# przykład 3.4

options(OutDec=",")
library("ElemStatLearn")
library("mlbench")
library("rpart")
library("adabag")

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

data(spam)
m<-nrow(spam)

ks<-seq(3,100,by=2)

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))

for (j in 1:10)
{
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/10
spam.ada.bl<-spam.ada.bl/10


####################################
data(Satellite)

m<-nrow(Satellite)

ks<-seq(3,100,by=2)

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

for (j in 1:10)
{
print(j)

ucz <- sample(1:m, size = 4290, replace = FALSE) 
sat.ucz<-Satellite[ucz,]
sat.test<-Satellite[-ucz,]

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

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

sat.ada<-adaboost.M1(classes~.,data=sat.ucz,mfinal=lmod, maxdepth=1)

sat.ada.pred <- predict.boosting(sat.ada,newdata=sat.test)

sat.ada.blad[k]<-sat.ada.pred$error

print(sat.ada.pred$error)

# bagging

sat.bag<-bagging(classes~.,data=sat.ucz,mfinal=lmod, maxdepth=3)

sat.bag.pred <- predict.bagging(sat.bag,newdata=sat.test)

sat.bag.blad[k]<-sat.bag.pred$error

print(sat.bag.pred$error)

}
sat.bag.bl<-sat.bag.bl+sat.bag.blad
sat.ada.bl<-sat.ada.bl+sat.ada.blad
}
sat.bag.bl<-sat.bag.bl/10
sat.ada.bl<-sat.ada.bl/10


########### rysunki

# rys. 3.8
plot(sat.bl[2:100],type="l",ylim=c(0.15,0.18),main="Zbiór Satellite",xlab="Liczba modeli bazowych",ylab="Błąd klasyfikacji")
plot(spam.bl[2:100],type="l",ylim=c(0.08,0.11),main="Zbiór Spam",xlab="Liczba modeli bazowych",ylab="Błąd klasyfikacji")

blad<-0.001
r<-numeric(length=1000)
for (i in 2: 1000) r[i]<-as.numeric(abs(1-spam.bl[i]/spam.bl[i-1])<blad)
#for (i in 2: 1000) r[i]<-as.numeric(abs(1-sat.bl[i]/sat.bl[i-1])<blad)
for (i in 2:990) if(sum(r[i:(i+10)])==10) print(i)