【python|ML】k-fold/leave-one-out 方法在对率回归实现（西瓜书习题3.4，数据UCI-iris）

# 经验：梯度下降要让目标确实减小（或者增大），此程序tempsum需要减小，故要着重检查！！

import numpy as np


def dlhg(x, d):
    x = np.c_[x, np.ones(d.shape[0])]
    w = np.random.randn(x[0].shape[0])
    miu = 0.01
    temp_sum_old = np.inf
    count = 0
    while True:
        tempsum = 0
        dw = 0
        for i in range(d.shape[0]):
            pdt = np.sum(w * x[i])
            tempsum = tempsum + np.log(1 + np.exp(pdt)) - d[i] * pdt
            dw = dw + (np.exp(pdt) / (1 + np.exp(pdt)) - d[i]) * x[i]
        if np.abs(tempsum - temp_sum_old) < 0.01:
            break
        w = w - miu * dw
        temp_sum_old = tempsum
        count = count + 1
        #print(tempsum)
    #print(count)
    #for i in range(d.shape[0]):
      #  # i 为第i个样本
     #   if np.sum(w * x[i]) > 0:
       #     print(1)
        #else:
         #   print(0)
    #print(w)
    ret