sklearn.preprocessing.LabelBinarizer

标签二值化：sklearn.preprocessing.LabelBinarizer(neg_label=0, pos_label=1,sparse_output=False)主要是将多类标签转化为二值标签，最终返回的是一个二值数组或稀疏矩阵

参数说明：

neg_label：输出消极标签值

pos_label：输出积极标签值

sparse_output：设置True时，以行压缩格式稀疏矩阵返回，否则返回数组

classes_属性：类标签的取值组成数组

①设置neg_label=2、pos_label=4，只能返回二值数组，理解neg_label、pos_label两标签值的含义

In [14]: from sklearn import preprocessing
    ...: lb = preprocessing.LabelBinarizer(neg_label=2,pos_label=4)
    ...: lb.fit([1, 2, 6, 4, 2])
    ...:
Out[14]: LabelBinarizer(neg_label=2, pos_label=4, sparse_output=False)

通过classes_属性获取训练集类标签值

In [15]: lb.classes_
Out[15]: array([1, 2, 4, 6])

通过transform方法转换成二值数组

In [16]: lb.transform([1, 2, 6, 4, 2])
Out[16]:
array([[4, 2, 2, 2],
       [2, 4, 2, 2],
       [2, 2, 2, 4],
       [2, 2, 4, 2],
       [2, 4, 2, 2]])

上述结果：1转换成 [4, 2, 2, 2]，由于标签有4个值，因此用4位表示，生成 [4, 2, 2, 2]的顺序，是按照数据与classes_属性获取的数组顺序匹配的，出现的位置为pos_label值，其他位置为neg_label值，通过下面测试集再验证该逻辑

In [17]: lb.transform([1, 6,3,5])
Out[17]:
array([[4, 2, 2, 2],
       [2, 2, 2, 4],
       [2, 2, 2, 2],
       [2, 2, 2, 2]])

neg_label值为非0数字时，不能设置sparse_output=True以稀疏矩阵输出，只能以数组形式输出

In [18]: from sklearn import preprocessing
    ...: lb = preprocessing.LabelBinarizer(neg_label=1,pos_label=2,sparse_outpu
    ...: t=True)
    ...: lb.fit([1, 2, 6, 4, 2])
    ...: lb.transform([1, 6])
    ...:
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-18-84f620fa6b11> in <module>()
      1 from sklearn import preprocessing
----> 2 lb = preprocessing.LabelBinarizer(neg_label=1,pos_label=2,sparse_output=
True)
      3 lb.fit([1, 2, 6, 4, 2])
      4 lb.transform([1, 6])

d:\softwore\python\lib\site-packages\sklearn\preprocessing\label.py in __init__(
self, neg_label, pos_label, sparse_output)
    275                              "zero pos_label and zero neg_label, got "
    276                              "pos_label={0} and neg_label={1}"
--> 277                              "".format(pos_label, neg_label))
    278
    279         self.neg_label = neg_label

ValueError: Sparse binarization is only supported with non zero pos_label and ze
ro neg_label, got pos_label=2 and neg_label=1

In [19]: from sklearn import preprocessing
    ...: lb = preprocessing.LabelBinarizer(neg_label=0,pos_label=2,sparse_outpu
    ...: t=True)
    ...: lb.fit([1, 2, 6, 4, 2])
    ...: lb.transform([1, 6])
    ...:
Out[19]:
<2x4 sparse matrix of type '<class 'numpy.int32'>'
        with 2 stored elements in Compressed Sparse Row format>

In [20]: lb.transform([1, 6]).toarray()
Out[20]:
array([[2, 0, 0, 0],
       [0, 0, 0, 2]])

对于二值标签训练，将返回单列数组

In [21]: from sklearn import preprocessing
    ...: lb = preprocessing.LabelBinarizer(neg_label=2,pos_label=4)
    ...: lb.fit_transform(['yes', 'no', 'no', 'yes'])
    ...:
Out[21]:
array([[4],
       [2],
       [2],
       [4]])

对于多标签分类

In [22]: from sklearn import preprocessing
    ...: import numpy as np
    ...: lb = preprocessing.LabelBinarizer(neg_label=2,pos_label=4)
    ...: lb.fit(np.array([[0, 1, 0, 1, 0],[1,1,0,0,1]]))
    ...: lb.classes_
    ...:
Out[22]: array([0, 1, 2, 3, 4])

In [23]: lb.transform([1, 6])
Out[23]:
array([[2, 4, 2, 2, 2],
       [2, 2, 2, 2, 2]])

In [24]: lb.y_type_
Out[24]: 'multilabel-indicator'