Data <- na.omit(read.table("C:/Documents and Settings/steeleb/My Documents/Math 542/BreastCancer.txt",header=FALSE))

names(Data) <- c("Obsvn","ID","thick","uniformSize","uniformShape","adhesion","cellSize","bareNuclei","Chromatin",
	"normalNuclei","Mitoses","Class")

dim(Data)
p <- dim(Data)[2] - 3

head(Data)
y <- 1+as.integer(Data$Class=="malignant")
X <- Data[,3:(2+p)]

n <- dim(Data)[1]
set.seed(123)
r <- sample(10,n,replace =T)

K <- 20
predictions.cv <- matrix(0,n,K)
predictions.app <- matrix(0,n,K)

u <- sort(unique(r))
for (k in 1:K){
	for (i in 1:length(u)){
		predictions.cv[r==i,k] <- knn(train = X[r!=i,], test = X[r==i,], cl=y[r!=i], k = k, l = 0, prob = FALSE, use.all = TRUE)
		predictions.app[,k] <- knn(train = X, test = X, cl=y, k = k, l = 0, prob = FALSE, use.all = TRUE)

	}
}

acc.cv <- rep(0,K)
acc.app <- rep(0,K)
for (k in 1:K){
	acc.cv[k] <- mean((as.integer(predictions.cv[,k])==as.integer(y)))
	acc.app[k] <- mean((as.integer(predictions.app[,k])==as.integer(y)))
}

acc.cv
plot(c(1,K),c(0.95,1),type="n",xlab="k",ylab="Accuracy estimate")
lines(1:K,acc.cv,col=1)
lines(1:K,acc.app,col=2)
legend("topright",legend=c("CV estimate","Apparent accuracy estimate"),col=1:2,lty=1)

############################## 10 repetitions

acc.cv <- matrix(0,10,K)
acc.app <- matrix(0,10,K)

for (rept in  1:10) {
	u <- sort(unique(r))
	for (k in 1:K){
		for (i in 1:length(u)){
			predictions.cv[r==i,k] <- knn(train = X[r!=i,], test = X[r==i,], cl=y[r!=i], k = k, l = 0, prob = FALSE, use.all = TRUE)
			predictions.app[,k] <- knn(train = X, test = X, cl=y, k = k, l = 0, prob = FALSE, use.all = TRUE)
		}
	}

	for (k in 1:K){
		acc.cv[rept,k] <- mean((as.integer(predictions.cv[,k])==as.integer(y)))
		acc.app[rept,k] <- mean((as.integer(predictions.app[,k])==as.integer(y)))
	}
}
acc.cv.10 <- colMeans(acc.cv)
acc.app.10 <- colMeans(acc.app)

plot(c(1,K),c(0.95,1),type="n",xlab="k",ylab="Accuracy estimate")
lines(1:K,acc.cv.10,col=1)
lines(1:K,acc.app.10,col=2)
legend("topright",legend=c("CV estimate","Apparent accuracy estimate"),col=1:2,lty=1)

lines(1:K,acc.cv[1,],col=3)
lines(1:K,acc.app[1,],col=4)

library(e1071)
tune.obj <- tune.knn(x=X, y=as.factor(y), k = 1:25, tunecontrol = tune.control(sampling = "cross",cross=10))
summary(tune.obj)