### question 4 hint ####
library(MASS)
library(class)
Boston$medv01 <- ifelse(Boston$medv > median(Boston$medv),1,0)
names(Boston)

# Choice of subsets
plot(Boston)

#subset 1: "lstat","rm"
#subset 2: "lstat", "rm","age","rad"
#subset 3: all predictors

Boston$medv <- NULL
names(Boston)

### try three values of K #######
##### K=1 #######
x.train <- Boston[,c("lstat","rm")]
y.train <- Boston[,"medv01"]
test.err <- mean(knn.cv(x.train, y.train, k=1) !=y.train)
test.err

x.train <- Boston[,c("lstat", "rm","age","rad")]
y.train <- Boston[,14]
test.err <- mean(knn.cv(x.train, y.train, k=1) !=y.train)
test.err

x.train <- Boston[,-14]### using all the predictors
y.train <- Boston[,14]
test.err <- mean(knn.cv(x.train, y.train, k=1) !=y.train)
test.err

x.train <- Boston[,c("lstat","rm")]
y.train <- Boston[,"medv01"]
test.err <- mean(knn.cv(x.train, y.train, k=3) !=y.train)
test.err

x.train <- Boston[,c("lstat", "rm","age","rad")]
y.train <- Boston[,14]
test.err <- mean(knn.cv(x.train, y.train, k=3) !=y.train)
test.err

x.train <- Boston[,-14]### using all the predictors
y.train <- Boston[,14]
test.err <- mean(knn.cv(x.train, y.train, k=3) !=y.train)
test.err

x.train <- Boston[,c("lstat","rm")]
y.train <- Boston[,"medv01"]
test.err <- mean(knn.cv(x.train, y.train, k=10) !=y.train)
test.err

x.train <- Boston[,c("lstat", "rm","age","rad")]
y.train <- Boston[,14]
test.err <- mean(knn.cv(x.train, y.train, k=10) !=y.train)
test.err

x.train <- Boston[,-14]### using all the predictors
y.train <- Boston[,14]
test.err <- mean(knn.cv(x.train, y.train, k=10) !=y.train)
test.err

###SCALED DATA####
X.train <- Boston[,c("lstat","rm")]
X.train.scaled <- scale(X.train)
y.train <- Boston[,"medv01"]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=1) !=y.train)
test.err

X.train <- Boston[,c("lstat", "rm","age","rad")]
X.train.scaled <- scale(X.train)
y.train.scaled <- Boston[,14]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=1) !=y.train.scaled)
test.err

X.train <- Boston[,-14]### using all the predictors
X.train.scaled <- scale(X.train)
y.train.scaled <- Boston[,14]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=1) !=y.train)
test.err

X.train <- Boston[,c("lstat","rm")]
X.train.scaled <- scale(X.train)
y.train <- Boston[,"medv01"]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=3) !=y.train)
test.err

X.train <- Boston[,c("lstat", "rm","age","rad")]
X.train.scaled <- scale(X.train)
y.train.scaled <- Boston[,14]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=3) !=y.train.scaled)
test.err

X.train <- Boston[,-14]### using all the predictors
X.train.scaled <- scale(X.train)
y.train.scaled <- Boston[,14]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=3) !=y.train)
test.err

X.train <- Boston[,c("lstat","rm")]
X.train.scaled <- scale(X.train)
y.train <- Boston[,"medv01"]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=10) !=y.train)
test.err

X.train <- Boston[,c("lstat", "rm","age","rad")]
X.train.scaled <- scale(X.train)
y.train.scaled <- Boston[,14]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=10) !=y.train.scaled)
test.err

X.train <- Boston[,-14]### using all the predictors
X.train.scaled <- scale(X.train)
y.train.scaled <- Boston[,14]
test.err <- mean(knn.cv(X.train.scaled, y.train, k=10) !=y.train)
test.err