R/tidy_output.R

In bmsasilva/soundClass: Sound Classification Using Convolutional Neural Networks

#' Format the classification table
#' @title Format the classification table
#' @description  This function eliminates calls: 1) classified as noise,
#' and 2) closer together than a specific threshold. It
#' then converts the numeric classes to text labels
#' @param output output of function classify_calls()
#' @param class_labels vector of character labels for the categories.
#' Must be in the same order as the numeric equivalent. For instance,
#' if the numeric categories are 0 = noise and 1 = bat, the vector must
#' be c("noise", "bat")
#' @param min_dist minimum distance in ms between consecutive calls.
#' @param remove_noise if TRUE remove the peaks identified as noise
#' from the formatted table
#' @param multiple_species if FALSE only the calls belonging to the
#' most represented species are kept in the formatted table. If TRUE all
#' calls classified as bats are returned in the formatted table
#' @usage tidy_output(output, class_labels, min_dist = 40, remove_noise = T,
#'                    multiple_species = F, fs, tx)
#' @return A formated classification table
#' @author Bruno Silva
#' @keywords internal
#' @noRd
tidy_output <- function(output, class_labels, min_dist = 40,
                        remove_noise = T, multiple_species = F, fs, tx) {
  try({
    if (remove_noise == T) {
      output <- output[output$spe > 0, ]
    }

    class_n <- seq(class_labels) - 1
    colnames(output)[1:length(class_n)] <- class_labels

    output$spe <- as.character(factor(output$spe,
      levels = c(class_n),
      labels = c(class_labels)
    ))

    if (multiple_species == F) {
      output <- output[output$spe == get_mode(output$spe), ]
    }

    intervals <- diff(output$peaks)
    filter <- which(intervals < (ms2samples(min_dist, fs = fs, tx = tx))) + 1

    if (length(filter) > 0) {
      output <- output[-filter, ]
    }

    return(output)
  })
}