R/confusion_matrix_df.R

In tidysdm: Species Distribution Models with Tidymodels

#' Make a confusion matrix dataframe for multiple thresholds
#'
#' Create the confusion matrix for multiple thresholds, using to optimise tss
#'
#' @param presences Probabilities for presences
#' @param absences probabilities for absences
#' @returns A data.frame of thresholds with columns *thres*, *tp*, *fp*, *fn*,
#'   and *tn*
#' @keywords internal


conf_matrix_df <- function(presences, absences) {
  # Threshold breaks (nice idea from flexsdm):
  if (length(presences) > 1000) {
    thresh <- as.vector(stats::quantile(presences, seq(0, 1, 0.001)))
  } else {
    thresh <- presences
  }
  if (length(absences) > 1000) {
    thresh <- c(thresh, as.vector(stats::quantile(absences, seq(0, 1, 0.001))))
  } else {
    thresh <- c(thresh, absences)
  }
  # think whether we want to round here
  thresh <- sort(unique(round(thresh, 8)))

  # confusion matrix at different thresholds, 1 flattened matrix per row
  return(data.frame(
    thresh = thresh,
    tp = unlist(lapply(thresh, FUN = function(x) sum(presences >= x))),
    fp = unlist(lapply(thresh, FUN = function(x) sum(absences >= x))),
    fn = unlist(lapply(thresh, FUN = function(x) sum(presences < x))),
    tn = unlist(lapply(thresh, FUN = function(x) sum(absences < x)))
  ))
}