#Daten

#install.packages("gdata")
library(gdata)

setwd("C:\\Users\\Jürgen Fassman\\Documents\\lowerhousing\\Abgabe 3")

files<-list.files("csv")

alldata<-list()

for(i in 1:length(files)) {
  alldata[[i]]<-read.csv(paste("csv\\",files[i],sep=""),header=TRUE,sep="*",stringsAsFactors=FALSE)
}

allsplitdata<-list()
for(i in c(1:length(files))) {
  dataset<-alldata[[i]]
  ordereddata<-dataset[order(dataset$Id,dataset$ServerTimestamp,dataset$timestamp),]
  splitdata<-split(ordereddata,ordereddata$Id)
  allsplitdata[[i]]<-splitdata
}

dataitems<-array()
for(i in 1:length(files)) {
  dataitems[i]<-nrow(alldata[i][[1]])
}

goodfiles<-c()
for(i in 1:length(files)) {
  if(dataitems[i]>=100) {
    goodfiles<-c(goodfiles,i)
  }
}

listofids<-c()
for(i in goodfiles) {
  listofids<-c(listofids,names(allsplitdata[[i]]))
}

ids=unique(listofids)

sumofids<-array()
for(i in c(1:length(ids))) {
  counter<-0
  for(j in c(1:length(listofids))) {
    if(ids[i]==listofids[j]) counter<-counter+1
  }
  sumofids[i]<-counter
}

idswithsums<-cbind(ids,sumofids)

goodids<-c()
for(i in c(1:length(ids))) {
  if(idswithsums[i,2]=="41") goodids<-c(goodids,idswithsums[i,1])
}

numbers<-data.frame()
for(i in goodfiles) {
  buffer<-array()
  buffer[1]<-i
  for(j in c(1:length(goodids))) {
    splitid<-which(goodids[j] == names(allsplitdata[[i]]))
	buffer[j+1]<-nrow(allsplitdata[[i]][[splitid]])
  }
  numbers<-rbind(numbers,buffer)
}
numbers<-setNames(numbers,c("filenumber",goodids))

minimum<-array()
for(i in 2:9) {
  minimum[i-1]<-min(numbers[i])
}

idswithminimum<-cbind(goodids,minimum)

reallygoodids<-c()
for(i in c(1:length(goodids))) {
  if(as.numeric(idswithminimum[i,2])>=10) reallygoodids<-c(reallygoodids,idswithminimum[i,1])
}

gooddata<-data.frame();
for(i in goodfiles) {
  buffer<-array()
  for(j in c(1:length(reallygoodids))) {
    splitid<-which(reallygoodids[j] == names(allsplitdata[[i]]))
	for(k in c(1:10)) {
	  buffer[(j-1)*10+k]<-as.numeric(allsplitdata[[i]][[splitid]]$value[nrow(allsplitdata[[i]][[splitid]])-(10-k)])
	}
  }
  gooddata<-rbind(gooddata,buffer)
}
gooddatanames=c()
for(i in c(1:length(reallygoodids))) {
  for(j in c(1:10)) {
    gooddatanames=c(gooddatanames,paste(reallygoodids[i],j,sep=""))
  }
}
gooddata<-setNames(gooddata,gooddatanames)

successdata<-read.xls("machning_success.xlsx",stringsAsFactors=FALSE)

j<-1
success<-array()
for(i in goodfiles) {
  success[j]<-successdata$Success[i]
  j<-j+1
}
gooddata<-cbind(gooddata,success)


#Supervised Classification
library(e1071)
library(mlbench)
library(leaps)
library(MASS)
library(caret)
library(randomForest)

set.seed(600)
goodframe <- data.frame(gooddata)
goodframe$success<-as.integer(as.logical(goodframe$success))

#https://machinelearningmastery.com/feature-selection-with-the-caret-r-package/
#feature selection(recursive feature elimination)
control <- rfeControl(functions=rfFuncs, method="cv", number=10)
results <- rfe(goodframe[,1:70], goodframe[,71], sizes=c(1:70), rfeControl=control)
print(results)
predictors(results)
plot(results, type=c("g", "o"))

reduceddata <- data.frame(goodframe$ns.2.s..Channel.MachineAxis.aaLoad.u1.1.7,
						goodframe$ns.2.s..Channel.Spindle.driveLoad8,
						goodframe$ns.2.s..Channel.MachineAxis.aaTorque.u1.1.7,
						goodframe$ns.2.s..Channel.MachineAxis.aaLoad.u1.1.6,
						goodframe$ns.2.s..Channel.Spindle.driveLoad7,
						goodframe$ns.2.s..Channel.MachineAxis.aaLoad.u1.1.8,
						goodframe$ns.2.s..Channel.MachineAxis.aaLoad.u1.1.5,
						goodframe$ns.2.s..Channel.MachineAxis.aaLoad.u1.1.10,
						goodframe$success)

reduceddata$goodframe.success<-as.factor(reduceddata$goodframe.success)
						
#Splitting data into test and training data 75/25
set.seed(600)
reduceddata[,"train"] <- ifelse(runif(nrow(reduceddata))<0.75,1,0)
trainset <- reduceddata[reduceddata$train==1,]
testset <- reduceddata[reduceddata$train==0,]
trainColNum <- grep("train",names(trainset))
trainset <- trainset[,-trainColNum]
testset <- testset[,-trainColNum]
typeColNum <- grep("success",names(reduceddata))

#SVM
#https://eight2late.wordpress.com/2017/02/07/a-gentle-introduction-to-support-vector-machines-using-r/
#Linear
set.seed(600)
svm_model <- svm(goodframe.success~., data=trainset, method="C-classification", kernel="linear")
svm_model
pred_train <-predict(svm_model,trainset)
mean(pred_train==trainset$goodframe.success)
pred_test <-predict(svm_model,testset)
mean(pred_test==testset$goodframe.success)

#Radial
set.seed(600)
svm_model <- svm(goodframe.success~ ., data=trainset, method="C-classification", kernel="radial")
svm_model
pred_train <-predict(svm_model,trainset)
mean(pred_train==trainset$goodframe.success)
pred_test <-predict(svm_model,testset)
mean(pred_test==testset$goodframe.success)

#Sigmoid
set.seed(600)
svm_model <- svm(goodframe.success~ ., data=trainset, method="C-classification", kernel="sigmoid")
svm_model
pred_train <-predict(svm_model,trainset)
mean(pred_train==trainset$goodframe.success)
pred_test <-predict(svm_model,testset)
mean(pred_test==testset$goodframe.success)

#Polynomial
set.seed(600)
svm_model <- svm(goodframe.success~ ., data=trainset, method="C-classification", kernel="polynomial")
svm_model
pred_train <-predict(svm_model,trainset)
mean(pred_train==trainset$goodframe.success)
pred_test <-predict(svm_model,testset)
mean(pred_test==testset$goodframe.success)

#Tuning for radial
set.seed(600)
tune_out <- tune.svm(x=trainset[,-typeColNum],y=trainset[,typeColNum],gamma=10^(-3:3),cost=c(0.01,0.1,1,10,100,1000),kernel="radial")
tune_out$best.parameters$cost
tune_out$best.parameters$gamma
svm_model <- svm(goodframe.success~ ., data=trainset, method="C-classification", kernel="radial",cost=tune_out$best.parameters$cost,gamma=tune_out$best.parameters$gamma)
pred_train <-predict(svm_model,trainset)
mean(pred_train==trainset$goodframe.success)
pred_test <-predict(svm_model,testset)
mean(pred_test==testset$goodframe.success)

0.9411765
0.8571429


#https://www.r-bloggers.com/understanding-naive-bayes-classifier-using-r/
#Naive Bayes
set.seed(600)
nb_model <- naiveBayes(goodframe.success ~ ., data =trainset)
summary(nb_model)
pred_train <-predict(nb_model,trainset)
mean(pred_train==trainset$goodframe.success)
pred_test <-predict(nb_model,testset)
mean(pred_test==testset$goodframe.success)

0.8235294
0.5714286