LR(逻辑回归)numpy简单代码实现

import numpy as np


# sigmod函数
def sigmod(x):
    return 1/(1+np.exp(-x))


# 逻辑回归
def logistic_regression(feature, target):
    feature['b'] = 1  # 截距项
    n = feature.shape[1]
    m = feature.shape[0]
    weight = np.random.randn(n)  # 初始化权重
    fx = (weight*feature).sum(1)  # 拟合函数
    loss =-(target['target']*np.log(sigmod(fx))+(1-target['target'])*np.log(1-sigmod(fx))).sum()/m  # 损失函数
    rate=0.01  # 学习率
    while True:  # 梯度下降停止条件
        if loss<0.3:
            break
        else:
            grad=(np.array(sigmod(fx)-target['target']).reshape(m,1)*feature).sum()/n  # 梯度计算
            weight=weight+rate*np.array(grad)  # 权重更新
        fx = (weight * feature).sum(1)  # 重新拟合函数
        loss = -(target['target'] * np.log(sigmod(fx)) + (1 - target['target']) * np.log(
            1 - sigmod(fx))).sum() / m  # 重新计算损失
    ans=[]
    for col in sigmod(fx).items():
        if col[1]>=0.5:
            ans.append(1)
        else:
            ans.append(0)
    return ans