二手車價預測 Baseline

Demo R code

library(tidyverse)
library(e1071)

# 本次競賽衡量標準 Rsquare
Rsq <- function(out){
  if(identical(c("price","predict"), names(out))){
    mn_price <- mean(out$price)
    res <- 1 - sum((out$price - out$predict)^2)/sum((out$price - mn_price)^2)
    res <- res * 100
    res <- ifelse(res>0, res, 0)
    sprintf("%.5f", res)
  }else{
    sprintf("%.5f", 0)
  }
}

# 匯入檔案
train <- read_csv("trainset.csv")
test <- read_csv("testset.csv")
dat <- full_join(train, test)

# 將兩份檔案合併，再做標準化
dat <- 
  mutate(
    dat,
    stpowerPS = (powerPS-mean(powerPS))/diff(range(powerPS)),
    stkilometer = (kilometer-mean(kilometer))/diff(range(kilometer)),
    stad_exist_time = (ad_exist_time-mean(ad_exist_time))/diff(range(ad_exist_time)),
    styears = (yearOfRegistration-mean(yearOfRegistration)/diff(range(yearOfRegistration)))
  ) %>%  
  select( 
         id, price, stpowerPS, stkilometer, stad_exist_time,
         vehicleType, gearbox, fuelType, styears, notRepairedDamage, brand
  ) %>% 
  mutate(
    vehicleType = as.factor(vehicleType),
    gearbox = as.factor(gearbox),
    fuelType = as.factor(fuelType),
    notRepairedDamage = as.factor(notRepairedDamage),
    brand = as.factor(brand) %>% relevel("sonstige_autos")
  )

# 重新篩選出 trainset / testset
train <- dat %>% slice(1:200000)
test <- dat %>% slice(-(1:200000))

# 用兩千筆 sub tainset 建 svr 模型
set.seed(123)
train_sub <- sample_n(train, 2000)

fit_svr <- svm(price~., train_sub[,-1], 
               type="eps", cost=16, gamma=0.25, epsilon=0.1, kernel="radial")

# 檢查 train set 的 Rsq
out_train <- data.frame(price=train_sub$price, predict=predict(fit_svr))
Rsq(out_train) # 72.40977

# 匯入submit.csv 並且將預測結果置入 predict 欄位
submit <- read_csv("submit.csv")
submit$predict <- predict(fit_svr, test)

# 結果寫入 submit01.csv 並將其上傳
submit %>% write_csv("submit01.csv") # 1.53