epclust/R/clustering.R

   1 #' Two-stage clustering, within one task (see \code{claws()})
   2 #'
   3 #' \code{clusteringTask1()} runs one full stage-1 task, which consists in iterated
   4 #' stage 1 clustering on nb_curves / ntasks energy contributions, computed through
   5 #' discrete wavelets coefficients.
   6 #' \code{clusteringTask2()} runs a full stage-2 task, which consists in WER distances
   7 #' computations between medoids (indices) output from stage 1, before applying
   8 #' the second clustering algorithm on the distances matrix.
   9 #'
  10 #' @param getContribs Function to retrieve contributions from initial series indices:
  11 #'   \code{getContribs(indices)} outputs a contributions matrix
  12 #' @inheritParams claws
  13 #' @inheritParams computeSynchrones
  14 #' @inheritParams computeWerDists
  15 #'
  16 #' @return The indices of the computed (resp. K1 and K2) medoids.
  17 #'
  18 #' @name clustering
  19 #' @rdname clustering
  20 #' @aliases clusteringTask1 clusteringTask2
  21 NULL
  22
  23 #' @rdname clustering
  24 #' @export
  25 clusteringTask1 = function(indices, getContribs, K1, algoClust1, nb_items_clust,
  26     ncores_clust=1, verbose=FALSE, parll=TRUE)
  27 {
  28     if (parll)
  29     {
  30         cl = parallel::makeCluster(ncores_clust, outfile = "")
  31         parallel::clusterExport(cl, c("getContribs","K1","verbose"), envir=environment())
  32     }
  33     # Iterate clustering algorithm 1 until K1 medoids are found
  34     while (length(indices) > K1)
  35     {
  36         # Balance tasks by splitting the indices set - as evenly as possible
  37         indices_workers = .splitIndices(indices, nb_items_clust, min_size=K1+1)
  38         if (verbose)
  39             cat(paste("*** [iterated] Clustering task 1 on ",length(indices)," series\n", sep=""))
  40         indices <-
  41             if (parll)
  42             {
  43                 unlist( parallel::parLapply(cl, indices_workers, function(inds) {
  44                     require("epclust", quietly=TRUE)
  45                     inds[ algoClust1(getContribs(inds), K1) ]
  46                 }) )
  47             }
  48             else
  49             {
  50                 unlist( lapply(indices_workers, function(inds)
  51                     inds[ algoClust1(getContribs(inds), K1) ]
  52                 ) )
  53             }
  54     }
  55     if (parll)
  56         parallel::stopCluster(cl)
  57
  58     indices #medoids
  59 }
  60
  61 #' @rdname clustering
  62 #' @export
  63 clusteringTask2 = function(indices, getSeries, K2, algoClust2, nb_series_per_chunk,
  64     nvoice, nbytes, endian, ncores_clust=1, verbose=FALSE, parll=TRUE)
  65 {
  66     if (verbose)
  67         cat(paste("*** Clustering task 2 on ",length(indices)," medoids\n", sep=""))
  68
  69     if (length(indices) <= K2)
  70         return (indices)
  71
  72     # A) Compute the WER distances (Wavelets Extended coefficient of deteRmination)
  73     distances = computeWerDists(indices, getSeries, nb_series_per_chunk,
  74         nvoice, nbytes, endian, ncores_clust, verbose, parll)
  75
  76     # B) Apply clustering algorithm 2 on the WER distances matrix
  77     if (verbose)
  78         cat(paste("*** algoClust2() on ",nrow(distances)," items\n", sep=""))
  79     indices[ algoClust2(distances,K2) ]
  80 }