######## Intro ########

### 2 vektory
area <- c(1:12)

rate <- c(1.2, 1.5, 1.02, 1.63, 1.14, 1.86, 2.1, 1.43, 1.76, 2.08, 1.32, 1.6)

### vypsat promenne
ls()

### zobrazit
area

### funkce
fact<-function(x){
	if(x>1) {
		return (x*fact(x-1)) 
	} else {
	  return(1)
	}
}
fact(5)


### matice
mat <- cbind(area, rate)

### zobrazit 2 radek matice
mat[2, ]

### vykreslit matici
plot(mat)

### dalsi vektor
area <- c(1:12)

### vytvoreni sekvenci
rate <- seq(length=12, from=1, by=.1)

mat2 <- cbind(area,rate)

plot(mat)
### okno se nepremaze
par(new=TRUE)
plot(mat2)

### rozlisit a mit stejnou stupnici

plot(mat, main="pokusne matice", ylim=c(1, 2.2));par(new=TRUE);plot(mat2, ylim=c(1, 2.2), pch=2)
### pridame popisky
legend(x="topleft", legend=c("mat", "mat2"), pch=c(1,2), bg="white", inset=0.01)
### pridame "ohraniceni"
abline(v=c(6, 8), lty=4)
### text
text(7,1.2,"spatna oblast",srt=90)


######################################################################
######### klasifikace, rozhodovaci stromy, asociacni pravidla
# I. Play or Not To Play

### nacteni souboru z dialogu (TPONTPNom.csv)
data <- read.csv(file.choose())

### Zobrazeni:
data

### Prvni radek:
data[1,]

### 3-5 sloupec:
data[,3:5]

### rozhodovaci strom (package "rpart")
library(rpart)

### strom s cilovym atributem Play:
tree<-rpart(Play~., data, method="class")

### vypsat strom:
tree

### parametr minsplit pomuze vytvorit lepsi strom:
tree<-rpart(Play~., data, method="class", control=rpart.control(minsplit=5))
tree

### prvni testovani stromu - na trenovacich datech (jen pro zacatek, jinak NEDELAT!!!)
tree.predictions <- predict(tree, data[,1:4], type="class")

### vytvoreni matice zmateni = confusion matrix
table(data[,5], tree.predictions) 

### cross validation - manualne
err.vect <- vector() 
for(j in 1:10) {    # 10 pokusů
	select <- sample(1:nrow(data), 0.9*nrow(data))  
					# nahodna permutace dat
	train <- data[select,]     	# 90% dat pro trenink 
	test <- data[-select,]  	# zbyla data pro test (10%)
	tree <- rpart(Play~., train, control=rpart.control(minsplit=5))
	pred <- predict(tree, test[,1:4], type="class") 
	cmx<-table(test[, "Play"], pred)  # sloupec Play pro ověření predikce
	err<- 1 - ( sum(diag(cmx)) / sum(cmx) )
	err.vect <- c(err.vect, err) 
} 
err.vect
mean(err.vect) 

### vypocet asociacnich pravidel
library(arules)
rules<-apriori(data, parameter = list(support = 0.2,confidence = 0.6))
rules



##################
# II. Iris dataset
### knihovna obsahujici funkce pro nacteni a zapsani arff souboru
library(foreign)

### nacteni souboru z dialogu (iris.arff)
data.iris <- read.arff(file.choose())

### Zobrazeni
data.iris

library(rpart)

### Vytvoreni stromu ze vsech dat, cilovy atribut tentokrat "class"
tree.iris<-rpart(class~., data.iris, method="class")
tree.iris

### Pro otestovani stromu na datech iris rozdelime data nahodne na 
### 2/3 trenovacich a tretinu testovacich:

### nahodna permutace delky 150, do hodnoty 150
idx<-sample(150,150) 	

### 2/3 dat pouzijeme jako trenovaci data
train.iris <- data.iris[idx[1:100]]   

### zbyla 1/3 dat pro test
test.iris <- data.iris[idx[101:150] 
tree.iris <- rpart(class~., train.iris)

### nechame zpocitat predikce pro testovaci data (vypustime posledni sloupec "class")
pred.iris <- predict(tree.iris, test.iris[,1:4], type="class") 

### “matice zmateni” = confusion matrix
cmx.iris <- table(test.iris[, 5], pred.iris)
cmx.iris  			

### priklad vypoctu error rate (pomer spatne klasifikovanych prikladu, ke vsem prikladum)
err.iris <- 1 - (sum(diag(cmx.iris)) / sum(cmx.iris) )
err.iris







########################################################################
######### tm ##########
### Pavel Brazdil, LIAAD, FEP Universidade do Porto ###
library(tm)

### data volne dostupna na internetu - 20newsgroups
### http://people.csail.mit.edu/jrennie/20Newsgroups/
electr.train <- Corpus(DirSource("20news-bydate-train/sci.electronics"), readerControl=list(reader=readPlain, language="en_US"))

print(electr.train)
summary(electr.train)
inspect(electr.train[1:3])

### dalsi 3 korpusy
religion.train <- Corpus(DirSource("20news-bydate-train/talk.religion.misc"), readerControl=list(reader=readPlain, language="en_US"))

electr.test <- Corpus(DirSource("20news-bydate-test/sci.electronics"), readerControl=list(reader=readPlain, language="en_US"))

religion.test <- Corpus(DirSource("20news-bydate-test/talk.religion.misc"), readerControl=list(reader=readPlain, language="en_US"))

# indexy jednotlivych dokumentu
#   	electr.train 	– documents 1 .. 591 
#   	religion.train 	– documents 592 .. 968  (377 docs)
#   	electr.test 	– documents 969 .. 1361 (393 docs)
#   	religion.test 	– documents 1362 .. 1612  (251 docs)

all <- c(electr.train, religion.train, electr.test, religion.test)

### predzpracovani
 all <- tm_map(all, PlainTextDocument)
 all <- tm_map(all, removeWords, stopwords(language="english")) 
 all <- tm_map(all, stripWhitespace)
 all <- tm_map(all, tolower)
 all <- tm_map(all, removePunctuation)
 all <- tm_map(all, removeNumbers)
### stemming - velmi dlouhe...
### library(rJava); library(rWeka); library(Snowball)
### all <- tm_map(all, stemDocument)

DocumentTermMatrix(all)
dtm.all <- DocumentTermMatrix( all, control=list(minWordLength=2, minDocFreq=5))

findFreqTerms(dtm.all, 40)

dtm.all.frame <- as.data.frame(inspect( dtm.all))
class <- c(rep("sci",591), rep("rel",377), rep("sci",393), rep("rel",251))
dtm.all.frame <- cbind(dtm.all.frame, class)

## pripravime trenovaci data
train <- dtm.all.frame[1:968, 1:ncol( dtm.all.frame )]

### testovaci
test <- dtm.all.frame[969:1612, 1:ncol( dtm.all.frame ) - 1]
testclass <- dtm.all.frame[969:1612,ncol(dtm.all.frame)]

### rozhodovaci strom
library(rpart)
dt <- rpart(class ~ cable + circuit + ground + neutral + outlets + subject + wire + wiring + judas + ra + christ + elohim + father + god + gods + jehovah + jesus + lord + mcconkie + ps + son + unto, train)

### predikce
dt.predictions <- predict(dt, test, type="class")
table(testclass, dt.predictions)