#### Model performance using prediction error : training and test dataset

x1<-runif(100,1,100)
x2<-rnorm(100,10,10)

y<-3+5*x1-7*x2+rnorm(100,0,3)

#make dataframe
mydata1<-as.data.frame(cbind(y,x1,x2))

#random partition for training and test
p<-0.7   # proportion of training set
n<-nrow(mydata1)


rep.n<-100  # number of repetition

mse<-rep(0,rep.n)
for (i in 1:rep.n){
        train.ind<-sample(n,as.integer(n*p))  #get train index randomly
	train.data<-mydata1[train.ind,]       #get train data
        test.data<-mydata1[-train.ind,]       #get test data

        lm.fit1<-lm(y~.,data=train.data)      #model fitting using train data
        lm.pred<-predict(lm.fit1,test.data)   #get fitted values for test data
        mse[i]<-mean((lm.pred-test.data$y)^2) #compute mse
	
}


#################################################################################
## Now Cross validation : 10 fold cv computation

x1<-runif(100,1,100)
x2<-rnorm(100,10,10)

y<-3+5*x1-7*x2+rnorm(100,0,3)

#make dataframe
mydata1<-as.data.frame(cbind(y,x1,x2))
K<-10  # 10-fold

myind<-sample(nrow(mydata1))
len1<-floor(length(myind)/K)
cv.err<-rep(0,K)
for (i in 1:K){
        startind<-(i-1)*len1+1   
        if (i==K){
                endind<-nrow(mydata1)} else
                {endind<-i*len1}
        testind<-myind[startind:endind]
        traindata<-mydata1[-testind,]
        testdata<-mydata1[testind,]

        lm.fit1<-lm(y~x1+x2,data=traindata)
        lm.pred<-predict(lm.fit1,testdata)
        cv.err[i]<-mean((lm.pred-testdata$y)^2)
}