BGD+交叉熵实现二分类，数据和题目来自CS229监督学习的课后题

#coding=utf-8
#CS229 BGD+LR
import numpy as np
import pandas as pd
import math
def sigmoid(x):
    return 1.0/(1+np.exp(-x))

def CrossEntropy(w0,x,y):
    predict_error=np.sum(w0 * x.T,axis=1)
    y=np.squeeze(y)
    predict_error = predict_error-y # 得到预测误差向量
    J = sum(y*np.log(sigmoid(predict_error))+(1-y)*np.log(1-sigmoid(predict_error)))
    return J/len(y)

def BGD(x,y):
    stp = 0.001
    x = np.append(x, np.ones((x.shape[0], 1)), axis=1)  # 加入常数列
    # 以整个数据集作为训练数据 BGD
    w0 = np.ones(x.shape[1]).T
    max_iters = 15000
    iter_count = 0
    eps = 0.001
    w0 = np.array(w0).reshape(len(w0), 1)
    w0 = w0.T
    loss = float(30)
    y = np.squeeze(y)
    x = x.T
    w0 = sum(w0)
    while (iter_count < max_iters):
        oldJ = CrossEntropy(w0, x, y)
        predict_error = np.matmul(w0, x) - y#得到预测误差向量
        print("w0_old",w0)
        for i in range(x.shape[0]):
            w0[i] = w0[i] - stp * sum(predict_error * x[i, :])  # x[i,:]#所有x样本的第i个分量
        print("w0_new",w0)
        iter_count = iter_count + 1
        newJ = CrossEntropy(w0,x,y)
        print("oldJ,newJ",oldJ,newJ)
        if newJ==oldJ:
            print("已收敛")
            print('\n iteration=', iter_count, 'loss=', newJ)
            return w0
        print('\n iteration=', iter_count, 'loss=', newJ)
    return w0

if __name__ =='__main__':
    q1x = pd.read_table(r'D:\研究生课程材料\模式识别与机器学习\第六章选做作业\q1x.dat',
                        names=['x_1', 'x_2'], sep='  ', encoding='utf-8')
    # q2x=pd.read_table(r'D:\研究生课程材料\模式识别与机器学习\第六章选做作业\q2x.dat',
    # names=['x_1','x_2'],sep='  ',encoding='utf-8')
    q1y = pd.read_table(r'D:\研究生课程材料\模式识别与机器学习\第六章选做作业\q1y.dat',
                        names=['y'], sep='  ', encoding='utf-8')
    # q2y=pd.read_table(r'D:\研究生课程材料\模式识别与机器学习\第六章选做作业\q2y.dat',
    # names=['x_1','x_2'],sep='  ',encoding='utf-8')
    x = q1x.values
    y = q1y.values
    w=BGD(x,y)
    print(w)