Traceback (most recent call last): File "D:/pycharm/projects/Pythoneeee/projects/例子.py", line 53, in <module> theta_ml = trust_region_newton(theta_init, X, y) File "D:/pycharm/projects/Pythoneeee/projects/例子.py", line 37, in trust_region_newton p = minimize(lambda p: np.dot(grad, p) + 0.5*np.dot(p.T, np.dot(H, p)), File "D:\pycharm\projects\venv\lib\site-packages\scipy\optimize\_minimize.py", line 626, in minimize constraints = standardize_constraints(constraints, x0, meth) File "D:\pycharm\projects\venv\lib\site-packages\scipy\optimize\_minimize.py", line 987, in standardize_constraints constraints[i] = old_constraint_to_new(i, con) File "D:\pycharm\projects\venv\lib\site-packages\scipy\optimize\_constraints.py", line 549, in old_constraint_to_new raise ValueError("Unknown constraint type '%s'." % con['type']) ValueError: Unknown constraint type 'trust-region'.

感谢提供错误信息。根据错误提示，似乎是在使用Scipy库中的`minimize`函数时出现了问题。具体来说，错误信息显示未知的约束类型`trust-region`，这可能是因为Scipy库的版本不同导致的。为了解决这个问题，您可以尝试使用`minimize`函数的替代方案。例如，可以使用`scipy.optimize.fmin_tnc`函数来代替`minimize`函数，并指定约束类型为`None`。下面是修改后的代码：

import numpy as np
from scipy.optimize import fmin_tnc

# 定义目标函数
def negative_log_likelihood(theta, X, y):
    # 计算模型预测值
    y_pred = np.dot(X, theta)
    # 计算负对数似然函数
    neg_log_likelihood = -np.sum(y*np.log(y_pred) + (1-y)*np.log(1-y_pred))
    return neg_log_likelihood

# 定义计算梯度的函数
def gradient(theta, X, y):
    # 计算模型预测值
    y_pred = np.dot(X, theta)
    # 计算梯度
    grad = np.dot(X.T, y_pred - y)
    return grad

# 定义计算海森矩阵的函数
def hessian(theta, X, y):
    # 计算模型预测值
    y_pred = np.dot(X, theta)
    # 计算海森矩阵
    H = np.dot(X.T * y_pred * (1 - y_pred), X)
    return H

# 定义信赖域和局部线性近似方法
def trust_region_newton(theta_init, X, y, radius=0.1, max_iter=100):
    theta = theta_init
    for i in range(max_iter):
        # 计算梯度和海森矩阵
        grad = gradient(theta, X, y)
        H = hessian(theta, X, y)
        # 使用信赖域方法求解更新量
        p = fmin_tnc(func=lambda p: np.dot(grad, p) + 0.5*np.dot(p.T, np.dot(H, p)),
                     x0=np.zeros_like(theta), fprime=lambda p: np.dot(H, p) + grad,
                     args=(X, y), bounds=None)
        # 更新参数
        theta += p[0]
    return theta

# 生成随机数据集
n_samples, n_features = 1000, 10
X = np.random.normal(size=(n_samples, n_features))
y = np.random.binomial(1, 0.5, size=n_samples)

# 初始化参数
theta_init = np.zeros(n_features)

# 求解最大似然估计
theta_ml = trust_region_newton(theta_init, X, y)

print("最大似然估计的参数为：", theta_ml)

在这个修改后的代码中，我们使用了`scipy.optimize.fmin_tnc`函数来代替`minimize`函数，并将约束类型指定为`None`。同时，我们还对计算梯度和海森矩阵的函数进行了微调，以适配`fmin_tnc`函数的接口。