setwd('D:/wonga/')
data<-read.table("cs-training.csv",header = TRUE,sep=",",na.strings="NA")
datatest<-read.table("cs-test.csv",header = TRUE,sep=",",na.strings="NA")
data<-rbind(data,datatest)
data$Sf<-as.factor(data$Serious)
#data<-data.frame(matrix(unlist(data22),ncol=ncol(data22),dimnames=list(NULL,colnames(data22))))
datafull<-data[complete.cases(data),]

#
# Impute the mising value
#
cor(data[,3:12],use="complete.obs")
data[is.na(data$Income),6]
data[is.na(data$Income)&data$DR>2,7]<-0#impute
data[is.na(data$Income)&data$DR==2,7]<-0

datacom<-data[complete.cases(data),]
library(rpart)

head(datacom)
datacom$NumOfDepF<-factor(datacom$NumOfDep)
fit<-rpart(NumOfDepF~Revolving+age+NumOfTime30.59+DR+NumRELL,method="class",data=datacom

library(DMwR)
nlist<-c(3,4,5,6,8,10,12)
data[data$age<30,nlist]<-knnImputation(data[data$age<30,nlist],k=3,meth="median")
dim(data[data$age<30&is.na(data$NumOfDep),])
dim(data[data$age>30&data$age<40,])
data[data$age>=30&data$age<35,nlist]<-knnImputation(data[data$age>=30&data$age<35,nlist],k=3,meth="median")
dim(data[data$age>=30&data$age<35&is.na(data$NumOfDep),])
data[data$age>=35&data$age<40,nlist]<-knnImputation(data[data$age>=35&data$age<40,nlist],k=3,meth="median")
data[data$age>=40&data$age<45,nlist]<-knnImputation(data[data$age>=40&data$age<45,nlist],k=3,meth="median")
data[data$age>=45&data$age<50,nlist]<-knnImputation(data[data$age>=45&data$age<50,nlist],k=3,meth="median")
data[data$age>=50&data$age<60,nlist]<-knnImputation(data[data$age>=50&data$age<60,nlist],k=3,meth="median")
data[data$age>=60&data$age<70,nlist]<-knnImputation(data[data$age>=60&data$age<70,nlist],k=3,meth="median")
data[data$age>=70,nlist]<-knnImputation(data[data$age>=70,nlist],k=3,meth="median")
        
nlist<-c(3,4,5,6,7,8,10)		
data[data$age<30,nlist]<-knnImputation(data[data$age<30,nlist],k=3)
data[data$age>=30&data$age<35,nlist]<-knnImputation(data[data$age>=30&data$age<35,nlist],k=3)
data[data$age>=35&data$age<40,nlist]<-knnImputation(data[data$age>=35&data$age<40,nlist],k=3)
data[data$age>=40&data$age<45,nlist]<-knnImputation(data[data$age>=40&data$age<45,nlist],k=3)
data[data$age>=45&data$age<50,nlist]<-knnImputation(data[data$age>=45&data$age<50,nlist],k=3)
data[data$age>=50&data$age<55,nlist]<-knnImputation(data[data$age>=50&data$age<55,nlist],k=3)
data[data$age>=55&data$age<60,nlist]<-knnImputation(data[data$age>=55&data$age<60,nlist],k=3)
data[data$age>=60&data$age<70,nlist]<-knnImputation(data[data$age>=60&data$age<70,nlist],k=3)
data[data$age>=70,nlist]<-knnImputation(data[data$age>=70,nlist],k=3)   
           
data[!complete.cases(data),"Serious"]<-0
data[complete.cases(data),]
		   
nlist<-c(3,4,5,6,7,8,19,12,13)
data[-c(1,2,12,13)]<-scale(data[-c(1,2,12,13)])#???
write.csv(data,file="normdata.csv")

#
# Adding new features
#
days<-data[5]*45+data[9]*90+data[11]*75
names(days)<-"days"
data<-data.frame(data,days)

netIncome<-data[7]-data[7]*data[6]
names(netIncome)<-"netIncome"
data<-data.frame(data,netIncome)

Incomepp<-ifelse(data$netIncome>0,data$netIncome/(data$NumOfDep+1),0)
names(Incomepp)<-"Incomepp"
data<-data.frame(data,Incomepp)

#
# Transform the data
#
dmatrix<-as.matrix(data)#??
digmatrix<-as.numeric(dmatrix)
dfmatrix<-matrix(digmatrix,nrow=150000)#150000 will change to 251503(dim(data)[1]) depending on how many data point you have
l2matrix<-l2norm(dfmatrix)
l2matrix[,2]<-data[,2]
l2matrix[,1]<-data[,1]
write.csv(l2matrix,file="normdata.csv")

data2<-data.frame(l2matrix)
names(data2)<-c("id","Serious","Revolving","age","NumOfTime30.59", "DR","Income","NumOfOCL","NumOfTimes90","NumRELL","NumOfTime60.89","NumOfDep","days",   "netIncome","Incomepp"  )

l2norm = function(matr){
  leng <- dim(matr)[2]
  dg<-sqrt(diag(t(matr)%*%matr))
  for(k in 1:leng)
  {
    matr[,k]<- matr[,k]/dg[k]
  }
  return(matr)
}

#
# Train models
#

#the best model is trained using
credit.model.nnet<-nnet(Serious~Revolving+age+NumOfTime30.59+DR+Income+NumOfOCL+NumOfTimes90+NumRELL+NumOfTime60.89+NumOfDep+days+netIncome,data=data[1:150000,],size=10,linout=FALSE,decay=0.025,maxit=500)
nnet.preds<-predict(credit.model.nnet,data[150001:251503,])
write.csv(nnet.preds,file="pred352.csv")

data<-read.table("train.csv",header = TRUE,sep=",",na.strings="NA")
credit.model.nnet<-nnet(Serious~Revolving+age+NumOfTime30.59+DR+Income+NumOfOCL+NumOfTimes90+NumRELL+NumOfTime60.89+NumOfDep+days+netIncome+Incomepp,data=data2[1:150000,],size=8,linout=FALSE,decay=0.025,maxit=500)
test<-read.table("test.csv",header = TRUE,sep=",",na.strings="NA")
nnet.preds<-predict(credit.model.nnet,data2[150001:251503,])
write.csv(nnet.preds,file="nnetpred.csv")

@Tools\liblinear-1.8\windows\run the following command
train -v 10 -c 1 -s 2 -w1 1 l2normsvmdata.txt