library(MASS); library(nnet)
X=as.matrix(crabs[,4:8]); y=c(rep(1,50),rep(2,50),rep(3,50),rep(4,50))

pe1=sample(50); pe2=sample(50)+50; pe3=sample(50)+100; pe4=sample(50)+150
ppp.qda=rep(NA,5); ppp.nnet=rep(NA,5)

for(i in 1:5){
  ind=((i-1)*10+1):(i*10)
  te1=pe1[ind]; te2=pe2[ind]; te3=pe3[ind]; te4=pe4[ind]
  te=c(te1,te2,te3,te4); tr=(1:200)[-te]

  K=qda(X[tr,],y[tr])
  pred=predict(K,X[te,])$class 
  ppp.qda[i] = mean( y[te] == pred )
               
  S=nnet(X[tr,],class.ind(y[tr]),maxit=200, trace=F,size=16,softmax=T)
  pred2=apply(predict(S,X[te,]),1,which.max)
  ppp.nnet[i] = mean( y[te] == pred2 )
}
cat("\nP-stwo poprawnej predykcji qda:\n",round(ppp.qda,2),"\n")
cat("P-stwo poprawnej predykcji nnet:\n",round(ppp.nnet,2),"\n")