library(leaps)

descriptors = read.csv("mix_desc.csv")
train = read.csv("mix_train.csv",sep=";")
dtrain = merge(train[,c(1,3)],descriptors,by.x="NSC",by.y="Name")
test = read.csv("mix_test.csv",sep=";")
dtest = merge(test[,c(1,3)],descriptors,by.x="NSC",by.y="Name")

dtrain = dtrain[complete.cases(dtrain),]
dtest = dtest[complete.cases(dtest),]

remove = c()
for(i in 3:1445){
  if(var(dtrain[i])<0.1){
    print(names(dtrain)[i])
    remove = append(remove,i)
  }
}
dtrain = dtrain[,-remove]

n = 45
model = regsubsets(pKa~.,data = dtrain[-1],method="forward",nvmax=n)
summary(model)$rsq
summary(model)$adjr2
models = c()
trsq = c()
for(i in 1:n){
  #print(which(summary(model)$which[i,]==TRUE))
  f = paste(names(which(summary(model)$which[i,]==TRUE))[-1],collapse="+")
  print(f)
  m = lm(paste("pKa~",f),data=dtrain)
  models[[i]] = m
  p = predict(m,newdata = dtest)
  trsq = append(trsq,cor(dtest$pKa,p)^2)
}

plot(summary(model)$rsq)
points(trsq,col=2)