library(MASS)
library(klaR)
library(e1071)

names <- c("clas", "alc","malic alc","ash","alcality of ash","magnesium","total phenols","flavanoids","nonflavanoids phenols","proanthocyanians","color intens","hue","od280","proline")
data <- read.csv(url("https://archive.ics.uci.edu/ml/machine-learning-databases/wine/wine.data"),col.names = names)

class.lda <- lda(clas ~ . , data)
class.qda <- qda(clas ~ ., data)
class.nb <- naiveBayes(clas ~ ., data)



CM.large <- function(org.class, pred.class) {

  CM <- table(org.class, pred.class)

  # Skuteczność klasyfikatora
  ACC <- sum(diag(CM)) / sum(CM)

  # Wartości true positive i true negative
  # zakładamy, że klasa "2" jest "pozytywna"
  TP <- CM[2,2]
  TN <- CM[1,1]

  sums <- apply(CM, 1, sum)

  TPR <- TP / sums[2]
  FPR <- 1 - TN / sums[1]

  return(c(ACC = round(ACC,4), TP = TP, TN = TN, TPR = round(TPR, 4), FPR = round(FPR, 4), row.names = NULL))
}

data.lda.old <- predict(class.lda, data)
data.qda.old <- predict(class.qda, data)
data.nb.old <- predict(class.nb, data)

res.old <- CM.large(data$clas, data.lda.old$clas)
res.old <- rbind(res.old, CM.large(data$clas, data.qda.old$clas))
res.old <- rbind(res.old, CM.large(data$clas, data.nb.old))

rownames(res.old) <- c("LDA", "QDA", "NB")

res.old