python计算AUC、Accuracy、Recall、Precision和PPV

 

import numpy as np
import cv2
import seaborn as sns
import matplotlib.pyplot as plt
# 也可以直接采用这个进行计算
from sklearn.metrics import confusion_matrix, accuracy_score, f1_score, roc_auc_score, recall_score, precision_score
# 加载confusion图,并显示对应的图像表示混淆矩阵的信息
img = cv2.imread("confusion_figure.jpg")
plt.imshow(img)

def calculate_metrics(gt, pred):
    """
    :param gt: 数据的真实标签，一般对应二分类的整数形式，例如:y=[1,0,1,0,1]
    :param pred: 输入数据的预测值，因为计算混淆矩阵的时候，内容必须是整数，所以对于float的值，应该先调整为整数
    :return: 返回相应的评估指标的值
    """
    """
        confusion_matrix(y_true,y_pred,labels,sample_weight,normalize)
        y_true:真实标签；
        y_pred:预测概率转化为标签；
        labels:用于标签重新排序或选择标签子集；
        sample_weight:样本权重；
        normalize:在真实（行）、预测（列）条件或所有总体上标准化混淆矩阵；
    """
    print("starting!!!-----------------------------------------------")
    sns.set()
    fig, (ax1, ax2) = plt.subplots(figsize=(10, 8), nrows=2)
    confusion = confusion_matrix(gt, pred)
    # 打印具体的混淆矩阵的每个部分的值
    print(confusion)
    # 从左到右依次表示TN、FP、FN、TP
    print(confusion.ravel())
    # 绘制混淆矩阵的图
    sns.heatmap(confusion,annot=True)
    ax2.set_title('sns_heatmap_confusion_matrix')
    ax2.set_xlabel('y_pred')
    ax2.set_ylabel('y_true')
    fig.savefig('sns_heatmap_confusion_matrix.jpg', bbox_inches='tight')
    # 混淆矩阵的每个值的表示
    TP = confusion[1, 1]
    TN = confusion[0, 0]
    FP = confusion[0, 1]
    FN = confusion[1, 0]
    # 通过混淆矩阵计算每个评估指标的值
    print('AUC:',roc_auc_score(gt, pred))
    print('Accuracy:', (TP + TN) / float(TP + TN + FP + FN))
    print('Sensitivity:', TP / float(TP + FN))
    print('Specificity:', TN / float(TN + FP))
    print('PPV:',TP / float(TP + FP))
    print('Recall:',TP / float(TP + FN))
    print('Precision:',TP / float(TP + FP))
    # 用于计算F1-score = 2*recall*precision/recall+precision,这个情况是比较多的
    P = TP / float(TP + FP)
    R = TP / float(TP + FN)
    print('F1-score:',(2*P*R)/(P+R))
    print('True Positive Rate:',round(TP / float(TP + FN)))
    print('False Positive Rate:',FP / float(FP + TN))
    print('Ending!!!------------------------------------------------------')

    # 采用sklearn提供的函数验证,用于对比混淆矩阵方法与这个方法的区别
    print("the result of sklearn package")
    auc = roc_auc_score(gt,pred)
    print("sklearn auc:",auc)
    accuracy = accuracy_score(gt,pred)
    print("sklearn accuracy:",accuracy)
    recal = recall_score(gt,pred)
    precision = precision_score(gt,pred)
    print("sklearn recall:{},precision:{}".format(recal,precision))
    print("sklearn F1-score:{}".format((2*recal*precision)/(recal+precision)))

# 输入真实的样本和对应的标签
y = [1,0,1,1,0]
pred = [0,1,1,0,0]
# 实例化
calculate_metrics(y, pred)