支持向量机学习之3-SVR(回归)


ϵSVR

SVR回归中,基本思路和SVM中是一样的,在 ϵSVR [Vapnic,1995] 需要解决如下的优化问题。 

min  12||w||2+Ci=1l(ξi+ξi)
s.t.  yi(wTxi+b)<ϵ+ξi(wTxi+b)yi<ϵ+ξiξi,ξi0

细心的读者可能已经发现了与 CSVM 中的具有相似的地方,但又不太一样,那么如何理解上述公式呢? 
假设我们的训练数据集是 {(x1,y1),(x2,y2),...(xn,yl)}  
我们的目标是找到一个函数,比如线性函数 f(x)=wTx+b ,使得 
如果数据离回归函数的偏差 |yiwTxb|<ϵ| (下图非色区域),我们是能接受的,不需要付出任何代价(即不需要在代价函数中体现)。我们只关注偏差大于 ϵ 的代价。举个例子来说,就好比我们在换外币时,我们并不关注少量 ϵ 损失,这部分损失是汇率引起的合理损失。 
所以约束条件是保证更多多的数据点都在灰色范围内(拟合最佳的线性回归函数,使得更多的点落在我们接受的精度范围内),即 |yiwTxb|<ϵ| 。但是我们发现,还是会有一部分点,偏差比较大,落在灰色区域之外,所以类似SVM中使用的方法,引入松弛因子,采取软边界的方法,而且上下采取不同的松弛因子 ξi,ξi0 ,这样就不难得出约束条件为: 
s.t.  yi(wTxi+b)<ϵ+ξi(wTxi+b)yi<ϵ+ξiξi,ξi0

这里写图片描述  
如同SVM中一样的,在多数情况下转换为对偶问题更容易计算。同时还可以计算出 w b ,直接看文献1吧。 
这里写图片描述  
这里写图片描述  
这里写图片描述  
详细推导过程看文献1。

使用核函数的 ϵSVR

文献2 
这里写图片描述

νSVR

这里写图片描述 
Chang and Lin (2002) prove that ϵ -SVR with parameters  (C¯¯¯̄ ,ϵ) has the same solution as  ν -SVR with parameters (lC¯¯¯̄ ,ν) . 
其实两种SVR在满足一定条件下,具有相同的解。

优缺点分析

Scikit代码

#-*-coding:utf-8-*- 
import numpy as np  
from sklearn.svm import SVR  
import matplotlib.pyplot as plt  

###############################################################################  
# Generate sample data  
X = np.sort(5 * np.random.rand(40, 1), axis=0)  #产生40组数据,每组一个数据,axis=0决定按列排列,=1表示行排列  
y = np.sin(X).ravel()   #np.sin()输出的是列,和X对应,ravel表示转换成行  

###############################################################################  
# Add noise to targets  
y[::5] += 3 * (0.5 - np.random.rand(8))  

###############################################################################  
# Fit regression model  
svr_rbf10 = SVR(kernel='rbf',C=100, gamma=10.0)  
svr_rbf1 = SVR(kernel='rbf', C=100, gamma=0.1)  
svr_rbf1 = SVR(kernel='rbf', C=100, gamma=0.1)  
#svr_lin = SVR(kernel='linear', C=1e3)  
#svr_poly = SVR(kernel='poly', C=1e3, degree=3)  
y_rbf10 = svr_rbf10.fit(X, y).predict(X)  
y_rbf1 = svr_rbf1.fit(X, y).predict(X) 
#y_lin = svr_lin.fit(X, y).predict(X)  
#y_poly = svr_poly.fit(X, y).predict(X)  

###############################################################################  
# look at the results  
lw = 2 #line width  
plt.scatter(X, y, color='darkorange', label='data')  
plt.hold('on')  
plt.plot(X, y_rbf10, color='navy', lw=lw, label='RBF gamma=10.0')  
plt.plot(X, y_rbf1, color='c', lw=lw, label='RBF gamma=1.0')  
#plt.plot(X, y_lin, color='c', lw=lw, label='Linear model')  
#plt.plot(X, y_poly, color='cornflowerblue', lw=lw, label='Polynomial model')  
plt.xlabel('data')  
plt.ylabel('target')  
plt.title('Support Vector Regression')  
plt.legend()  
plt.show()
 
 
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40

RBF不同参数: 
这里写图片描述 
不同核函数: 
这里写图片描述

  • 5
    点赞
  • 32
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值