编程语言
首页 > 编程语言> > 分类算法-逻辑回归与二分类

分类算法-逻辑回归与二分类

作者:互联网

分类算法-逻辑回归与二分类

1、逻辑回归的应用场景

看到上面的例子,我们可以发现其中的特点,那就是都属于两个类别之间的判断。逻辑回归就是解决二分类问题的利器

2、 逻辑回归的原理

2.1 输入

逻辑回归的输入就是一个线性回归的结果,然后把这个结果映射到0-1之间。

2.2 激活函数

逻辑回归最终的分类是通过属于某个类别的概率值来判断是否属于某个类别,并且这个类别默认标记为1(正例),另外的一个类别会标记为0(反例)。(方便损失计算)

输出结果解释(重要):假设有两个类别A,B,并且假设我们的概率值为属于A(1)这个类别的概率值。现在有一个样本的输入到逻辑回归输出结果0.6,那么这个概率值超过0.5,意味着我们训练或者预测的结果就是A(1)类别。那么反之,如果得出结果为0.3那么,训练或者预测结果就为B(0)类别。

2.3 损失以及优化

2.3.1 对数似然损失

逻辑回归的损失,称之为对数似然损失,公式如下:

​ 我们已经知道,log(P), P值越大,结果越小,所以我们可以对着这个损失的式子去分析

2.3.2 优化

同样使用梯度下降优化算法,去减少损失函数的值。这样去更新逻辑回归前面对应算法的权重参数,提升原本属于1类别的概率,降低原本是0类别的概率。

3、逻辑回归API

默认将类别数量少的当做正例

LogisticRegression方法相当于 SGDClassifier(loss="log", penalty=" "),SGDClassifier实现了一个普通的随机梯度下降学习,也支持平均随机梯度下降法(ASGD),可以通过设置average=True。而使用LogisticRegression(实现了SAG)

4、 案例:癌症分类预测-良/恶性乳腺癌肿瘤预测

原始数据的下载地址:https://archive.ics.uci.edu/ml/machine-learning-databases/

数据描述

(1)699条样本,共11列数据,第一列用语检索的id,后9列分别是与肿瘤

相关的医学特征,最后一列表示肿瘤类型的数值。

(2)包含16个缺失值,用”?”标出。

4.1 分析

4.2 代码

def logisticregression():
    """
    逻辑回归进行癌症预测
    :return: None
    """
    # 1、读取数据,处理缺失值以及标准化
    column_name = ['Sample code number', 'Clump Thickness', 'Uniformity of Cell Size', 'Uniformity of Cell Shape',
                   'Marginal Adhesion', 'Single Epithelial Cell Size', 'Bare Nuclei', 'Bland Chromatin',
                   'Normal Nucleoli', 'Mitoses', 'Class']

    data = pd.read_csv("https://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/breast-cancer-wisconsin.data",
                       names=column_name)

    # 删除缺失值
    data = data.replace(to_replace='?', value=np.nan)

    data = data.dropna()

    # 取出特征值
    x = data[column_name[1:10]]

    y = data[column_name[10]]

    # 分割数据集
    x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.3)

    # 进行标准化
    std = StandardScaler()

    x_train = std.fit_transform(x_train)

    x_test = std.transform(x_test)

    # 使用逻辑回归
    lr = LogisticRegression()

    lr.fit(x_train, y_train)

    print("得出来的权重:", lr.coef_)

    # 预测类别
    print("预测的类别:", lr.predict(x_test))

    # 得出准确率
    print("预测的准确率:", lr.score(x_test, y_test))
    return None

5、分类的评估方法

5.1 精确率与召回率

5.1.1混淆矩阵

在分类任务下,预测结果(Predicted Condition)与正确标记(True Condition)之间存在四种不同的组合,构成混淆矩阵(适用于多分类)

![image-20220404225416011]

5.1.2 精确率(Precision)与召回率(Recall)

5.1.3 分类评估报告API

print("精确率和召回率为:", classification_report(y_test, lr.predict(x_test), labels=[2, 4], target_names=['良性', '恶性']))

假设这样一个情况,如果99个样本癌症,1个样本非癌症,不管怎样我全都预测正例(默认癌症为正例),准确率就为99%但是这样效果并不好,这就是样本不均衡下的评估问题

问题:如何衡量样本不均衡下的评估?

5.2 ROC曲线与AUC指标

ROC全称是“受试者工作特征”(Receiver Operating Characteristic)。ROC曲线的面积就是AUC(Area Under the Curve)。AUC用于衡量“二分类问题“机器学习算法性能(泛化能力)。

5.2.1 知道TPR与FPR

5.2.2 ROC曲线

5.2.3 AUC指标

最终AUC的范围在[0.5, 1]之间,并且越接近1越好

5.2.4 AUC计算API

# 0.5~1之间,越接近于1约好
y_test = np.where(y_test > 2.5, 1, 0)

print("AUC指标:", roc_auc_score(y_test, lr.predict(x_test)))
案例(逻辑回归进行癌症预测)
def logisticregression():
    """
    逻辑回归进行癌症预测
    :return: None
    """
    # 1、读取数据,处理缺失值以及标准化
    column_name = ['Sample code number', 'Clump Thickness', 'Uniformity of Cell Size', 'Uniformity of Cell Shape',
                   'Marginal Adhesion', 'Single Epithelial Cell Size', 'Bare Nuclei', 'Bland Chromatin',
                   'Normal Nucleoli', 'Mitoses', 'Class']

    data = pd.read_csv("https://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/breast-cancer-wisconsin.data",
                       names=column_name)

    # 删除缺失值
    data = data.replace(to_replace='?', value=np.nan)

    data = data.dropna()

    # 取出特征值
    x = data[column_name[1:10]]

    y = data[column_name[10]]

    # 分割数据集
    x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.3)

    # 进行标准化
    std = StandardScaler()

    x_train = std.fit_transform(x_train)

    x_test = std.transform(x_test)

    # 使用逻辑回归
    lr = LogisticRegression()

    lr.fit(x_train, y_train)

    print("得出来的权重:", lr.coef_)

    # 预测类别
    print("预测的类别:", lr.predict(x_test))

    # 得出准确率
    print("预测的准确率:", lr.score(x_test, y_test))

    print("精确率和召回率为:", classification_report(y_test, lr.predict(x_test), labels=[2, 4], target_names=['良性', '恶性']))

    # 0.5~1之间,越接近于1约好
    y_test = np.where(y_test > 2.5, 1, 0)

    print("AUC指标:", roc_auc_score(y_test, lr.predict(x_test)))
    return None

返回结果:

得出来的权重: [[1.18088011 0.17960576 0.64716029 0.84256205 0.13629304 1.40238555
  1.05954948 0.68190687 0.86153865]]
预测的类别: [4 4 2 2 4 2 4 2 2 2 2 2 2 4 4 4 2 4 2 4 2 2 4 2 2 2 2 2 2 4 2 4 2 2 2 2 2
 2 4 2 4 2 2 2 4 2 2 2 2 2 2 4 4 2 2 2 2 2 2 2 4 2 4 2 2 2 2 2 4 2 4 2 2 2
 2 4 4 2 4 2 2 2 4 2 2 4 4 2 2 2 2 4 2 2 2 2 4 2 2 2 2 2 2 2 2 2 2 4 4 2 2
 2 4 2 2 2 4 2 2 2 4 2 2 2 2 4 2 4 2 2 2 4 2 2 2 4 2 2 2 2 2 4 2 2 2 2 2 2
 2 4 4 2 2 4 2 2 2 2 2 4 4 4 4 2 2 2 2 2 2 2 2 2 2 2 4 4 2 2 2 4 2 2 4 4 2
 4 4 4 4 4 4 4 4 2 2 2 2 2 2 2 4 2 4 4 4]
预测的准确率: 0.9707317073170731
精确率和召回率为:               precision    recall  f1-score   support

          良性       0.98      0.98      0.98       142
          恶性       0.95      0.95      0.95        63

    accuracy                           0.97       205
   macro avg       0.97      0.97      0.97       205
weighted avg       0.97      0.97      0.97       205

AUC指标: 0.965627095908786

5.2.5 小结


注:参考了黑马程序员的资料

标签:AUC,分类,逻辑,类别,算法,train,lr,test,data
来源: https://www.cnblogs.com/rainbow-1/p/16101426.html