Numpy实现逻辑回归

实验环境

系统：Windows
CPU：i7-6820HK 超频至3.8G （Numpy默认使用CPU）
Python：3.8.6
Numpy：1.19.2

引入所需的库

import numpy as np  # 用于矩阵运算
from sklearn.metrics import f1_score, accuracy_score  # 用于模型评估
from time import time # 用于计算程序用时

导入训练数据和测试数据

"""
设置文件路径
"""
Train_Path = "train_data.txt"  # 训练集和标签
Test_Date_Path = "test_data.txt"  # 测试集
Test_Label_Path = "answer.txt"  # 测试标签
"""
将文件读取为numpy矩阵
"""
Train = np.genfromtxt(Train_Path, dtype='float', delimiter=',')
Test_Date = np.genfromtxt(Test_Date_Path, dtype='float', delimiter=',')
Test_Label = np.genfromtxt(Test_Label_Path, dtype='float', delimiter=',')
"""
分割训练集中的Date和Label
"""
Train_Date = np.array([x[:-1] for x in Train])
Train_Label = np.array([x[-1] for x in Train])

逻辑回归代码

class Lr:
    """
    Lr模型
    参数说明：
        alpha：学习速率
        max_iter：最大迭代次数
    """

    def __init__(self, alpha=0.01, max_iter=10000):
        """
        参数初始化
        """
        self.alpha = alpha
        self.max_iter = max_iter

    def sigmoid(self, z):
        """
        sigmoid函数运算
        """
        return 1/(1 + np.exp(-z))

    def gd(self, X, Y, W, b):
        """
        梯度下降
        参数说明：
            X：数据矩阵
            Y：标签矩阵
            W：权值
            b：偏置
        """
        for i in range(self.max_iter):  # 迭代更新
            z = np.dot(W, X.T) + b
            dz = self.sigmoid(z) - Y
            dw = 1/self.m * np.dot(dz, X)
            db = 1/self.m * np.sum(dz)
            W = W - self.alpha * dw  # 更新W
            b = b - self.alpha * db  # 更新b
        return W, b

    def fit(self, X, Y):
        """
        模型训练
        """
        m, nx = X.shape
        self.m = m
        self.nx = nx
        W = np.random.random((1, nx))*0.1  # 初始化W
        b = 0  # 初始化b
        self.W, self.b = self.gd(X, Y, W, b)  # 梯度下降更新

    def predict(self, testX):
        """
        模型预测
        """
        result = self.sigmoid(np.dot(self.W, testX.T) + self.b)  # Y=WX+b
        Y = []
        for i in list(result)[0]:
            if i > 0.5:  # 值大于0.5设为1
                Y.append(1)
            else:  # 否则设为0
                Y.append(0)
        return Y

评估函数

def score(y, ym):
    """
    评估函数
    """
    print("Accuracy：{:.2f}%".format(accuracy_score(y, ym)*100))  # 正确率
    print("F1 Score：{:.4f}".format(f1_score(y, ym)))  # F1 Score

数据测试

start = time()  # 计时开始
model = Lr()  # 初始化模型
model.fit(Train_Date, Train_Label)  # 模型训练
Ym = model.predict(Test_Date)  # 模型与与测
score(Test_Label, Ym)  # 模型评估
end = time()  # 计时结束
print("Time: {:.2f}s".format(end-start))  # 输出用时

Accuracy：85.00%
F1 Score：0.7475
Time: 43.30s