利用ROC和AUC度量分类器性能

淡淡的烟草味﹌ 2022-03-21 19:44 195阅读 0赞

关于ROC及AUC的介绍，可以参考：[https://blog.csdn.net/qq\_41080850/article/details/86560645][https_blog.csdn.net_qq_41080850_article_details_86560645]

本文主要讨论用Python3实现ROC的绘制。另：以下代码均运行于Jupyter Notebook中。

代码实现：

# 读入岩石-水雷数据集：
    data = pd.read_csv('rock_mine.csv',header=None,prefix='V')
    data.head()

data表结构如下所示：

![20190204211931917.PNG][]

![20190204212444748.PNG][]

# 导入要用到的相关包
    import numpy as np
    import random
    from sklearn import datasets, linear_model
    from sklearn.metrics import roc_curve, auc
    import pylab as pl
    
    
    # 定义混淆矩阵函数：
    def confusionMatrix(predicted, actual, threshold):
        if len(predicted) != len(actual): return -1
        tp = 0.0
        fp = 0.0
        tn = 0.0
        fn = 0.0
        for i in range(len(actual)):
            if actual[i] > 0.5: # 实际值是正例
                if predicted[i] > threshold:
                    tp += 1.0 # 预测值也是正例(真正例)
                else:
                    fn += 1.0 # 预测值是反例(假反例)
            else:              # 实际值是反例
                if predicted[i] < threshold:
                    tn += 1.0 # 预测值是反例(真反例)
                else:
                    fp += 1.0 # 预测值是正例(假正例)
        rtn = [tp, fn, fp, tn]
        return rtn
    
    
    # 提取特征集和标签集
    xList = list(data.iloc[:,0:-1].values)
    labels = data['V60'].map({'M':1,'R':0}).tolist()
    
    
    # 采用留出法将原始数据集划分成训练集(2/3)和测试集(1/3)  //准确的说，此处的测试集应该叫做验证集
    indices = range(len(xList))
    xListTest = [xList[i] for i in indices if i%3 == 0 ]
    xListTrain = [xList[i] for i in indices if i%3 != 0 ]
    labelsTest = [labels[i] for i in indices if i%3 == 0]
    labelsTrain = [labels[i] for i in indices if i%3 != 0]
    
    
    # 将训练集和测试集分别转化成与scikit-learn中linear model的输入相匹配的数组对象
    xTrain = np.array(xListTrain)
    yTrain = np.array(labelsTrain)
    xTest = np.array(xListTest)
    yTest = np.array(labelsTest)
    
    
    # 训练线性回归模型
    rocksVMinesModel = linear_model.LinearRegression()
    rocksVMinesModel.fit(xTrain,yTrain)
    
    # 利用训练好的模型计算在训练集上的预测值：
    trainingPredictions = rocksVMinesModel.predict(xTrain)
    
    
    # 利用混淆函数计算训练集上的混淆矩阵：
    confusionMatTrain = confusionMatrix(trainingPredictions, yTrain, 0.5)
    tp_train = confusionMatTrain[0]
    fn_train = confusionMatTrain[1] 
    fp_train = confusionMatTrain[2] 
    tn_train = confusionMatTrain[3]
    
    # 打印出tp_train、fn_train、fp_train和tn_train的值
    print("tp_train = " + str(tp_train) + "\tfn_train = " + str(fn_train) + "\n" + \
          "fp_train = " + str(fp_train) + "\ttn_train = " + str(tn_train) + '\n')
    # 打印结果为：
    # tp_train = 68.0	fn_train = 6.0
    # fp_train = 7.0	tn_train = 57.0
    
    
    # 利用训练好的模型计算在测试集上的测试值：
    testPredictions = rocksVMinesModel.predict(xTest)
    
    # 利用混淆函数计算测试集上的混淆矩阵：
    conMatTest = confusionMatrix(testPredictions, yTest, 0.5)
    tp_test = conMatTest[0]
    fn_test = conMatTest[1]
    fp_test = conMatTest[2]
    tn_test = conMatTest[3]
    
    # 打印出tp_test,fn_test,fp_test和tn_test的值
    print("tp_test = " + str(tp_test) + "\tfn_test = " + str(fn_test) + "\n" + \
          "fp_test = " + str(fp_test) + "\ttn_test = " + str(tn_test) + '\n')
    # 打印结果为：
    # tp_test = 28.0	fn_test = 9.0
    # fp_test = 9.0	        tn_test = 24.0

# 利用roc_curve绘制训练集上的ROC
    fpr_train, tpr_train, thresholds_train = roc_curve(yTrain,trainingPredictions)
    roc_auc_train = auc(fpr_train, tpr_train)
    print( 'AUC for in-sample ROC curve: %f' % roc_auc_train)
    
    pl.clf()  # 清除当前figure
    pl.plot(fpr_train, tpr_train, label='ROC curve (area = %0.2f)' % roc_auc_train)
    pl.plot([0, 1], [0, 1], 'k--')   # 绘制随机猜测分类器对应的ROC
    pl.xlim([0.0, 1.0])
    pl.ylim([0.0, 1.0])
    pl.xlabel('False Positive Rate')
    pl.ylabel('True Positive Rate')
    pl.title('In sample ROC rocks versus mines')
    pl.legend(loc="best")

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70][]

# 利用roc_curve绘制测试集上的ROC
    fpr_test, tpr_test, thresholds_test = roc_curve(yTest,testPredictions)
    roc_auc_test = auc(fpr_test, tpr_test)
    print( 'AUC for out-of-sample ROC curve: %f' % roc_auc_test)
    
    pl.clf()
    pl.plot(fpr_test, tpr_test, label='ROC curve (area = %0.2f)' % roc_auc_test)
    pl.plot([0, 1], [0, 1], 'k--')
    pl.xlim([0.0, 1.0])
    pl.ylim([0.0, 1.0])
    pl.xlabel('False Positive Rate')
    pl.ylabel('True Positive Rate')
    pl.title('Out-of-sample ROC rocks versus mines')
    pl.legend(loc="best")

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 1][]

参考：《Python机器学习——预测分析核心算法》Michael Bowles著

[https_blog.csdn.net_qq_41080850_article_details_86560645]: https://blog.csdn.net/qq_41080850/article/details/86560645
[20190204211931917.PNG]: /images/20220321/4d43e0a07ef84bd38af820fa81ca6023.png
[20190204212444748.PNG]: /images/20220321/14f6a98e2d8444869149727c3d8148ad.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70]: /images/20220321/2c875bf8f17a40b2bf4eb53b00594532.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 1]: /images/20220321/4e8a5c9c03e34b158458abcdb418483a.png