# spam
library("MASS")
library("class")
library("nnet")
library("tree")
attach(spam)

set.seed(123)
m <-length(spam$spam)
ucz <- sample(1:m, size = round(m/3), replace = FALSE, prob = rep(1/m, m)) 
spam.ucz <- data.frame(spam[ucz,])
spam.test <- data.frame(spam[-ucz,])

ks<-c(3:10,15, 20, 25, seq(30,100,by=10))
nks<-length(ks)
blad.ucz<-numeric(length=nks)
blad.test<-numeric(length=nks)
names(blad.ucz)<-names(blad.test)<-ks
spam.tr<-tree(spam~.,spam.ucz, mindev = 0, minsize = 2)

for (i in seq(along=ks)) 
{
spam.pr<-prune.tree(spam.tr,best=ks[i])
Z.ucz<-predict.tree(spam.pr,spam.ucz,type ="class")
Z.test<-predict.tree(spam.pr,spam.test,type ="class")
blad.ucz[i]<-1-sum(Z.ucz==spam.ucz$spam)/nrow(spam.ucz)
blad.test[i]<-1-sum(Z.test==spam.test$spam)/nrow(spam.test)
}

plot(ks,blad.ucz,xlab="Wielkość drzewa",ylab="Błąd klasyfikacji",ylim=c(0,0.2),type="n")
lines(ks,blad.test,type="b",col='blue',pch=1)
lines(ks,blad.ucz,type="b",col='red',pch=3)
legend("topright",lty=1,pch=c(3,1), col=c("red","blue"),legend = c("Zbiór uczący ", "Zbiór testowy "))