感知机（代码实现）

方法一：损失函数L(ω,b)=−sum(yi * (ω * xi + b))

import numpy as np


def perceptron(xi, yi):
    res, leaning_rate, b, w = [], 0.1, 0, np.array([0.5, 0.5])  # res:存储算得的预测值 b:偏置项  w:x的权重
    # np.sign(np.dot(w, xi) + b)  # 激活函数，输出结果为{-1, +1}， -1为假， +1为真
    for i in range(len(yi)):
        if yi[i] * (np.dot(w, xi[i]) + b) <= 0:
            w += leaning_rate * yi[i] * xi[i]  # 对w进行更新
            b += leaning_rate * yi[i]          # 对b进行更新
    for j in range(len(xi)):
        y = np.sign(np.dot(w, xi[j]) + b)
        res.append(y)
    print(f"梯度下降后的权重：{w}, 梯度下降后的偏置项：{b}, 梯度下降后的输出值：{res}")


def AND():  # and运算
    x = np.array([[-1, -1], [-1, 1], [1, -1], [1, 1]])  # -1为假  1为真
    y = np.array([-1, -1, -1, 1])
    return [x, y]


perceptron(AND()[0], AND()[1])

运行结果：
梯度下降后的权重：[0.5 0.5], 梯度下降后的偏置项：-0.2, 梯度下降后的输出值：[-1.0, -1.0, -1.0, 1.0]

方法二：损失函数L(ω,b)=−0.5 * sum((ω * xi + b - yi)**2)

# 方法二
import numpy as np


def perceptron(xi, yi):
    count, res, leaning_rate, b, w = 0, [], 0.0000001, 0, np.array([0.5, 0.5])  # b:偏置项  w:x的权重
    # np.sign(np.dot(w, xi) + b)  # 激活函数，输出结果为{-1, +1}， -1为假， +1为真
    for i in range(len(yi)):
        while count < 1000000:
            J = (1 / 2) * (np.dot(w, xi[i]) + b - yi[i]) ** 2
            w += leaning_rate * (np.dot(w, xi[i]) + b - yi[i]) * xi[i]
            b += leaning_rate * (np.dot(w, xi[i]) + b - yi[i])
            if (J - (1 / 2) * (np.dot(w, xi[i]) + b - yi[i]) ** 2).all() == 0 and \
                    ((np.dot(w, xi[i]) + b - yi[i]) * xi[i]).all() == 0 \
                    and (np.dot(w, xi[i]) + b - yi[i]).all() == 0 and np.sign(np.dot(w, xi[i]) + b) * yi[i] == 1:
                break
            count += 1
    for j in range(len(xi)):
        y = np.sign(np.dot(w, xi[j]) + b)
        res.append(y)
    print(f"梯度下降后的权重：{w}, 梯度下降后的偏置项：{b}, 梯度下降后的输出值：{res}")


def AND():  # and运算
    x = np.array([[-1, -1], [-1, 1], [1, -1], [1, 1]])  # -1为假  1为真
    y = np.array([-1, -1, -1, 1])
    return [x, y]


perceptron(AND()[0], AND()[1])

运行结果：
梯度下降后的权重w：[0.38338042, 0.61661958]
梯度下降后的偏置项b：0.11661960682579162
梯度下降后的输出值：[-1.0, 1.0, -1.0, 1.0]----结果依旧存在误差