#
# RESITVE
#

##########################################################
#
# pomozne funkcije (kopirane iz skripte)
#

CA <- function(obs, pred)
{
	t <- table(obs, pred)

	sum(diag(t)) / sum(t)
}

Sensitivity <- function(obs, pred, pos.class)
{
	tab <- table(obs, pred)

	tab[pos.class, pos.class] / sum(tab[pos.class,])
}

Specificity <- function(obs, pred, pos.class)
{
	tab <- table(obs, pred)
	neg.class <- which(row.names(tab) != pos.class)

	tab[neg.class, neg.class] / sum(tab[neg.class,])
}

brier.score <- function(obsMat, predMat)
{
	sum((obsMat - predMat) ^ 2) / nrow(predMat)
}

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

library(rpart)
library(rpart.plot)
library(nnet)





mdata <- read.table("movies.txt", header = T, sep = ",")

for (i in 18:24)
	mdata[,i] <- as.factor(mdata[,i])

mdata$mpaa <- as.factor(mdata$mpaa)

mdata$title <- NULL
mdata$budget <- NULL

summary(mdata)


learn <- mdata[mdata$year < 2004,]
test <- mdata[mdata$year >= 2004,]

# atribut "year" nam ni vec uporaben, zato ga odstranimo
learn$year <- NULL
test$year <- NULL
	
dt <- rpart(Comedy~., learn)
rpart.plot(dt)

observed <- test$Comedy
predicted <- predict(dt, test, type="class")

CA(observed, predicted)
Sensitivity(observed, predicted, "1")
Specificity(observed, predicted, "1")


predMat <- predict(dt, test, type = "prob")
obsMat <- class.ind(test$Comedy)

brier.score(obsMat, predMat)




predMat <- predict(dt, test, type = "prob")
p <- unique(sort(predMat[,"1"]))
threshVec <- sort(c(-Inf, p[-length(p)] + diff(p)/2, Inf), decreasing=T)

sensVec <- vector()
specVec <- vector()
for (th in threshVec)
{
	predicted <- ifelse(predMat[,"1"] > th, "1", "0")
	predicted <- factor(predicted, levels=levels(mdata$Comedy))

	sensVec <- append(sensVec, Sensitivity(observed, predicted, "1"))
	specVec <- append(specVec, Specificity(observed, predicted, "1"))
}

plot(x=1-specVec, y=sensVec, type="l", xlab="1-Specificity", ylab="Sensitivity", main="ROC krivulja")
abline(a=0, b=1, col="grey")


# izris ROC krivulj omogoca knjiznica pROC
library(pROC)
rocobj <- roc(test$Comedy, predMat[,"1"])
plot(rocobj)




#######################################################################################
#
# Resitev racunske naloge:	
#
#
# a) klasifikacijska tocnost:
#
#     (300+120)/(300+0+80+120)
#     = 0.84
#
#
# b) pricakovana tocnost vecinskega klasifikatorja: 
#
#    vecinski razred je "0"     
#
#    matrika zmot vecinskega klasifikatorja:
#
#        0   1
#    --+---+---+								 
#    0 |300| 0 |
#    --+---+---+
#    1 |200| 0 |
#    --+---+---+
#
#    300/500 = 0.6
#
# c) senzitivnost ("0" je pozitivni razred)
#
#    TP/POS = 300 / 300 = 1
#
# d) specificnost ("0" je pozitivni razred)
#
#    TN/NEG = 120 / 200 = 0.6
#
#######################################################################################