ricardo.R

load("paintings_train.Rdata")
load("paintings_test.Rdata")
load("paintings_validation.Rdata")

paintings_full <- rbind(paintings_train, paintings_test)
paint_full <- clean_data(paintings_full) %>% distinct()
paint_valid <- clean_data(paintings_validation)


main.lm <- lm(logprice ~ dealer + year + Interm + origin_cat + endbuyer +
                        engraved + prevcoll + finished + lrgfont + discauth + logSurface +
                        winningbiddertype + winningbiddertype*dealer + logSurface*position +
                        dealer*position +
                        ## interactions
                        
                        dealer*I(year^3) + lrgfont*year + lrgfont*Interm +
                        endbuyer*dealer + endbuyer*Interm + origin_cat*logSurface,
                        
              data = paint_full)


aux.lm <- lm(logprice ~ dealer + year + Interm + origin_cat + endbuyer +
                        engraved + prevcoll + finished + lrgfont + discauth +
                        winningbiddertype + winningbiddertype*dealer +
                        dealer*position +
                        ## interactions
                        
                        dealer*I(year^3) + lrgfont*year + lrgfont*Interm +
                        endbuyer*dealer + endbuyer*Interm,
                        
              data = paint_full)

predict_main <- as.data.frame(
  exp(predict(main.lm, newdata=paint_valid,
              interval = "pred")))
predict_aux <- as.data.frame(
  exp(predict(aux.lm, newdata=paint_valid[is.na(paint_valid$logSurface),],
              interval = "pred")))

predict_valid <- predict_main
predict_valid[is.na(predict_valid$fit),] <- predict_aux

predictions <- predict_valid

top.price <- data.frame(id = 1:nrow(predictions), price = predictions$fit, paint_valid) %>% arrange(desc(price)) %>% head(10) %>% 
  select(id, price, year, winningbiddertype, Interm, engraved, finished, prevcoll, logSurface)

top.price %>% kable()