# przyklad 6.4
options(OutDec=",")
library("ElemStatLearn")
library("mlbench")
library("rpart")
library("adabag")
library("gbm")

#####################################################

data(spam)
m<-nrow(spam)

ks<-c(1,seq(5,100,by=5))

spam.bag.bl<-numeric(length=length(ks))
spam.ada.bl<-numeric(length=length(ks))

spam.ada.blad<-numeric(length=length(ks))
spam.bag.blad<-numeric(length=length(ks))

iter<-50

for (j in 1:iter)
{
print(j)
ucz <- sample(1:m, size = 3000, replace = FALSE) 
spam.ucz<-spam[ucz,]
spam.test<-spam[-ucz,]

for (k in seq(along=ks)) 
{
#### petla ###

lmod<-ks[k] # liczba modeli pojedynczych do klasyfikacji

print(lmod)
# boosting

spam.ada<-adaboost.M1(spam ~.,data=spam.ucz,mfinal=lmod, maxdepth=1)
spam.ada.pred <- predict.boosting(spam.ada,newdata=spam.test)
spam.ada.blad[k]<-spam.ada.pred$error

## bagging

spam.bag<-bagging(spam ~.,data=spam.ucz,mfinal=lmod, maxdepth=1)
spam.bag.pred <- predict.bagging(spam.bag,newdata=spam.test)
spam.bag.blad[k]<-spam.bag.pred$error

}
spam.bag.bl<-spam.bag.bl+spam.bag.blad
spam.ada.bl<-spam.ada.bl+spam.ada.blad
}
spam.bag.bl<-spam.bag.bl/iter
spam.ada.bl<-spam.ada.bl/iter

# rys. 6.4

aa<-min(min(spam.bag.bl),min(spam.ada.bl))
bb<-max(max(spam.bag.bl),max(spam.ada.bl))

plot(ks,spam.ada.bl,type="n",ylim=c(aa,bb),xlab="Liczba modeli bazowych",ylab="Błąd predykcji",main="Zbiór Spam")
lines(ks,spam.ada.bl,type="l",lty=1)
lines(ks,spam.bag.bl,type="l",lty=2)
points(ks,spam.ada.bl,pch=19)
points(ks,spam.bag.bl,pch=19)
legend("right",lty=c(1,2),legend = c("Boosting","Bagging"))