# Small example to illustrate rpart() on the kyphosis data


library("rpart")
set.seed(784)
kyphosis[sample(nrow(kyphosis), 4), ]

fit <- rpart(Kyphosis ~ ., data = kyphosis, method = "class")

fit
printcp(fit)
summary(fit)
names(fit) 



if(FALSE) {
  postscript("kyphosis.eps",
             horiz = FALSE, onefile = FALSE, print.it = FALSE,
             width = 9, height = 5)
} else {
  pdf("kyphosis.pdf",
      onefile = FALSE,
      width = 9, height = 5)
}

par(mar = c(5, 4, 2, 2) + 0.8, mfrow = c(1, 1), xpd = TRUE)
plot(fit)
text(fit)
dev.off() 

pfit <- predict(fit, kyphosis, type = "class")
tab <- table(with(kyphosis, Kyphosis), pfit)

# Error checking: the table must be square
if(!all(dimnames(tab)[[1]] == dimnames(tab)[[2]]))
  stop("not all dimnames are the same")

# Obtain the (resubstitution) misclassification rate
1 - sum(diag(tab)) / sum(tab)
# Other ways of obtaining the (resubstitution) misclassification rate
rev(fit$cptable[,"rel error"])[1] *  rev(fit$parms$prior)[1]
tail(fit$cptable[,"rel error"], 1) *  tail(fit$parms$prior, 1)