R语言中的svm实例

svm

WANGChang

2016年2月18日

n=150

p=2

sigma = 1

meanpos = 0

meanneg =3

npos = round(n/2)

nneg = n-npos

##生成数据中正负实例样本特征值

xpos = matrix(rnorm(npos*p,mean =meanpos,sd = sigma),npos,p)

xneg = matrix(rnorm(nneg*p,mean =meanneg,sd = sigma),npos,p)

x = rbind(xpos,xneg)

##生成类别标签

y = matrix(c(rep(1,time =npos),rep(-1,nneg)))

##rep()函数指的是后面重复前面次数

##观察数据发现数据线性可分

plot(x,pch = ifelse(y>0,1,2),xlab = "",ylab = "")

legend("topleft",c('Positive','negative'),pch = seq(2),text.col = seq(2))



##pch为点的形状类别，text.col指颜色类别

###在数据集中随机选取80%的数据作为学习样本，剩余20%的作为预测样本

ntrain <- round(n*0.8)

tindex <- sample(n,ntrain)

xtrain <- x[tindex,]

xtest <- x[-tindex,]

ytrain <- y[tindex]

ytest <- y[-tindex]

istrain <- rep(0 ,n)

istrain[tindex] = 1

#观察学习样本与预测样本的特征值与类别情况

plot(x,col = ifelse(y>0,1,2),pch = ifelse(istrain==1,1,2),xlab = "",ylab ="")

legend("topleft",c('PositiveTrain','PositiveTest','NegativeTrain','NegativeTest'),

pch = c(1,2,1,2), col = c(1,1,2,2))



#载入程序包，利用ksvm函数对样本数据进行学习

library(kernlab)

## Warning: package 'kernlab'was built under R version 3.2.3

svp <- ksvm(xtrain,ytrain,type = "C-svc",kernel = 'vanilladot',C = 100,scaled= c())

## Setting default kernel parameters

##观察学习后的svm模型

svp

## Support Vector Machineobject of class "ksvm"

##

## SV type: C-svc (classification)

## parameter : cost C = 100

##

## Linear (vanilla) kernel function.

##

## Number of Support Vectors : 7

##

## Objective Function Value : -507.8632

## Training error : 0.025

##详细观察模型各属性

##attributes(svp)

##观察学习样本中的支持向量样本

alpha(svp)

## [[1]]

## [1] 100.000000 5.543819 100.000000 48.882801 100.000000 100.000000

## [7] 54.426620

alphaindex(svp)

## [[1]]

## [1] 10 40 48 51 57 71 81

b(svp)

## [1] -2.682805

plot(svp,data =xtrain)



#可视化模型结果，颜色表示样本分类的确信度（集合间隔），加粗的样本点为支持向量

##使用学习后的模型进行预测

ypred = predict(svp,xtest)

table(ytest,ypred)

## ypred

## ytest -1 1

## -110 0

## 1 0 20

##计算结果的准确率

sum(ypred== ytest)/length(ytest)

## [1] 1

##计算分类函数值f(x)即：W(T)x+b

ypredscore = predict(svp,xtest,type="decision")

##载入ROCR包，用于刻画ROC曲线，计算预测结果分类的各项指标

library(ROCR)

## Warning: package 'ROCR'was built under R version 3.2.3

## Loading required package:gplots

## Warning: package 'gplots'was built under R version 3.2.3

##

## Attaching package: 'gplots'

##

## The following object is masked from'package:stats':

##

## lowess

pred <- prediction(ypredscore,ytest)

#刻画ROC曲线

perf <- performance(pred,measure = "tpr",x.measure = "fpr")

plot(perf)



#刻画查全率和查准率

perf <- performance(pred,measure = "prec",x.measure = "rec")

plot(perf)