gistfile1.txt

library(caret)
library(dplyr)

data(diamonds)


train_i <- sample(c(FALSE, TRUE), nrow(diamonds), replace = TRUE, prob = c(0.2, 0.8))
diamonds_train <- diamonds[train_i, ]
diamonds_test <- diamonds[!train_i, ]

folds <- createFolds(diamonds_train$price, k = 5, returnTrain = TRUE)


trcntrl <- trainControl(index = folds, savePredictions = TRUE, search = 'grid')


m1 <- train(price ~ carat + cut + depth + table,
            data = diamonds_train,
            method = 'enet',
            tuneLength = 10,
            trControl = trcntrl)

m1$results$RMSE %>% min

m1_error <- sqrt(mean((predict(m1, diamonds_test) - diamonds_test$price) ^ 2))



m2 <- train(price ~ carat + cut + depth + table,
            data = diamonds_train,
            method = 'ranger',
            tuneLength = 4,
            trControl = trcntrl)

m2$results$RMSE %>% min

m2_error <- sqrt(mean((predict(m2, diamonds_test) - diamonds_test$price) ^ 2))



m3 <- train(price ~ carat + cut + depth + table,
            data = diamonds_train,
            method = 'nnet',
            tuneLength = 10,
            trControl = trcntrl,
            linout = TRUE)

m3$results$RMSE %>% min

m3_error <- sqrt(mean((predict(m3, diamonds_test) - diamonds_test$price) ^ 2))





df <- data.frame(method = c('enet', 'rf', 'nnet'),
                 validation = c(min(m1$results$RMSE), min(m2$results$RMSE), min(m3$results$RMSE)),
                 test = c(m1_error, m2_error, m3_error))

ggplot(df, aes(validation, test, colour = method)) + 
  geom_point() +
  geom_abline(slope = 1, intercept = 0)