library(randomForest)
library(MASS)
set.seed(123)
Mode <- function(x) {
  ux <- unique(x)
  return(ux[which.max(tabulate(match(x, ux))) & is.na(ux) == F])
}
df <- read.csv("train.csv")
df1 <- df
df2 <- read.csv("test.csv")
tt <- read.csv("sample_submission.csv")

for(i in 1:ncol(df))
{
  col <- df[[i]]
  if(is.numeric(col))
  {
    mn <- (mean(col, na.rm = T))
    col[is.na(col)] <- mn
    col <- log(col + 1)
  }
  if(is.factor(col[1]))
  {
    md <- Mode(col)
    #   print(md[1])
    col[is.na(col)] <- md[1]
  }
  df1[[i]] <- col
}

apply(df1, 2, function(x) length(unique(x)))
#df1 <- ifelse(length(unique(df1)) < 8, as.factor(df1), df1)
View(df1)
summary(df1)
#apply(df, 2, function(x){sum(is.na(x))})

#df$PoolQC <- ifelse(is.na(df$PoolQC)==TRUE,mode(df$PoolQC), df$PoolQC)


#fit <- lm(SalePrice ~ ., df1, family = "lm")

#ifit <- step(fit, direction = "backward")



#df1$y <- predict(ifit, df1)



for(i in 1:ncol(df2))
{
  col <- df2[[i]]
  if(is.numeric(col))
  {
    mn <- (mean(col, na.rm = T))
    col[is.na(col)] <- mn
    col <- log(col + 1)
  }
  if(is.factor(col[1]))
  {
    md <- Mode(col)
    #   print(md[1])
    col[is.na(col)] <- md[1]
  }
  df2[[i]] <- col
}

#model3 <- randomForest(SalePrice~., df1)

#install.packages("ROSE")
#library(ROSE)
#roc.curve(df1$SalePric e, predict(model3, df2))

rf <- randomForest(SalePrice ~ ., df1, mtry = 9, ntree = 500)

length(levels(df1))
df2$SalePrice <- 0
df1 <- rbind(df1, df2)

tt$SalePrice <- predict(rf, df1[1461:2919, 1:80])
#tt$SalePrice <- predict(ifit, df2)

#df2$SalePrice <- predict(ifit, df2)

#tt$SalePrice <- df2$SalePrice
tt$SalePrice <- exp(tt$SalePrice) - 1


write.csv(tt, file = "sm.csv", row.names = F)
q <- read.csv("sm.csv")
q