#CHAPTER 4 EXAMPLES IN R


#Exploring the iris data set.
head(iris)
attach(iris)
plot(Petal.Length,Petal.Width)
plot(Petal.Length,Petal.Width,col=Species)
plot(Petal.Length,Petal.Width,col=c('blue','red','purple')[Species])


#Decision trees with the rpart package.
library(rpart)
library(rattle)

iristree=rpart(Species~Sepal.Length+Sepal.Width+Petal.Length+Petal.Width,  data=iris)
iristree=rpart(Species~.,data=iris)

fancyRpartPlot(iristree)


#Confusion matrix.
predSpecies=predict(iristree,newdata=iris,type="class")
confusionmatrix=table(Species,predSpecies)
confusionmatrix


#A second look at the iris scatterplot.
plot(jitter(Petal.Length),jitter(Petal.Width),col=c('blue','red','purple')[Species])
lines(1:7,rep(1.8,7),col='black')
lines(rep(2.4,4),0:3,col='black')


#Accuracy for rpart tree.
accuracy=sum(diag(confusionmatrix))/sum(confusionmatrix)
accuracy

#confmatrix function
confmatrix=function(y,predy){
  matrix=table(y,predy)
  accuracy=sum(diag(matrix))/sum(matrix)
  return(list(matrix=matrix,accuracy=accuracy,error=1-accuracy))
}


#confmatrix for iristree
confmatrix(iris$Species,
           predict(iristree,newdata=iris,type="class"))   


#The party package.
library(party)
iristree2=ctree(Species~.,data=iris)
plot(iristree2)


#Simple plot.
plot(iristree2,type='simple')


#ctree confusion matrix.
predSpecies=predict(iristree2,newdata=iris)
confusionmatrix=table(Species,predSpecies)
confusionmatrix


#Controling the depth of the tree.
iristree3=ctree(Species~.,data=iris, controls=ctree_control(maxdepth=2))
plot(iristree3)



#########################################
#Training and Testing Sets for Iris Data#
#########################################

train=sample(150,105)
iris[train,]    #Training data  (70% of the data)
iris[-train,]   #Test data      (30% of the data)

#More general code for training set
train=sample(nrow(iris),round(0.7*nrow(iris),0))


#Use training data to construct a tree.
library(rpart)
library(rattle)
iristree=rpart(Species~.,data=iris[train,])
fancyRpartPlot(iristree)

#confmatrix function
confmatrix=function(y,predy){
  matrix=table(y,predy)
  accuracy=sum(diag(matrix))/sum(matrix)
  return(list(matrix=matrix,accuracy=accuracy,error=1-accuracy))
}

#Confusion matrix, accuracy, and error rate for training data.
confmatrix(iris$Species[train],
           predict(iristree,newdata=iris[train,],type="class"))


#Confusion matrix, accuracy, and error rate for test data.
confmatrix(iris$Species[-train],
           predict(iristree,newdata=iris[-train,],type="class"))



##############
#Example Data#
##############

#Import traindata.csv and testdata.csv
#Make sure class variable is a factor
traindata$class=as.factor(traindata$class)
testdata$class=as.factor(testdata$class)

#Quick Data Exploration
dim(traindata)
head(traindata)
plot(traindata$x,traindata$y,col=traindata$class)

dim(testdata)
head(testdata)
plot(testdata$x,testdata$y,col=testdata$class)

#Building Tree 1 from the Slides
extree1=rpart(class~.,data=traindata)
plot(extree1)
confmatrix(traindata$class,predict(extree1,newdata=traindata,type='class'))
confmatrix(testdata$class,predict(extree1,newdata=testdata,type='class'))

#Number of Nodes for Tree 1
extree1$frame       #Frame of information about the nodes
dim(extree1$frame)  #First entry tells us how many nodes there are


#Class Breakdown for Training and Testing Data
table(traindata$class)
table(traindata$class)/900

table(testdata$class)
table(testdata$class)/2100


##################################################
#Confidence Intervals for Classification Accuracy#
##################################################

#Exact binomial test.
#Example test data had 2100 records, and 1488 were classified correctly.
confmatrix(testdata$class,predict(extree1,newdata=testdata,type='class'))
binom.test(1488,2100)


#Building tree 2
extree2=rpart(class~.,data=traindata,
              control=rpart.control(minsplit=1,cp=0))
plot(extree2)
confmatrix(traindata$class,predict(extree2,newdata=traindata,type='class'))
confmatrix(testdata$class,predict(extree2,newdata=testdata,type='class'))

#Building accuracy vectors
accvector1=(testdata$class==predict(extree1,newdata=testdata,type='class'))
table(accvector1)
table(accvector1)/2100

accvector2=(testdata$class==predict(extree2,newdata=testdata,type='class'))
table(accvector2)
table(accvector2)/2100

#McNemar Table
mcnemartable=table(accvector1,accvector2)
mcnemartable

#Chi-square statistic and p-value
chisq=(abs(174-278)-1)^2/(174+278)
chisq
pchisq(chisq,df=1,lower.tail=FALSE)


#Built-in Function
mcnemar.test(mcnemartable)

#Exact McNemar Test
library(exact2x2)
mcnemar.exact(mcnemartable)




####10-fold Cross-validation####

#Combine traindata and testdata.
Exdata=rbind(traindata,testdata)
folds=cvFolds(nrow(Exdata),K=10,type='random')


#createfolds function
createfolds=function(n,K){
  reps=ceiling(n/K)
  folds=sample(rep(1:K,reps))
  return(folds[1:n])
}

#Folds for Exdata
set.seed(5364)
folds=createfolds(nrow(Exdata),10)

#Accuracy for first fold
temptest=Exdata[folds==1,]
temptrain=Exdata[folds!=1,]

dim(temptest)
dim(temptrain)

colSums(Exdata[,1:2])
colSums(temptest[,1:2])+colSums(temptrain[,1:2])

temptree=rpart(class~.,data=temptrain)
tempacc=confmatrix(temptest$class,
                   predict(temptree,newdata=temptest,type="class"))$accuracy

tempacc


#Accuracy for all folds using a loop.
accvector=1:10

for(k in 1:10){
  temptest=Exdata[folds==k,]
  temptrain=Exdata[folds!=k,]
  
  temptree=rpart(class~.,data=temptrain)
  accvector[k]=confmatrix(temptest$class,
                     predict(temptree,newdata=temptest,type="class"))$accuracy
}

mean(accvector)


#Delete-d Hints
#Let d=20

index=sample(nrow(Exdata))
index[1:20]
index[21:nrow(Exdata)]


#Bootstrap Hints

index=sample(nrow(Exdata),replace=TRUE)