准确率、精确率、召回率等指标计算 “Accuracy“, “Precision“, “Recall“, “Specificity“, “Npv“

import numpy as np
from prettytable import PrettyTable
class ConfusionMatrix(object):
    """
    matrix format:
     tp   fp
     fn   tn
    """

    def __init__(self, num_classes: int, labels: list):
        self.matrix = np.zeros((num_classes, num_classes))
        self.num_classes = num_classes
        self.labels = labels

    def update(self, preds:list, label:list):
        table1 = PrettyTable()
        self.matrix = np.zeros((self.num_classes, self.num_classes))
        for p, t in zip(preds, label):
            self.matrix[t, p] += 1
        table1.field_names = [""] + self.labels

        for i, row in enumerate(self.matrix):
            row_data = [self.labels[i]] + list(map(int, row))  # Convert to integer for better display
            table1.add_row(row_data)

        print("Confusion Matrix:")
        print(table1)


    def summary(self, whether_ci=False):
        self.sensitivity = []
        self.precision = []
        self.specificity = []
        self.accuracy = []
        self.npv = []
        ci_result = {}

        # calculate accuracy
        sum_TP = 0
        for i in range(self.num_classes):
            sum_TP += self.matrix[i, i]
        self.acc = round(sum_TP / np.sum(self.matrix), 3)
        print("the model accuracy is ", self.acc)

        # precision, recall, specificity
        table = PrettyTable()
        table.field_names = ["", "Accuracy", "Precision", "Recall", "Specificity", "Npv"]
        for i in range(self.num_classes):
            TP = self.matrix[i, i]  # 对角线
            FN = np.sum(self.matrix[i, :]) - TP  # 第i行的所有列
            FP = np.sum(self.matrix[:, i]) - TP  # 所有行的第i列
            TN = np.sum(self.matrix) - TP - FP - FN

            self.Precision = round(TP / (TP + FP), 3) if TP + FP != 0 else 0.
            self.Recall = round(TP / (TP + FN), 3) if TP + FN != 0 else 0.
            self.Specificity = round(TN / (TN + FP), 3) if TN + FP != 0 else 0.
            self.Npv = round(TN / (TN + FN), 3) if TN + FN != 0 else 0.

            table.add_row([self.labels[i], self.acc, self.Precision, self.Recall, self.Specificity, self.Npv])
            # print(self.Precision, self.Recall, self.Specificity)

            self.sensitivity.append(self.Recall)
            self.precision.append(self.Precision)
            self.specificity.append(self.Specificity)
            self.npv.append(self.Npv)
        avg_precision = round(np.mean(self.precision), 3)
        avg_recall = round(np.mean(self.sensitivity), 3)
        avg_specificity = round(np.mean(self.specificity), 3)
        avg_npv = round(np.mean(self.npv), 3)
        table.add_row(["Average", self.acc, avg_precision, avg_recall, avg_specificity, avg_npv])
        print(table)

1.首先初始化类：ac = ConfusionMatrix（numclass（分类的个数），labels（标签数）

）

示例输入：

ac = ConfusionMatrix（3，['类别1'，'类别2'，'类别3'）

2.调用方式：

#调用方式
ac.update([预测值]，[实际值])

3.输出样式：