## R code for Chapter 11 in The Elements of Statistical Learning (ESL)

## Load R package for the book
library("ElemStatLearn")

## load the training data
data(vowel.train)
str(vowel.train)
# 'data.frame':	528 obs. of  11 variables:
# $ y   : int  1 2 3 4 5 6 7 8 9 10 ...
# $ x.1 : num  -3.64 -3.33 -2.12 -2.29 -2.6 ...
# $ x.2 : num  0.418 0.496 0.894 1.809 1.938 ...

## use neural network method
## need package "nnet"
library("nnet")
y <- class.ind(vowel.train$y) # Generates class indicator from the given factor "xclass"
X <- vowel.train[,2:11]
## fit neural network model based on Cross-Validation
size.nn <- 10          # number of units in the hidden layer
decay.nn <- 0.001      # decay.nn.opt[18] = 0.0009765625 
rang.nn <- 0.3         # Initial random weights on [-rang, rang]
maxit.nn <- 5000        # maximum number of iterations
set.seed(123)
fit.nn <- nnet(X, y, size=size.nn, rang=rang.nn, decay=decay.nn, maxit=maxit.nn)
## check training error rate
train.label.nn <- apply(predict(fit.nn, X), 1, which.max)
sum(train.label.nn!=vowel.train$y) # 39
sum(train.label.nn!=vowel.train$y)/dim(y)[1]
#0.07386364

## check testing error rate
data(vowel.test)
Xt=vowel.test[,2:11]
tclass=vowel.test[,1]
test.label.nn <- apply(predict(fit.nn, Xt), 1, which.max)
sum(test.label.nn!=tclass) # 275
sum(test.label.nn!=tclass)/length(tclass) # 0.5952381