knn<-function(x0,x,y,k){
      x=as.matrix(x)
       p=dim(x)[2]	
       n=dim(x)[1]        
       dis=rep(0,n)
       for(i in 1:p){
         dis= (x0[i]-x[,i])^2+dis
       }
       ind=order(dis)[1:k]
       mean(y[ind])
}

bls<-function(x0,x,y){
  x0=as.matrix(x0)
  x = as.matrix(x)
  y = as.matrix(y)
  x0%*%solve(t(x)%*%x,t(x)%*%y)
}

# read the files
trainAll <- read.csv(file="train.csv",head=FALSE,sep=",")
testAll <- read.csv(file="test.csv",head=FALSE,sep=",")

#choose instances with code 2 and 3 only
train = rbind(trainAll[which(trainAll[,1,1] == 2),],trainAll[which(trainAll[,1,1] == 3),])
test = rbind(testAll[which(testAll[,1,1] == 2),],testAll[which(testAll[,1,1] == 3),])

# define possible k values we want to try
kValues = c(1,3,5,7,15)

# calculate and print out the error rates for knn and linear regression for each k value
for(kCounter in 1:(length(kValues)-4)){
	k = kValues[kCounter]

	# define variables to keep track of actual and estimated classes
	actualsTrain = train[,1]
	prdKnnTrain = matrix(0,length(actualsTrain),1)
	prdRegTrain = matrix(0,length(actualsTrain),1)


	# training error calculation
	for(i in 1:length(train[,1])){
		tmpAns1 = knn(train[i,2:length(train)],train[,2:length(train)],train[,1],k)
		if(tmpAns1<2.5){
		  prdKnnTrain[i]=2
		}
		else{
		  prdKnnTrain[i] = 3
		}
		tmpAns2 = bls(train[i,2:length(train)],train[,2:length(train)],train[,1])
		if(tmpAns2<2.5){
		  prdRegTrain[i]=2
		}
		else{
		  prdRegTrain[i] = 3
		}
	}
	knnTrainAccuracy = length(which(prdKnnTrain == actualsTrain))*100/length(actualsTrain)
	regTrainAccuracy = length(which(prdRegTrain == actualsTrain))*100/length(actualsTrain)
	cat("k = ",k, "; k-NN Train Accuracy: ", knnTrainAccuracy, " Percent; Regression Train Acccuracy: ", regTrainAccuracy, " Percent\n")
 	
	# now define variables for testing
	actualsTest = test[,1]
	prdKnnTest = matrix(0,length(actualsTest),1)
	prdRegTest = matrix(0,length(actualsTest),1)
	# testing error calculation
	for(i in 1:length(test[,1])){
		prdKnnTest[i] = knn(test[i,2:length(test)],train[,2:length(train)],train[,1],k)
		tmpAns = bls(test[i,2:length(train)],train[,2:length(train)],train[,1])

		if(tmpAns<2.5){
		  prdRegTest[i]=2
		}
		else{
		  prdRegTest[i] = 3
		}
	}
	knnTestAccuracy = length(which(prdKnnTest == actualsTest))*100/length(actualsTest)
	regTestAccuracy = length(which(prdRegTest == actualsTest))*100/length(actualsTest)
	cat("k = ",k, "; k-NN Test Accuracy: ", knnTestAccuracy, "Percent; Regression Test Acccuracy: ", regTrainAccuracy, "Percent\n")
}