R/edge_nn.R

In spatsoc: Group Animal Relocation Data by Spatial and Temporal Relationship

#' Nearest neighbour based edge lists
#'
#'
#' \code{edge_nn} returns edge lists defined by the nearest neighbour. The
#' function accepts a \code{data.table} with relocation data, individual
#' identifiers and a threshold argument. The threshold argument is used to
#' specify the criteria for distance between points which defines a group.
#' Relocation data should be in two columns representing the X and Y
#' coordinates.
#'
#'
#' The \code{DT} must be a \code{data.table}. If your data is a
#' \code{data.frame}, you can convert it by reference using
#' \code{\link[data.table:setDT]{data.table::setDT}}.
#'
#' The \code{id}, \code{coords}, \code{timegroup}  (and optional \code{splitBy})
#' arguments expect the names of a column in \code{DT} which correspond to the
#' individual identifier, X and Y coordinates, timegroup (generated by
#' \code{group_times}) and additional grouping columns.
#'
#' The \code{threshold} must be provided in the units of the coordinates. The
#' \code{threshold} must be larger than 0. The coordinates must be planar
#' coordinates (e.g.: UTM). In the case of UTM, a \code{threshold} = 50 would
#' indicate a 50m distance threshold.
#'
#' The \code{timegroup} argument is required to define the temporal groups
#' within which edge nearest neighbours are calculated. The intended framework
#' is to group rows temporally with \code{\link{group_times}} then spatially
#' with \code{edge_nn}. If you have already calculated temporal groups without
#' \code{\link{group_times}}, you can pass this column to the \code{timegroup}
#' argument. Note that the expectation is that each individual will be observed
#' only once per timegroup. Caution that accidentally including huge numbers of
#' rows within timegroups can overload your machine since all pairwise distances
#' are calculated within each timegroup.
#'
#' The \code{splitBy} argument offers further control over grouping. If within
#' your \code{DT}, you have multiple populations, subgroups or other distinct
#' parts, you can provide the name of the column which identifies them to
#' \code{splitBy}. \code{edge_nn} will only consider rows within each
#' \code{splitBy} subgroup.
#'
#' @param threshold (optional) spatial distance threshold to set maximum
#'   distance between an individual and their neighbour.
#' @param returnDist boolean indicating if the distance between individuals
#'   should be returned. If FALSE (default), only ID, NN columns (and timegroup,
#'   splitBy columns if provided) are returned. If TRUE, another column
#'   "distance" is returned indicating the distance between ID and NN.
#' @inheritParams group_pts
#'
#' @return \code{edge_nn} returns a \code{data.table}  with three columns:
#'   timegroup, ID and NN. If 'returnDist' is TRUE, column 'distance' is
#'   returned indicating the distance between ID and NN.
#'
#'   The ID and NN columns represent the edges defined by the nearest neighbours
#'   (and temporal thresholds with \code{group_times}).
#'
#'   If an individual was alone in a timegroup or splitBy, or did not have any
#'   neighbours within the threshold distance, they are assigned NA for nearest
#'   neighbour.
#'
#' @export
#'
#' @family Edge-list generation
#'
#' @examples
#' # Load data.table
#' library(data.table)
#' \dontshow{data.table::setDTthreads(1)}
#'
#' # Read example data
#' DT <- fread(system.file("extdata", "DT.csv", package = "spatsoc"))
#'
#' # Select only individuals A, B, C for this example
#' DT <- DT[ID %in% c('A', 'B', 'C')]
#'
#' # Cast the character column to POSIXct
#' DT[, datetime := as.POSIXct(datetime, tz = 'UTC')]
#'
#' # Temporal grouping
#' group_times(DT, datetime = 'datetime', threshold = '20 minutes')
#'
#' # Edge list generation
#' edges <- edge_nn(DT, id = 'ID', coords = c('X', 'Y'),
#'         timegroup = 'timegroup')
#'
#' # Edge list generation using maximum distance threshold
#' edges <- edge_nn(DT, id = 'ID', coords = c('X', 'Y'),
#'         timegroup = 'timegroup', threshold = 100)
#'
#' # Edge list generation, returning distance between nearest neighbours
#' edge_nn(DT, id = 'ID', coords = c('X', 'Y'),
#'         timegroup = 'timegroup', threshold = 100,
#'         returnDist = TRUE)
#'
edge_nn <- function(DT = NULL,
                    id = NULL,
                    coords = NULL,
                    timegroup,
                    splitBy = NULL,
                    threshold = NULL,
                    returnDist = FALSE) {
  # NSE
  N <- NULL

  if (is.null(DT)) {
    stop('input DT required')
  }

  if (!is.null(threshold)) {
    if (!is.numeric(threshold)) {
      stop('threshold must be numeric')
    }
    if (threshold <= 0) {
      stop('threshold must be greater than 0')
    }
  }


  if (is.null(id)) {
    stop('ID field required')
  }

  if (length(coords) != 2) {
    stop('coords requires a vector of column names for coordinates X and Y')
  }

  if (missing(timegroup)) {
    stop('timegroup required')
  }

  if (any(!(
    c(timegroup, id, coords, splitBy) %in% colnames(DT)
  ))) {
    stop(paste0(
      as.character(paste(setdiff(
        c(timegroup, id, coords, splitBy),
        colnames(DT)
      ), collapse = ', ')),
      ' field(s) provided are not present in input DT'
    ))
  }

  if (any(!(DT[, vapply(.SD, is.numeric, TRUE), .SDcols = coords]))) {
    stop('coords must be numeric')
  }

  if (!is.null(timegroup)) {
    if (any(unlist(lapply(DT[, .SD, .SDcols = timegroup], class)) %in%
            c('POSIXct', 'POSIXlt', 'Date', 'IDate', 'ITime', 'character'))) {
      warning(
        strwrap(
          prefix = " ",
          initial = "",
          x = 'timegroup provided is a date/time
          or character type, did you use group_times?'
        )
      )
    }
  }

  if ('splitBy' %in% colnames(DT)) {
    warning(
      strwrap(x = 'a column named "splitBy" was found in your data.table,
              renamed to "split_by" to avoid confusion with the argument
              "splitBy"')
    )
    setnames(DT, 'splitBy', 'split_by')
  }


  if (is.null(timegroup) && is.null(splitBy)) {
    splitBy <- NULL
  } else {
    splitBy <- c(splitBy, timegroup)
    if (DT[, .N, by = c(id, splitBy, timegroup)][N > 1, sum(N)] != 0) {
      warning(
        strwrap(
          prefix = " ",
          initial = "",
          x = 'found duplicate id in a
          timegroup and/or splitBy -
          does your group_times threshold match the fix rate?'
        )
      )
    }
  }

  DT[, {

    distMatrix <-
      as.matrix(stats::dist(.SD[, 2:3], method = 'euclidean'))
    diag(distMatrix) <- NA

    if (is.null(threshold)) {
      wm <- apply(distMatrix, MARGIN = 2, which.min)
    } else {
      distMatrix[distMatrix > threshold] <- NA
      wm <- apply(distMatrix, MARGIN = 2,
                  function(x) ifelse(sum(!is.na(x)) > 0, which.min(x), NA))
    }

    if (returnDist) {
      w <- wm + (length(wm) * (as.numeric(names(wm)) - 1))
      l <- list(ID = .SD[[1]][as.numeric(names(wm))],
                NN = .SD[[1]][wm],
                distance = distMatrix[w])
    } else {
      l <- list(ID = .SD[[1]][as.numeric(names(wm))],
                NN = .SD[[1]][wm])
    }
    l
  },
  by = splitBy, .SDcols = c(id, coords)]
}