pandas.cut()方法介绍

  连续值经常需要离散化，或者分离成“箱子”进行分析，假设我们有某项研究中一组人群的数据，需要将其进行分组，放入离散的年龄框中：

ages = [2, 67, 40, 32, 4, 15, 82, 99, 26, 30, 11, 18, 53, 66]

  现在将这些年龄分为12以下，12_19，1961以及61以上，为实现这个，可以使用pandas中的cut，该语句含义是“根据ages列表中的最大值和最小值计算出等长的3个箱”：

cats = pd.cut(ages, bins=3)
cats

[(1.903, 34.333], (66.667, 99.0], (34.333, 66.667], (1.903, 34.333], (1.903, 34.333], ..., (1.903, 34.333], (1.903, 34.333], (1.903, 34.333], (34.333, 66.667], (34.333, 66.667]]
Length: 14
Categories (3, interval[float64, right]): [(1.903, 34.333] < (34.333, 66.667] < (66.667, 99.0]]

  分析输出数据可以看到，pandas返回的是一个特殊的Catagorical对象，输出描述了由pandas.cut计算出的箱。在这个箱中，第一部分列表和ages中的元素一一对应，分别表示ages中2属于(1.903, 34.333]，67属于(66.667, 99.0]，...；第二部分表示了被分割列表的长度；第三部分便是Catagorical对象的属性：平均分为3段、右侧为闭区间、每段的范围。
  我们可以把这个Catagorical对象当作一个表示箱名的字符串数组，它内部包含一个categories（类别）数组，它指定了不同的类别名称以及codes属性中的ages数据标签：

cats.codes

array([0, 2, 1, 0, 0, 0, 2, 2, 0, 0, 0, 0, 1, 1], dtype=int8)

  该列表表示，ages列表中2被分配在第0个箱，即(1.903, 34.333]，6被分配在第2个箱，即(66.667, 99.0]，...

cats.categories

IntervalIndex([(1.903, 34.333], (34.333, 66.667], (66.667, 99.0]], dtype='interval[float64, right]')

pd.value_counts(cats)

(1.903, 34.333]     8
(34.333, 66.667]    3
(66.667, 99.0]      3
dtype: int64

  注意pd.value_counts(cats)是对pandas.cut的结果中的箱数量的计数。
  在默认情况下，分配的箱为左开右闭（不包含左侧数值，包含右侧数值），可以通过传递right=False来改变哪一边是封闭的。

cats = pd.cut(ages, bins=3, right=False)
cats

[[2.0, 34.333), [66.667, 99.097), [34.333, 66.667), [2.0, 34.333), [2.0, 34.333), ..., [2.0, 34.333), [2.0, 34.333), [2.0, 34.333), [34.333, 66.667), [34.333, 66.667)]
Length: 14
Categories (3, interval[float64, left]): [[2.0, 34.333) < [34.333, 66.667) < [66.667, 99.097)]

cut()中的参数bins

  在上一部分中，我们传入了参数3，代表将列表均分为3个箱，这便意味着当给bins传入一个n时，cut()将为我们根据最大值和最小值把箱均分为n份。同理，还可以给bins传入一个列表，如[a, b, c]，此时cut()会按照(a_b]，(bc]分配，如下所示：

bins = [0, 12, 19, 61, 100]
cats = pd.cut(ages, bins=bins)
cats

[(0, 12], (61, 100], (19, 61], (19, 61], (0, 12], ..., (19, 61], (0, 12], (12, 19], (19, 61], (61, 100]]
Length: 14
Categories (4, interval[int64, right]): [(0, 12] < (12, 19] < (19, 61] < (61, 100]]

cut()中的参数labels与precision

  labels可以帮助我们自定义箱名，我们可以向labels选项传递一个列表或数组来传入自定义的箱名：

labels=['Teens', 'Youth', 'Adult', 'Older']
bins = [0, 12, 19, 61, 100]

cats = pd.cut(ages, bins=bins, labels=labels)
cats

['Teens', 'Older', 'Adult', 'Adult', 'Teens', ..., 'Adult', 'Teens', 'Youth', 'Adult', 'Older']
Length: 14
Categories (4, object): ['Teens' < 'Youth' < 'Adult' < 'Older']

  precision可以精确小数位如下所示：

cats = pd.cut(ages, bins=3, precision=2)
cats

[(1.9, 34.33], (66.67, 99.0], (34.33, 66.67], (1.9, 34.33], (1.9, 34.33], ..., (1.9, 34.33], (1.9, 34.33], (1.9, 34.33], (34.33, 66.67], (34.33, 66.67]]
Length: 14
Categories (3, interval[float64, right]): [(1.9, 34.33] < (34.33, 66.67] < (66.67, 99.0]]

pandas.qcut()方法介绍

  qcut()是一个与分箱密切相关的函数，它基于样本分位数进行分箱，即等频分箱（每个箱中个数一致），而cut()是等距分箱，直接举例：

data = pd.DataFrame(np.random.rand(12), columns=['number'])
data

	number
0	0.240733
1	0.143113
2	0.395232
3	0.845771
4	0.714535
5	0.561660
6	0.596787
7	0.921728
8	0.397136
9	0.422436
10	0.322760
11	0.588082

data['cut_group'] = pd.qcut(data['number'], q=4)  # 将data切分成4份
data

	number	        cut_group
0	0.240733	(0.142, 0.377]
1	0.143113	(0.142, 0.377]
2	0.395232	(0.377, 0.492]
3	0.845771	(0.626, 0.922]
4	0.714535	(0.626, 0.922]
5	0.561660	(0.492, 0.626]
6	0.596787	(0.492, 0.626]
7	0.921728	(0.626, 0.922]
8	0.397136	(0.377, 0.492]
9	0.422436	(0.377, 0.492]
10	0.322760	(0.142, 0.377]
11	0.588082	(0.492, 0.626]

pd.value_counts(data['cut_group'])

(0.142, 0.377]    3
(0.377, 0.492]    3
(0.492, 0.626]    3
(0.626, 0.922]    3
Name: cut_group, dtype: int64

  可以看到，qcut()把data切分成了每个箱中内容数量都一致的4份，与cut类似，我们也可以传入自定义的分位数，但切分出的箱便不再具有以上特性：

data['cut_group'] = pd.qcut(data['number'], q=[0, 0.3, 0.6, 1])
data

        number	        cut_group
0	0.240733	(0.142, 0.396]
1	0.143113	(0.142, 0.396]
2	0.395232	(0.142, 0.396]
3	0.845771	(0.578, 0.922]
4	0.714535	(0.578, 0.922]
5	0.561660	(0.396, 0.578]
6	0.596787	(0.578, 0.922]
7	0.921728	(0.578, 0.922]
8	0.397136	(0.396, 0.578]
9	0.422436	(0.396, 0.578]
10	0.322760	(0.142, 0.396]
11	0.588082	(0.578, 0.922]

pd.value_counts(data['cut_group'])

(0.578, 0.922]    5
(0.142, 0.396]    4
(0.396, 0.578]    3
Name: cut_group, dtype: int64