【机器学习】周志华西瓜书3.3题——Python代码实现，牛顿法与梯度下降法

# -*-coding:utf-8-*-

"""
牛顿法实现对率回归(Logistic Regression)
来源: '机器学习, 周志华'
模型: P69 problem 3.3
数据集: P89 watermelon_3.0a (watermelon_3.0a.npy)
"""
import matplotlib.pyplot as plt
import numpy as np
import xlrd


def sigmoid(x):
    """
    Sigmoid function.
    Input:
        x:np.array
    Return:
        y: the same shape with x
    """
    y = 1.0 / (1 + np.exp(-x))
    return y


def newton(X, y):
    """
    Input:
        X: np.array with shape [N, 3]. Input.
        y: np.array with shape [N, 1]. Label.
    Return:
        beta: np.array with shape [1, 3]. Optimal params with newton method
    """
    N = X.shape[0]
    # initialization
    beta = np.ones((1, 3))

    # shape [N, 1]
    z = X.dot(beta.T)

    # log-likehood
    old_l = 0
    new_l = np.sum(-y * z + np.log(1 + np.exp(z)))  # 计算对数似然的代价函数值
    iters = 0
    while (np.abs(old_l - new_l) > 1e-6):
        # shape [N, 1]
        p1 = np.exp(z) / (1 + np.exp(z))

        # shape [N, N]
        p = np.diag((p1 * (