R/PR.R

In weecology/NeonTreeEvaluation_package: Evaluate Individual Tree Segmentation Algorithms against Benchmark Data

#' Precision and recall for a set of prediction results
#'
#' \code{PR} is an internal function not called by the user for summary statistics
#' @param results A dataframe returned from \code{evaluate_plot}
#' @param threshold intersection-over-union threshold (0-1)
#' @return Summary recall and precision

PR <- function(results, threshold) {
  true_positives <- as.numeric(results$IoU) > threshold

  n <- results %>%
    group_by(plot_name) %>%
    distinct(total_ground, total_prediction)
  total_ground <- sum(n$total_ground)
  total_prediction <- sum(n$total_prediction)
  recall <- round(sum(true_positives, na.rm = TRUE) / total_ground, 3)
  precision <- round(sum(true_positives, na.rm = TRUE) / total_prediction, 3)
  statistic <- data.frame(recall, precision)
  return(statistic)
}