -1, 5, 6, 5, 4, 7])
In [67]: bag = np.array([5, 7, -1, 6, 4])

In [68]: sampler = np.random.randint(0, len(bag), size=10)

In [69]: sampler
Out[69]: array([4, 4, 2, 2, 2, 0, 3, 0, 4, 1])

In [70]: draws = bag.take(sampler)

In [71]: draws
Out[71]: array([ 4,  4, -1, -1, -1,  5,  6,  5,  4,  7])


9、计算指标/哑变量

另一种常用于统计建模或机器学习的转换方式是：将分类变量（categorical variable）转换为“哑变量矩阵”（dummy matrix）或“指标矩阵”（indicator matrix）。如果DataFrame的某一列中含有k个不同的值，则可以派生出一个k列矩阵或DataFrame（其值全为1和0）。pandas有一个get_dummies函数可以实现该功能（其实自己动手做一个也不难）。拿之前的一个例子来说：

[python]
view plaincopyprint?





In [72]: df = pd.DataFrame({'key': ['b', 'b', 'a', 'c', 'a', 'b'],
....: 'data1': range(6)})

In [73]: pd.get_dummies(df['key'])
Out[73]:
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

[6 rows x 3 columns]

In [72]: df = pd.DataFrame({'key': ['b', 'b', 'a', 'c', 'a', 'b'],
....:                    'data1': range(6)})

In [73]: pd.get_dummies(df['key'])
Out[73]:
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

[6 rows x 3 columns]
有时候，你可能想给指标DataFrame的列加上一个前缀，以便能够跟其他数据进行合并。get_dummies的prefix参数可以实现该功能：

[python]
view plaincopyprint?





In [74]: dummies = pd.get_dummies(df['key'], prefix='key')

In [75]: df_with_dummy = df[['data1']].join(dummies)

In [76]: df_with_dummy
Out[76]:
data1 key_a key_b key_c
0 0 0 1 0
1 1 0 1 0
2 2 1 0 0
3 3 0 0 1
4 4 1 0 0
5 5 0 1 0

[6 rows x 4 columns]

In [74]: dummies = pd.get_dummies(df['key'], prefix='key')

In [75]: df_with_dummy = df[['data1']].join(dummies)

In [76]: df_with_dummy
Out[76]:
data1  key_a  key_b  key_c
0      0      0      1      0
1      1      0      1      0
2      2      1      0      0
3      3      0      0      1
4      4      1      0      0
5      5      0      1      0

[6 rows x 4 columns]
如果DataFrame中的某行同属于多个分类，则事情就会有点复杂。根据MovieLens 1M数据集，如下所示：

[python]
view plaincopyprint?





In [77]: mnames = ['movie_id', 'title', 'genres']
In [78]: movies = pd.read_table('movies.dat', sep='::', header=None,
.....: names=mnames)

In [79]: movies[:10]
Out[79]:
movie_id title genres
0 1 Toy Story (1995) Animation|Children's|Comedy
1 2 Jumanji (1995) Adventure|Children's|Fantasy
2 3 Grumpier Old Men (1995) Comedy|Romance
3 4 Waiting to Exhale (1995) Comedy|Drama
4 5 Father of the Bride Part II (1995) Comedy
5 6 Heat (1995) Action|Crime|Thriller
6 7 Sabrina (1995) Comedy|Romance
7 8 Tom and Huck (1995) Adventure|Children's
8 9 Sudden Death (1995) Action
9 10 GoldenEye (1995) Action|Adventure|Thriller

In [77]: mnames = ['movie_id', 'title', 'genres']
In [78]: movies = pd.read_table('movies.dat', sep='::', header=None,
.....: names=mnames)

In [79]: movies[:10]
Out[79]:
movie_id                              title                       genres
0         1 		      Toy Story (1995)  Animation|Children's|Comedy
1         2   			Jumanji (1995) Adventure|Children's|Fantasy
2         3            Grumpier Old Men (1995)               Comedy|Romance
3         4           Waiting to Exhale (1995)                 Comedy|Drama
4         5 Father of the Bride Part II (1995)                       Comedy
5         6                        Heat (1995)        Action|Crime|Thriller
6         7                     Sabrina (1995)               Comedy|Romance
7         8                Tom and Huck (1995)         Adventure|Children's
8         9                Sudden Death (1995)                       Action
9        10                   GoldenEye (1995)    Action|Adventure|Thriller
要为每个genre添加指标变量就需要做一些数据规整操作。首先，我们从数据集中抽取出不同的genre值（注意巧用set.union）：

[python]
view plaincopyprint?





In [80]: genre_iter = (set(x.split('|')) for x in movies.genres)

In [81]: genres = sorted(set.union(*genre_iter))

In [80]: genre_iter = (set(x.split('|')) for x in movies.genres)

In [81]: genres = sorted(set.union(*genre_iter))
现在，我们从一个全零DataFrame开始构建指标DataFrame：

[python]
view plaincopyprint?





In [82]: dummies = DataFrame(np.zeros((len(movies), len(genres))), columns=genres)

In [82]: dummies = DataFrame(np.zeros((len(movies), len(genres))), columns=genres)
接下来，迭代每一部电影并将dummies各行的项设置为1：

[python]
view plaincopyprint?





In [83]: for i, gen in enumerate(movies.genres):
.....: dummies.ix[i, gen.split('|')] = 1

In [83]: for i, gen in enumerate(movies.genres):
  .....: dummies.ix[i, gen.split('|')] = 1
然后，再将其与movies合并起来：

[python]
view plaincopyprint?





In [84]: movies_windic = movies.join(dummies.add_prefix('Genre_'))

In [85]: movies_windic.ix[0]
Out[85]:
movie_id 1
title Toy Story (1995)
genres Animation|Children's|Comedy
Genre_Action 0
Genre_Adventure 0
Genre_Animation 1
Genre_Children's 1
Genre_Comedy 1
Genre_Crime 0
Genre_Documentary 0
Genre_Drama 0
Genre_Fantasy 0
Genre_Film-Noir 0
Genre_Horror 0
Genre_Musical 0
Genre_Mystery 0
Genre_Romance 0
Genre_Sci-Fi 0
Genre_Thriller 0
Genre_War 0
Genre_Western 0
Name: 0

In [84]: movies_windic = movies.join(dummies.add_prefix('Genre_'))

In [85]: movies_windic.ix[0]
Out[85]:
movie_id		                  1
title 		           Toy Story (1995)
genres 		Animation|Children's|Comedy
Genre_Action                              0
Genre_Adventure                           0
Genre_Animation                           1
Genre_Children's                          1
Genre_Comedy                              1
Genre_Crime                               0
Genre_Documentary                         0
Genre_Drama                               0
Genre_Fantasy                             0
Genre_Film-Noir                           0
Genre_Horror                              0
Genre_Musical                             0
Genre_Mystery                             0
Genre_Romance                             0
Genre_Sci-Fi                              0
Genre_Thriller                            0
Genre_War                                 0
Genre_Western                             0
Name: 0

注意：

对于很大的数据，用这种方式构建多成员指标变量就会变得非常慢。肯定需要编写一个能够利用DataFrame内部机制的更低级的函数才行。

一个对统计应用有用的秘诀是：结合get_dummies和诸如cut之类的离散化函数。

[python]
view plaincopyprint?





In [86]: values = np.random.rand(10)

In [87]: values
Out[87]:
array([ 0.75603383, 0.90830844, 0.96588737, 0.17373658, 0.87592824,
0.75415641, 0.163486 , 0.23784062, 0.85564381, 0.58743194])

In [88]: bins = [0, 0.2, 0.4, 0.6, 0.8, 1]

In [89]: pd.get_dummies(pd.cut(values, bins))
Out[89]:
(0, 0.2] (0.2, 0.4] (0.4, 0.6] (0.6, 0.8] (0.8, 1]
0 0 0 0 1 0
1 0 0 0 0 1
2 0 0 0 0 1
3 1 0 0 0 0
4 0 0 0 0 1
5 0 0 0 1 0
6 1 0 0 0 0
7 0 1 0 0 0
8 0 0 0 0 1
9 0 0 1 0 0

[10 rows x 5 columns]
内容来自用户分享和网络整理，不保证内容的准确性，如有侵权内容，可联系管理员处理 点击这里给我发消息
标签： 
相关文章推荐
新的分享
章节导航