机器学习习题3.3

最新推荐文章于 2023-07-30 15:26:19 发布

原创最新推荐文章于 2023-07-30 15:26:19 发布 · 1.1k 阅读

3 ·

CC 4.0 BY-SA版权

机器学习专栏收录该内容

7 篇文章

订阅专栏

西瓜3.0a数据集如下：

密度	含糖量	好瓜
0.6792	0.46	1
0.774	0.376	1
0.634	0.264	1
0.608	0.318	1
0.556	0.215	1
0.403	0.237	1
0.481	0.149	1
0.437	0.211	1
0.666	0.091	0
0.243	0.267	0
0.245	0.057	0
0.343	0.099	0
0.639	0.161	0
0.657	0.198	0
0.36	0.37	0
0.593	0.042	0
0.719	0.103	0

由‘机器学习第三章’中的推导公式，将其转换为代码如下

将 $p_{1}(x^{'}_{i};\beta )$ 转化为如下代码：

def P1(x_data,beta):
    per1=np.exp(np.dot(beta.T,x_data))/(1+np.exp(np.dot(beta.T,x_data)))
    return per1

将 $(\frac{\partial^2 l(\beta )}{\partial \beta \partial \beta ^{T}})^{-1}\frac{\partial l(\beta )}{\partial \beta}$ 转化如下：

def derivative(x_data,y_data,beta):
    sum1 = 0
    sum2 = 0
    for x,y in zip(x_data,y_data):
        xi = np.append(x,1)
        sum1 -=xi*(y-P1(xi,beta))#(3.30)
        sum2 +=xi*xi.T*P1(xi,beta)*(1-P1(xi,beta)) #(3.31)
        #print(sum1/sum2)
    return sum1/sum2

将最后的牛顿算法进行转换

def Newton_method(x_data,y_data,iterations,alpha=1):
    print('start')
    w = np.zeros(x_data.shape[1])
    b = 0
    old_beta=beta = np.append(w,b)
    
    for i in range(iterations):
        beta = old_beta- alpha*derivative(x_data,y_data,old_beta) #(3.29)
        #beta值判定
        if converge(old_beta,beta):
            print(i)
            break
            
        old_beta = beta
    print('end')
    return beta

中间添加了beta值判定，代码如下：当误差降到0.001后停止

def converge(old_beta,beta):
    sub = beta - old_beta
    for i,j in zip(sub,beta):
        if j != 0 and abs(i/j) >= 0.001:
            return False
    return True

全部工程代码如下：

# -*- coding: utf-8 -*-
"""
Created on Fri Sep 21 21:58:19 2018

@author: Zch
"""

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

#load dataset
df_train = pd.read_csv('wm3.0a.csv')
X = np.array(df_train[['密度','含糖量']])
y = np.array(df_train['好瓜'])

def Newton_method(x_data,y_data,iterations,alpha=1):
    #(3.29)P59
    print('start')
    w = np.zeros(x_data.shape[1])
    b = 0
    old_beta=beta = np.append(w,b)
    
    for i in range(iterations):
        beta = old_beta- alpha*derivative(x_data,y_data,old_beta) #(3.29)
        
        if converge(old_beta,beta):
            print(i)
            break
            
        old_beta = beta
    print('end')
    return beta

def converge(old_beta,beta):
    sub = beta - old_beta
    for i,j in zip(sub,beta):
        if j != 0 and abs(i/j) >= 0.001:
            return False
    return True

def derivative(x_data,y_data,beta):
    #(3.30)(3.31) P60
    sum1 = 0
    sum2 = 0
    for x,y in zip(x_data,y_data):
        xi = np.append(x,1)
        sum1 -=xi*(y-P1(xi,beta))#(3.30)
        sum2 +=xi*xi.T*P1(xi,beta)*(1-P1(xi,beta)) #(3.31)
        #print(sum1/sum2)
    return sum1/sum2
        
def P1(x_data,beta):
    #(3.23) P59
    per1=np.exp(np.dot(beta.T,x_data))/(1+np.exp(np.dot(beta.T,x_data)))
    return per1

def predic(x_data,beta):
    y = 1/(1+np.exp(-np.dot(beta.T,x_data)))
    if y < 0.5:
        return False
    elif y > 0.5:
        return True

if __name__ == '__main__':
    beta = Newton_method(X,y,1000,0.05)
    print(beta)
#plt.scatter(x,y)

由于没有找到相关测试集，所以还没进行调优，路过的大佬，有相关测试集的，麻烦提供一下