gistfile1.R

library(fpp3)
state_tourism <- tourism |>
  group_by(State) |>
  summarise(Trips = sum(Trips))
training <- state_tourism |>
  filter(year(Quarter) <= 2014)
fit <- training |>
  model(
    ets = ETS(Trips),
    arima = ARIMA(Trips),
    snaive = SNAIVE(Trips)
  )
fc <- fit |>
  forecast(h = "3 years")
fc_accuracy <- accuracy(fc, state_tourism,
  measures = list(
    point_accuracy_measures,
    interval_accuracy_measures,
    distribution_accuracy_measures
  )
)

model_performance <- fc_accuracy |>
  group_by(State, .model) |>
  summarise(
    RMSE = mean(RMSE),
    MAE = mean(MAE),
    MAPE = mean(MAPE),
    Winkler = mean(winkler),
    CRPS = mean(CRPS)
  ) |>
  ungroup() |>
  arrange(MAPE)

#build table of best models based on MAE
best_models <- model_performance |>
  group_by(State) |>
  filter(MAE==min(MAE)) |>
  select(State, .model)

best_fit <- fit |>
  left_join(best_models, by = join_by(State)) 

best_fit <- best_fit |> 
  mutate(best_model = purrr::map2(row_number(), .model, ~best_fit[[.y]][[.x]]))
class(best_fit$best_model) <- class(best_fit$ets)
best_fit <- best_fit |>
  select(State, best_model)

best_fit |> 
  forecast(h = "1 year")