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 #' 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, in columns
  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=3, verbose=FALSE)
  27 {
  28     if (verbose)
  29         cat(paste("*** Clustering task 1 on ",length(indices)," series [start]\n", sep=""))
  30
  31     if (length(indices) <= K1)
  32         return (indices)
  33
  34     parll <- (ncores_clust > 1)
  35     if (parll)
  36     {
  37         # outfile=="" to see stderr/stdout on terminal
  38         cl <-
  39             if (verbose)
  40                 parallel::makeCluster(ncores_clust, outfile = "")
  41             else
  42                 parallel::makeCluster(ncores_clust)
  43         parallel::clusterExport(cl, c("getContribs","K1","verbose"), envir=environment())
  44     }
  45     # Iterate clustering algorithm 1 until K1 medoids are found
  46     while (length(indices) > K1)
  47     {
  48         # Balance tasks by splitting the indices set - as evenly as possible
  49         indices_workers <- .splitIndices(indices, nb_items_clust, min_size=K1+1)
  50         indices <-
  51             if (parll)
  52             {
  53                 unlist( parallel::parLapply(cl, indices_workers, function(inds) {
  54                     require("epclust", quietly=TRUE)
  55                     inds[ algoClust1(getContribs(inds), K1) ]
  56                 }) )
  57             }
  58             else
  59             {
  60                 unlist( lapply(indices_workers, function(inds)
  61                     inds[ algoClust1(getContribs(inds), K1) ]
  62                 ) )
  63             }
  64         if (verbose)
  65         {
  66             cat(paste("*** Clustering task 1 on ",length(indices)," medoids [iter]\n", sep=""))
  67         }
  68     }
  69     if (parll)
  70         parallel::stopCluster(cl)
  71
  72     indices #medoids
  73 }
  74
  75 #' @rdname clustering
  76 #' @export
  77 clusteringTask2 <- function(indices, getSeries, K2, algoClust2, nb_series_per_chunk,
  78     smooth_lvl, nvoice, nbytes, endian, ncores_clust=3, verbose=FALSE)
  79 {
  80     if (verbose)
  81         cat(paste("*** Clustering task 2 on ",length(indices)," medoids\n", sep=""))
  82
  83     if (length(indices) <= K2)
  84         return (indices)
  85
  86     # A) Compute the WER distances (Wavelets Extended coefficient of deteRmination)
  87     distances <- computeWerDists(indices, getSeries, nb_series_per_chunk,
  88         smooth_lvl, nvoice, nbytes, endian, ncores_clust, verbose)
  89
  90     # B) Apply clustering algorithm 2 on the WER distances matrix
  91     if (verbose)
  92         cat(paste("*** algoClust2() on ",nrow(distances)," items\n", sep=""))
  93     indices[ algoClust2(distances,K2) ]
  94 }