library(MASS)
library(class)
data(fgl)

head(fgl)
table(fgl$type)

predictions <- knn(train = fgl[,1:9], test = fgl[,1:9], cl=fgl[,10], k = 1, l = 0, prob = FALSE, use.all = TRUE)

mean(predictions ==  fgl[,10])

predictions <- knn.cv(train = fgl[,1:9],  cl=fgl[,10], k = 1, l = 0, prob = FALSE, use.all = TRUE)
mean(predictions ==  fgl[,10]) # n-fold cross-validation estimate

X <- as.matrix(fgl[,1:9])
R <- cor(X)
e <- eigen(R)
PC <- scale(X)%*%e$vectors[,1:2]

library(lattice)
xyplot(PC[,2]~PC[,1],groups=fgl[,10],col=1:10,pch=16)

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

predictions <- fgl[,10]
u <- sort(unique(r))
for (i in 1:length(u)){
	predictions[r==i] <- knn(train = fgl[r!=i,1:9], test = fgl[r==i,1:9], cl=fgl[r!=i,10], k = 1, l = 0, prob = FALSE, use.all = TRUE)
	cat("fold",i,"acc=",mean(predictions[r==i] ==  fgl[r==i,10]),"\n",sep=" ")
}
mean(predictions ==  fgl[,10]) # 10-fold cross-validation estimate

reps <- 100
Acc.cv <- rep(0,reps)

for  (j in 1:reps) {

	set.seed(j)
	r <- sample(10,n,replace =T)

	u <- sort(unique(r))
	for (i in 1:length(u)){
		predictions[r==i] <- knn(train = fgl[r!=i,1:9], test = fgl[r==i,1:9], cl=fgl[r!=i,10], k = 1, l = 0, prob = FALSE, use.all = TRUE)
	}
	Acc.cv[j] <- mean(predictions ==  fgl[,10])}

mean(Acc.cv) # better 10-fold cross-validation estimate

y <- fgl[,10]

tabl <- table(fgl[,10],predictions)

options(digits=3)

100*diag(tabl/rowSums(tabl))
rowSums(tabl)
100*sum(diag(tabl))/sum(tabl)

tabl2 <- rbind(cbind(tabl,rowSums(tabl)),c(colSums(tabl),sum(sum(tabl))))
rownames(tabl2) <- c(rownames(tabl),"Total")
colnames(tabl2) <- c(colnames(tabl),"Total")
"Rows = actual       columns = predicted"
tabl2