分类指标计算 Precision、Recall、F-score、TPR、FPR、TNR、FNR、AUC、Accuracy

【机器学习】一文读懂分类算法常用评价指标

https://blog.csdn.net/deephub/article/details/113806330

诊断试验评价——真实性评价以及医疗图像这些指标的含义功能

import numpy as np
from sklearn import metrics
from sklearn.metrics import roc_auc_score
# import precisionplt


def calculate_TP(y, y_pred):
    tp = 0
    for i, j in zip(y, y_pred):
        if i == j == 1:
            tp += 1
    return tp
def calculate_TN(y, y_pred):
    tn = 0
    for i, j in zip(y, y_pred):
        if i == j == 0:
            tn += 1
    return tn
def calculate_FP(y, y_pred):
    fp = 0
    for i, j in zip(y, y_pred):
        if i == 0 and j == 1:
            fp += 1
    return fp
def calculate_FN(y, y_pred):
    fn = 0
    for i, j in zip(y, y_pred):
        if i == 1 and j == 0:
            fn += 1
    return fn



# acc 这两个值是一样的
def calculate_accuracy_sklearn(y, y_pred):
    return metrics.accuracy_score(y, y_pred)
# accu
def calculate_accuracy(y, y_pred):
    tp = calculate_TP(y, y_pred)
    tn = calculate_TN(y, y_pred)
    fp = calculate_FP(y, y_pred)
    fn = calculate_FN(y, y_pred)
    return (tp+tn) / (tp+tn+fp+fn)


# 精度 Precision
def calculate_precision(y, y_pred):
    tp = calculate_TP(y, y_pred)
    fp = calculate_FP(y, y_pred)
    return tp / (tp + fp)
# 召回率 Recall  也是 TPR  有多少被预测成正类（正类预测正确）
def calculate_recall(y, y_pred):
    tp = calculate_TP(y, y_pred)
    fn = calculate_FN(y, y_pred)
    return tp / (tp + fn)
def precision_recall_curve(y, y_pred):
    y_pred_class,precision,recall = [],[],[]
    thresholds = [0.1, 0.2, 0.3, 0.6, 0.65]
    for thresh in thresholds:
        for i in y_pred:     #y_pred holds prob value for class 1
            if i>=thresh: y_pred_class.append(1)
            else: y_pred_class.append(0)
        precision.append(calculate_precision(y, y_pred_class))
        recall.append(calculate_recall(y, y_pred_class))
    return recall, precisionplt.plot(recall, precision)

# F1分数 F1结合了Precision和Recall得分，得到一个单一的数字，可以帮助直接比较不同的模型。 可以将其视为P和R的谐波均值
def calculate_F1(y, y_pred):
    p = calculate_precision(y, y_pred)
    r = calculate_recall(y, y_pred)
    return 2*p*r / (p+r)


# AUC-ROC是用于二分类问题的非常常见的评估指标之一。 这是一条曲线，绘制在y轴的TPR（正确率）和x轴的FPR（错误率）之间，
# ROC曲线下的AUC（曲线下的面积）值越接近1，模型越好
def roc_auc(y, y_pred):
    return roc_auc_score(y, y_pred)


# 所有反类中，有多少被预测成正类（正类预测错误）
def FPR(y, y_pred):
    fp= calculate_FP(y, y_pred)
    tn= calculate_TN(y, y_pred)
    tp = calculate_TP(y, y_pred)
    fn = calculate_FN(y, y_pred)
    return fp / (fp + tn)

# 所有正类中，有多少被预测成反类（反类预测错误）
def FNR(y, y_pred):#
    tp = calculate_TP(y, y_pred)
    fn = calculate_FN(y, y_pred)
    return fn / (fn + tp)
# TNR= TN / (FP + TN)   , return tp / (tp + fp)
def TNR(y,y_pred):
    tn = calculate_TN(y, y_pred)
    fp = calculate_FP(y, y_pred)
    return tn / (fp + tn)

#TPR=TP/ (TP+ FN)  TPR即为敏感度（sensitivity）  也是recall
def TPR(y,y_pred):
    tp = calculate_TP(y, y_pred)
    fn = calculate_FN(y, y_pred)
    return tp / (fn + tp)

    
    


# Recall	F1_Score	precision	FPR假阳性率	FNR假阴性率
# AUC	AUC910%CI	ACC准确,TPR敏感,TNR特异度(TPR即为敏感度（sensitivity），TNR即为特异度（specificity）)
y=[]
y_pred=[]

a="/home/syy/code/PaddleClas/school_pre/pred3.txt"
f=open(a)
for line in f.readlines():
    line = line.strip().split()
    y.append(int(line[0]))
    y_pred.append(int(line[1]))

print(y)
print(y_pred)

Recall=calculate_recall(y, y_pred)
precision=calculate_precision(y, y_pred)
F1_Score=calculate_F1(y, y_pred)
FPR=FPR(y, y_pred)
FNR=FNR(y, y_pred)

auc = roc_auc(y, y_pred)
accuracy=calculate_accuracy_sklearn(y, y_pred)
TPR = TPR(y,y_pred)
TNR = TNR(y,y_pred)

print("Recall",round(Recall,4))
print("precision",round(precision,4))
print("F1_Score",round(F1_Score,4))
print("FPR",round(FPR,4))
print("FNR",round(FNR,4))

print("auc",round(auc,4))
print("accuracy",round(accuracy,4))
print("TPR",round(TPR,4))
print("TNR",round(TNR,4))