线性回归

题目

题目分析

考察线性回归的知识，可以采用最小二乘法、正规方程、以及建立线性回归模型

基础知识

线性回归模型

模型使用的准则是LMS准则，即最小均方误差准则

代码实现

未加正则化

import numpy as np
import matplotlib.pyplot as plt
#准备数据
##注意数据格式：矩阵的维数一定要看好
data_x=[[1,0],[1,1],[1,2],[1,3],[1,4],[1,5],[1,6],[1,7],[1,8],[1,9],[1,10],[1,11],[1,12],[1,13]] #进行数据归一化
data_y=[[2.0],[2.5],[2.9],[3.147],[4.515],[4.903],[5.365],[5.704],[6.853],[7.971],[8.561],[10],[11.280],[12.9]]

#计算当前权重下代价函数的值
def cost(xMat,weights,yMat):

    m,n=xMat.shape
    hypothesis=np.dot(xMat,weights)#预测值
    cost=(1/(2*m))*np.sum(np.square(hypothesis-yMat))

    return cost

#采用批处理梯度下降算法
def Training(data_x, data_y, alpha=0.01, maxepochs=3000, epslion=0.0000000001):

    xMat=np.mat(data_x)#构造矩阵
    yMat=np.mat(data_y)
    m,n=xMat.shape
    weights=np.zeros((n,1)) # 模型参数n*1的矩阵

    epochs_count=0
    loss_list=[]

    while epochs_count<maxepochs:
        loss=cost(xMat,weights,yMat)#上次的损失值
        hypothesis=np.dot(xMat,weights)#预测值
        error=hypothesis-yMat#（即实际值与预测值的差距）
        grad=(1/m)*np.dot(xMat.T,error)
        weights=weights-alpha*grad#参数更新，梯度下降
        loss_new=cost(xMat,weights,yMat)#weights更新后，损失的值
        if abs(loss_new-loss)<epslion:  #如果损失值变化小于一定值，表示收敛
            break
        loss_list.append(loss_new)
        epochs_count+=1
        print("第"+str(epochs_count)+"轮，weight0="+str(weights[0]+weights[1] * -2000)+"，weight1="+str(weights[1]))
    weights[0] = weights[0] + weights[1] * -2000# 参数还原
    x = 2014
    a = weights[0]
    b = weights[1]
    print("2014年房价：")
    print(a + x * b)

    plt.plot(loss_list)
    plt.show()

    return weights
weights_bgd = Training(data_x, data_y, alpha=0.01, maxepochs=200, epslion=0.000001)

运算结果

加入正则化

import numpy as np
import matplotlib.pyplot as plt
#准备数据
##注意数据格式：矩阵的维数一定要看好
data_x=[[1,0],[1,1],[1,2],[1,3],[1,4],[1,5],[1,6],[1,7],[1,8],[1,9],[1,10],[1,11],[1,12],[1,13]]
data_y=[[2.0],[2.5],[2.9],[3.147],[4.515],[4.903],[5.365],[5.704],[6.853],[7.971],[8.561],[10],[11.280],[12.9]]

#计算当前权重下代价函数的值
def cost(xMat,weights,yMat,Lambda):

    m,n=xMat.shape
    hypothesis=np.dot(xMat,weights)#预测值
   # cost=(1/(2*m))*np.sum(np.square(hypothesis-yMat))
    cost=(1/(2*m))*(np.sum(np.square(hypothesis-yMat))+Lambda*np.sum(np.square(weights)) )

    return cost

#采用批处理梯度下降算法
def Training(data_x, data_y, alpha=0.01,Lambda=0.01 ,maxepochs=3000, epslion=0.0000000001):

    xMat=np.mat(data_x)#构造矩阵
    yMat=np.mat(data_y)
    m,n=xMat.shape
    weights=np.zeros((n,1)) # 模型参数n*1的矩阵

    epochs_count=0
    loss_list=[]

    while epochs_count<maxepochs:
        loss=cost(xMat,weights,yMat,Lambda)#上次的损失值
        hypothesis=np.dot(xMat,weights)#预测值
        error=hypothesis-yMat#（即实际值与预测值的差距）
        grad=(1/m)*np.dot(xMat.T,error)
        weights=weights*(1-alpha*Lambda/m)-alpha*grad#参数更新，梯度下降
        loss_new=cost(xMat,weights,yMat,Lambda)#weights更新后，损失的值
        if abs(loss_new-loss)<epslion:
            break
        loss_list.append(loss_new)
        epochs_count+=1
        print("第"+str(epochs_count)+"轮，weight0="+str(weights[0]+weights[1] * -2000)+"，weight1="+str(weights[1]))
    weights[0] = weights[0] + weights[1] * -2000# 参数还原
    x = 2014
    a = weights[0]
    b = weights[1]
    print("2014年房价：")
    print(a + x * b)

    plt.plot(loss_list)
    plt.show()

    return weights
weights_bgd = Training(data_x, data_y, alpha=0.01,Lambda=0.01 ,maxepochs=200, epslion=0.000001)

运算结果

两者对比

根据代价函数下降曲线可以看出，当加入正则化时，曲线并没有发生很大的变化