R/get_training_data.R

In reconhub/epichange: Detect Changes in Temporal Trends

#' Set training data
#'
#' This function adds a column 'training' to a `data.frame`, a `logical`
#' indicating if data points are part of the training set ('callibrartion
#' window'), or not. The last 'k' time units from a dataset to define a training
#' set. Time is known through the name of a column indicating time.
#'
#' @param x A `data.frame` or a `tibble`.
#' @param date_index The name of the variable in `df` to be used as time.
#' @param k The number of the last data points to exclude from the training
#'   set. These will be the last `k` unique values of the `time` column.
#' @noRd
set_training_data <- function(x, date_index, k) {

  if (k <= 0) {
    x$.training <- rep(TRUE, nrow(x))
    return(x)
  }

  # get time, sort unique values by decreasing order, and find the k-th value;
  # this will set the (exclusive) upper bound of data retained in the training
  # set
  dates <- x[[date_index]]
  unique_dates <- sort(unique(dates), decreasing = TRUE)
  max_date <- unique_dates[k]
  x$.training <- dates < max_date
  x
}



#' Get training data
#'
#' This function gets the set of training data defined by `set_training_data()`.
#'
#' @inheritParams set_training_data
#' @noRd
get_training_data <- function(x) {
  var_to_keep <- names(x) != ".training"
  x[x$.training, var_to_keep, drop = FALSE]
}