R/method_nnd.R

In blocking: Various Blocking Methods for Entity Resolution

#'
#' @importFrom rnndescent rnnd_build
#' @importFrom rnndescent rnnd_query
#' @importFrom data.table data.table
#'
#' @title An internal function to use the NN descent algorithm via the \link[rnndescent]{rnndescent} package.
#' @author Maciej Beręsewicz
#'
#' @param x deduplication or reference data,
#' @param y query data,
#' @param k number of neighbours to return,
#' @param distance 	type of distance to calculate,
#' @param deduplication whether the deduplication is applied,
#' @param verbose if TRUE, log messages to the console,
#' @param n_threads maximum number of threads to use,
#' @param control controls for the NN descent algorithm.
#'
#' @description
#' See details of \link[rnndescent]{rnnd_build} and \link[rnndescent]{rnnd_query}.
#'
#' @keywords internal
method_nnd <- function(x,
                       y,
                       k,
                       distance,
                       deduplication,
                       verbose,
                       n_threads,
                       control,
                       seed) {

  set.seed(seed)
  l_ind <- rnndescent::rnnd_build(data = x,
                                  k = if (nrow(x) < control$nnd$k_build) nrow(x) else control$nnd$k_build,
                                  metric = distance,
                                  verbose = verbose,
                                  n_threads = n_threads,
                                  use_alt_metric = control$nnd$use_alt_metric,
                                  init = control$nnd$init,
                                  n_trees = control$nnd$n_trees,
                                  leaf_size = control$nnd$leaf_size,
                                  max_tree_depth = control$nnd$max_tree_depth,
                                  margin = control$nnd$margin,
                                  n_iters = control$nnd$n_iters,
                                  delta = control$nnd$delta,
                                  max_candidates = control$nnd$max_candidates,
                                  low_memory = control$nnd$low_memory,
                                  n_search_trees = control$nnd$n_search_trees,
                                  pruning_degree_multiplier = control$nnd$pruning_degree_multiplier,
                                  diversify_prob = control$nnd$diversify_prob,
                                  weight_by_degree = control$nnd$weight_by_degree,
                                  prune_reverse = control$nnd$prune_reverse,
                                  progress = control$nnd$progress,
                                  obs = control$nnd$obs)

  l_1nn <- rnndescent::rnnd_query(index = l_ind,
                                  query = y,
                                  k = if (nrow(x) < control$k_search) nrow(x) else control$k_search,
                                  epsilon = control$nnd$epsilon,
                                  max_search_fraction = control$nnd$max_search_fraction,
                                  init = NULL,
                                  verbose = verbose,
                                  n_threads = n_threads,
                                  obs = control$nnd$obs)

  l_df <- data.table::data.table(y = 1:NROW(y),
                                 x = l_1nn$idx[, k],
                                 dist = l_1nn$dist[,k])


  l_df
}