#Moderni metody DA budeme opet ilustrovat na datovem souboru IRIS;
iris
Species=iris$Species   #vektor spravnych trid

a=as.numeric(Species)  #vektor spravnych trid - ciselny

#matice scatterplotu
plot(iris[,c(1,2,3,4)],col=a)

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

#          Nejprve zopakujeme LDA a QDA  

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

library(MASS)
lda(Species~.,data=iris)
predict(lda(Species~.,data=iris))

PRED_LDA=predict(lda(Species~.,data=iris))$class

table(Species,PRED_LDA)   #matice zamen

qda(Species~.,data=iris)
predict(qda(Species~.,data=iris))

PRED_QDA=predict(qda(Species~.,data=iris))$class

table(Species,PRED_QDA)

#matice scatterplotu - klasifikace
plot(iris[,c(1,2,3,4)],col=a,pch=as.numeric(PRED_QDA))
##################################################################

#            Moderni metody DA

##################################################################
# DA zalozena na hloubce dat
library(ddalpha)

#pocitejme hloubku vsech bodu vzhledem ke kazde ze tri skupin
hloubka_Setosa <- depth.halfspace(iris[,c(1,2,3,4)], iris[Species=='setosa',c(1,2,3,4)])
hloubka_Versicolor <- depth.halfspace(iris[,c(1,2,3,4)], iris[Species=='versicolor',c(1,2,3,4)])
hloubka_Virginica <- depth.halfspace(iris[,c(1,2,3,4)], iris[Species=='virginica',c(1,2,3,4)])

cbind(hloubka_Setosa,hloubka_Versicolor,hloubka_Virginica)

#klasifikator maximalni hloubky
dd=ddalpha.train(Species ~ ., iris,depth = "halfspace",
                         separator = "maxD",outsider.methods = "LDA")
plot(dd)

classes_halfspace <- ddalpha.classify(dd,iris[,c(1,2,3,4)])
unlist(classes_halfspace)
table(Species,unlist(classes_halfspace))

#uvazujme nyni Mahalanobisovu hloubku vsech bodu vzhledem ke kazde ze tri skupin
hloubka_Setosa <- depth.Mahalanobis(iris[,c(1,2,3,4)], iris[Species=='setosa',c(1,2,3,4)])
hloubka_Versicolor <- depth.Mahalanobis(iris[,c(1,2,3,4)], iris[Species=='versicolor',c(1,2,3,4)])
hloubka_Virginica <- depth.Mahalanobis(iris[,c(1,2,3,4)], iris[Species=='virginica',c(1,2,3,4)])

cbind(hloubka_Setosa,hloubka_Versicolor,hloubka_Virginica)


dd=ddalpha.train(Species ~ ., iris,depth = "Mahalanobis",
                         separator = "maxD",outsider.methods = "LDA")

classes_Mahalanobis <- ddalpha.classify(dd,iris[,c(1,2,3,4)])
unlist(classes_Mahalanobis)
table(Species,unlist(classes_Mahalanobis))

#klasifikator, separacni utvar je polynom
dd=ddalpha.train(Species ~ ., iris,depth = "Mahalanobis",
                         separator = "polynomial",outsider.methods = "LDA")

classes_Mahalanobis <- ddalpha.classify(dd,iris[,c(1,2,3,4)])
unlist(classes_Mahalanobis)
table(Species,unlist(classes_Mahalanobis))

#dd-alpha klasifikator
dd=ddalpha.train(Species ~ ., iris,depth = "Mahalanobis",
                         separator = "alpha",outsider.methods = "LDA")

classes_Mahalanobis <- ddalpha.classify(dd,iris[,c(1,2,3,4)])
unlist(classes_Mahalanobis)
table(Species,unlist(classes_Mahalanobis))

#dd-alpha klasifikator s poloprostorovou hloubkou
dd=ddalpha.train(Species ~ ., iris,depth = "halfspace",
                         separator = "alpha",outsider.methods = "LDA")

classes_halfspace <- ddalpha.classify(dd,iris[,c(1,2,3,4)])
unlist(classes_halfspace)
table(Species,unlist(classes_halfspace))


library(mda)
#DA smes (mixture DA)

model=mda(Species~.,data=iris,subclasses = c(3,3,3))
model
PRED_MDA=predict(model)
table(Species,PRED_MDA)

#pridame dalsi prototypy
model=mda(Species~.,data=iris,subclasses = c(3,10,10))
model
PRED_MDA=predict(model)
table(Species,PRED_MDA)

#Regularizova DA
library(klaR)

#s jedinym parametrem lambda
rda(Species~.,data=iris,gamma=0)
PRED_RDA=predict(rda(Species~.,data=iris))$class
table(Species,PRED_RDA)

#s obema parametry lambda a gamma
rda(Species~.,data=iris)
PRED_RDA2=predict(rda(Species~.,data=iris))$class
table(Species,PRED_RDA2)