test_source/old scripts/get_neural_network.R

In opoyc/autoforecast: Forecast Development Kit

#' Fit Neural Network model
#'
#' @param .data Data frame or tibble with a response variable.
#' @param y_var String. Column name of the time series to be forecasted.
#' @param x_data Data frame. Design matrix to calculate predicted/forecast figures.
#' @param parameter List. Parameter to be used for estimation.
#'
#' @import forecast
#' @return data-frame
#' @export
#'
#' @examples
#' \dontrun{
#' get_neural_network()
#' }
get_neural_network <- function(.data, y_var, x_data = NULL, parameter = NULL){
  if(is.null(attributes(.data)[["prescription"]]) == FALSE) {
    prescription <- attributes(.data)[["prescription"]]
    y_var <- prescription$y_var
    date_var <- prescription$date_var
    freq <- prescription$freq
    na_exclude <- unique(c(prescription$key, y_var, date_var))
  }
  y_var_int <- ts(.data[[y_var]], frequency = freq) # maybe not optimal
  model_fit <- nnetar(y = y_var_int)
  # Output
  .fit_output <- list(model = "neural_network"
                      , model_fit = model_fit
                      , y_var_pred = as.numeric(model_fit$fitted)
                      , parameter = as.numeric(unlist(str_extract_all(model_fit$method, "[0-9]{1,2}")
                                                      )
                                               )
                      )
  
  attr(.fit_output, "prescription") <- prescription
  class(.fit_output) <- ".fit_output"
  
  return(.fit_output)
}