epclust/R/clustering.R

   1 #' @name clustering
   2 #' @rdname clustering
   3 #' @aliases clusteringTask1 computeClusters1 computeClusters2
   4 #'
   5 #' @title Two-stage clustering, withing one task (see \code{claws()})
   6 #'
   7 #' @description \code{clusteringTask1()} runs one full stage-1 task, which consists in
   8 #'   iterated stage 1 clustering (on nb_curves / ntasks energy contributions, computed
   9 #'   through discrete wavelets coefficients).
  10 #'   \code{clusteringTask2()} runs a full stage-2 task, which consists in synchrones
  11 #'   and then WER distances computations, before applying the clustering algorithm.
  12 #'   \code{computeClusters1()} and \code{computeClusters2()} correspond to the atomic
  13 #'   clustering procedures respectively for stage 1 and 2. The former applies the
  14 #'   clustering algorithm (PAM) on a contributions matrix, while the latter clusters
  15 #'   a chunk of series inside one task (~max nb_series_per_chunk)
  16 #'
  17 #' @param indices Range of series indices to cluster in parallel (initial data)
  18 #' @param getContribs Function to retrieve contributions from initial series indices:
  19 #'   \code{getContribs(indices)} outpus a contributions matrix
  20 #' @param contribs matrix of contributions (e.g. output of \code{curvesToContribs()})
  21 #' @param distances matrix of K1 x K1 (WER) distances between synchrones
  22 #' @inheritParams computeSynchrones
  23 #' @inheritParams claws
  24 #'
  25 #' @return For \code{clusteringTask1()} and \code{computeClusters1()}, the indices of the
  26 #'   computed (K1) medoids. Indices are irrelevant for stage 2 clustering, thus
  27 #'   \code{computeClusters2()} outputs a big.matrix of medoids
  28 #'   (of size limited by nb_series_per_chunk)
  29 NULL
  30
  31 #' @rdname clustering
  32 #' @export
  33 clusteringTask1 = function(
  34     indices, getContribs, K1, nb_series_per_chunk, ncores_clust=1, verbose=FALSE, parll=TRUE)
  35 {
  36     if (verbose)
  37         cat(paste("*** Clustering task 1 on ",length(indices)," lines\n", sep=""))
  38
  39     if (parll)
  40     {
  41         cl = parallel::makeCluster(ncores_clust)
  42         parallel::clusterExport(cl, varlist=c("getContribs","K1","verbose"), envir=environment())
  43     }
  44     while (length(indices) > K1)
  45     {
  46         indices_workers = .spreadIndices(indices, nb_series_per_chunk)
  47         indices <-
  48             if (parll)
  49             {
  50                 unlist( parallel::parLapply(cl, indices_workers, function(inds) {
  51                     require("epclust", quietly=TRUE)
  52                     inds[ computeClusters1(getContribs(inds), K1, verbose) ]
  53                 }) )
  54             }
  55             else
  56             {
  57                 unlist( lapply(indices_workers, function(inds)
  58                     inds[ computeClusters1(getContribs(inds), K1, verbose) ]
  59                 ) )
  60             }
  61     }
  62     if (parll)
  63         parallel::stopCluster(cl)
  64
  65     indices #medoids
  66 }
  67
  68 #' @rdname clustering
  69 #' @export
  70 clusteringTask2 = function(medoids, K2,
  71     getRefSeries, nb_ref_curves, nb_series_per_chunk, ncores_clust=1,verbose=FALSE,parll=TRUE)
  72 {
  73     if (verbose)
  74         cat(paste("*** Clustering task 2 on ",nrow(medoids)," lines\n", sep=""))
  75
  76     if (nrow(medoids) <= K2)
  77         return (medoids)
  78     synchrones = computeSynchrones(medoids,
  79         getRefSeries, nb_ref_curves, nb_series_per_chunk, ncores_clust, verbose, parll)
  80     distances = computeWerDists(synchrones, ncores_clust, verbose, parll)
  81     medoids[ computeClusters2(distances,K2,verbose), ]
  82 }
  83
  84 #' @rdname clustering
  85 #' @export
  86 computeClusters1 = function(contribs, K1, verbose=FALSE)
  87 {
  88     if (verbose)
  89         cat(paste("   computeClusters1() on ",nrow(contribs)," lines\n", sep=""))
  90     cluster::pam(contribs, K1, diss=FALSE)$id.med
  91 }
  92
  93 #' @rdname clustering
  94 #' @export
  95 computeClusters2 = function(distances, K2, verbose=FALSE)
  96 {
  97     if (verbose)
  98         cat(paste("   computeClusters2() on ",nrow(distances)," lines\n", sep=""))
  99     cluster::pam(distances, K2, diss=TRUE)$id.med
 100 }
 101
 102 #' computeSynchrones
 103 #'
 104 #' Compute the synchrones curves (sum of clusters elements) from a matrix of medoids,
 105 #' using L2 distances.
 106 #'
 107 #' @param medoids big.matrix of medoids (curves of same length as initial series)
 108 #' @param getRefSeries Function to retrieve initial series (e.g. in stage 2 after series
 109 #'   have been replaced by stage-1 medoids)
 110 #' @param nb_ref_curves How many reference series? (This number is known at this stage)
 111 #' @inheritParams claws
 112 #'
 113 #' @return A big.matrix of size K1 x L where L = data_length
 114 #'
 115 #' @export
 116 computeSynchrones = function(medoids, getRefSeries,
 117     nb_ref_curves, nb_series_per_chunk, ncores_clust=1,verbose=FALSE,parll=TRUE)
 118 {
 119     if (verbose)
 120         cat(paste("--- Compute synchrones\n", sep=""))
 121
 122     computeSynchronesChunk = function(indices)
 123     {
 124         ref_series = getRefSeries(indices)
 125         nb_series = nrow(ref_series)
 126
 127         if (parll)
 128         {
 129             require("bigmemory", quietly=TRUE)
 130             require("synchronicity", quietly=TRUE)
 131             require("epclust", quietly=TRUE)
 132             synchrones <- bigmemory::attach.big.matrix(synchrones_desc)
 133             medoids <- bigmemory::attach.big.matrix(medoids_desc)
 134             m <- synchronicity::attach.mutex(m_desc)
 135         }
 136
 137
 138
 139 #TODO: use dbs(),
 140         #https://www.r-bloggers.com/debugging-parallel-code-with-dbs/
 141         #http://gforge.se/2015/02/how-to-go-parallel-in-r-basics-tips/
 142
 143 #OK ::
 144 #write(length(indices), file="TOTO")
 145 #write( computeMedoidsIndices(medoids@address, getRefSeries(indices[1:600])), file="TOTO")
 146 #stop()
 147
 148 #       write(indices, file="TOTO", ncolumns=10, append=TRUE)
 149 #write("medoids", file = "TOTO", ncolumns=1, append=TRUE)
 150 #write(medoids[1,1:3], file = "TOTO", ncolumns=1, append=TRUE)
 151 #write("synchrones", file = "TOTO", ncolumns=1, append=TRUE)
 152 #write(synchrones[1,1:3], file = "TOTO", ncolumns=1, append=TRUE)
 153
 154 #NOT OK :: (should just be "ref_series") ...or yes ? race problems mutex then ? ?!
 155         #get medoids indices for this chunk of series
 156         mi = computeMedoidsIndices(medoids@address, getRefSeries(indices[1:600]))  #ref_series)
 157 write("MI ::::", file = "TOTO", ncolumns=1, append=TRUE)
 158 write(mi[1:3], file = "TOTO", ncolumns=1, append=TRUE)
 159
 160 #       #R-equivalent, requiring a matrix (thus potentially breaking "fit-in-memory" hope)
 161 #       mat_meds = medoids[,]
 162 #       mi = rep(NA,nb_series)
 163 #       for (i in 1:nb_series)
 164 #           mi[i] <- which.min( rowSums( sweep(mat_meds, 2, ref_series[i,], '-')^2 ) )
 165 #       rm(mat_meds); gc()
 166
 167         for (i in seq_len(nb_series))
 168         {
 169             if (parll)
 170                 synchronicity::lock(m)
 171             synchrones[mi[i],] = synchrones[mi[i],] + ref_series[i,]
 172             counts[mi[i],1] = counts[mi[i],1] + 1
 173             if (parll)
 174                 synchronicity::unlock(m)
 175         }
 176     }
 177
 178     K = nrow(medoids) ; L = ncol(medoids)
 179     # Use bigmemory (shared==TRUE by default) + synchronicity to fill synchrones in //
 180     # TODO: if size > RAM (not our case), use file-backed big.matrix
 181     synchrones = bigmemory::big.matrix(nrow=K, ncol=L, type="double", init=0.)
 182     counts = bigmemory::big.matrix(nrow=K, ncol=1, type="double", init=0)
 183     # synchronicity is only for Linux & MacOS; on Windows: run sequentially
 184     parll = (requireNamespace("synchronicity",quietly=TRUE)
 185         && parll && Sys.info()['sysname'] != "Windows")
 186     if (parll)
 187     {
 188         m <- synchronicity::boost.mutex()
 189         m_desc <- synchronicity::describe(m)
 190         synchrones_desc = bigmemory::describe(synchrones)
 191         medoids_desc = bigmemory::describe(medoids)
 192
 193         cl = parallel::makeCluster(ncores_clust)
 194         parallel::clusterExport(cl,
 195             varlist=c("synchrones_desc","counts","verbose","m_desc","medoids_desc","getRefSeries"),
 196             envir=environment())
 197     }
 198
 199     indices_workers = .spreadIndices(seq_len(nb_ref_curves), nb_series_per_chunk)
 200     ignored <-
 201         if (parll)
 202             parallel::parLapply(cl, indices_workers, computeSynchronesChunk)
 203         else
 204             lapply(indices_workers, computeSynchronesChunk)
 205
 206     if (parll)
 207         parallel::stopCluster(cl)
 208
 209     #TODO: can we avoid this loop? ( synchrones = sweep(synchrones, 1, counts, '/') )
 210     for (i in seq_len(K))
 211         synchrones[i,] = synchrones[i,] / counts[i,1]
 212     #NOTE: odds for some clusters to be empty? (when series already come from stage 2)
 213     #      ...maybe; but let's hope resulting K1' be still quite bigger than K2
 214     noNA_rows = sapply(seq_len(K), function(i) all(!is.nan(synchrones[i,])))
 215     if (all(noNA_rows))
 216         return (synchrones)
 217     # Else: some clusters are empty, need to slice synchrones
 218     synchrones[noNA_rows,]
 219 }
 220
 221 #' computeWerDists
 222 #'
 223 #' Compute the WER distances between the synchrones curves (in rows), which are
 224 #' returned (e.g.) by \code{computeSynchrones()}
 225 #'
 226 #' @param synchrones A big.matrix of synchrones, in rows. The series have same length
 227 #'   as the series in the initial dataset
 228 #' @inheritParams claws
 229 #'
 230 #' @return A matrix of size K1 x K1
 231 #'
 232 #' @export
 233 computeWerDists = function(synchrones, ncores_clust=1,verbose=FALSE,parll=TRUE)
 234 {
 235     if (verbose)
 236         cat(paste("--- Compute WER dists\n", sep=""))
 237
 238     n <- nrow(synchrones)
 239     delta <- ncol(synchrones)
 240     #TODO: automatic tune of all these parameters ? (for other users)
 241     nvoice   <- 4
 242     # noctave = 2^13 = 8192 half hours ~ 180 days ; ~log2(ncol(synchrones))
 243     noctave = 13
 244     # 4 here represent 2^5 = 32 half-hours ~ 1 day
 245     #NOTE: default scalevector == 2^(0:(noctave * nvoice) / nvoice) * s0 (?)
 246     scalevector  <- 2^(4:(noctave * nvoice) / nvoice + 1)
 247     #condition: ( log2(s0*w0/(2*pi)) - 1 ) * nvoice + 1.5 >= 1
 248     s0=2
 249     w0=2*pi
 250     scaled=FALSE
 251     s0log = as.integer( (log2( s0*w0/(2*pi) ) - 1) * nvoice + 1.5 )
 252     totnoct = noctave + as.integer(s0log/nvoice) + 1
 253
 254     Xwer_dist <- bigmemory::big.matrix(nrow=n, ncol=n, type="double")
 255     fcoefs = rep(1/3, 3) #moving average on 3 values
 256
 257     # Generate n(n-1)/2 pairs for WER distances computations
 258     pairs = list()
 259     V = seq_len(n)
 260     for (i in 1:n)
 261     {
 262         V = V[-1]
 263         pairs = c(pairs, lapply(V, function(v) c(i,v)))
 264     }
 265
 266     # Distance between rows i and j
 267     computeDistancesIJ = function(pair)
 268     {
 269         require("bigmemory", quietly=TRUE)
 270         require("epclust", quietly=TRUE)
 271         synchrones <- bigmemory::attach.big.matrix(synchrones_desc)
 272         Xwer_dist <- bigmemory::attach.big.matrix(Xwer_dist_desc)
 273
 274         computeCWT = function(i)
 275         {
 276             ts <- scale(ts(synchrones[i,]), center=TRUE, scale=scaled)
 277             totts.cwt = Rwave::cwt(ts, totnoct, nvoice, w0, plot=FALSE)
 278             ts.cwt = totts.cwt[,s0log:(s0log+noctave*nvoice)]
 279             #Normalization
 280             sqs <- sqrt(2^(0:(noctave*nvoice)/nvoice)*s0)
 281             sqres <- sweep(ts.cwt,2,sqs,'*')
 282             sqres / max(Mod(sqres))
 283         }
 284
 285         i = pair[1] ; j = pair[2]
 286         if (verbose && j==i+1)
 287             cat(paste("   Distances (",i,",",j,"), (",i,",",j+1,") ...\n", sep=""))
 288         cwt_i = computeCWT(i)
 289         cwt_j = computeCWT(j)
 290         num <- epclustFilter(Mod(cwt_i * Conj(cwt_j)))
 291         WX  <- epclustFilter(Mod(cwt_i * Conj(cwt_i)))
 292         WY  <- epclustFilter(Mod(cwt_j * Conj(cwt_j)))
 293         wer2 <- sum(colSums(num)^2) / sum(colSums(WX) * colSums(WY))
 294         Xwer_dist[i,j] <- sqrt(delta * ncol(cwt_i) * (1 - wer2))
 295         Xwer_dist[j,i] <- Xwer_dist[i,j]
 296         Xwer_dist[i,i] = 0.
 297     }
 298
 299     if (parll)
 300     {
 301         cl = parallel::makeCluster(ncores_clust)
 302         synchrones_desc <- bigmemory::describe(synchrones)
 303         Xwer_dist_desc_desc <- bigmemory::describe(Xwer_dist)
 304
 305         parallel::clusterExport(cl, varlist=c("synchrones_desc","Xwer_dist_desc","totnoct",
 306             "nvoice","w0","s0log","noctave","s0","verbose"), envir=environment())
 307     }
 308
 309     ignored <-
 310         if (parll)
 311             parallel::parLapply(cl, pairs, computeDistancesIJ)
 312         else
 313             lapply(pairs, computeDistancesIJ)
 314
 315     if (parll)
 316         parallel::stopCluster(cl)
 317
 318     Xwer_dist[n,n] = 0.
 319     distances <- Xwer_dist[,]
 320     rm(Xwer_dist) ; gc()
 321     distances #~small matrix K1 x K1
 322 }
 323
 324 # Helper function to divide indices into balanced sets
 325 .spreadIndices = function(indices, nb_per_chunk)
 326 {
 327     L = length(indices)
 328     nb_workers = floor( L / nb_per_chunk )
 329     if (nb_workers == 0)
 330     {
 331         # L < nb_series_per_chunk, simple case
 332         indices_workers = list(indices)
 333     }
 334     else
 335     {
 336         indices_workers = lapply( seq_len(nb_workers), function(i)
 337             indices[(nb_per_chunk*(i-1)+1):(nb_per_chunk*i)] )
 338         # Spread the remaining load among the workers
 339         rem = L %% nb_per_chunk
 340         while (rem > 0)
 341         {
 342             index = rem%%nb_workers + 1
 343             indices_workers[[index]] = c(indices_workers[[index]], indices[L-rem+1])
 344             rem = rem - 1
 345         }
 346     }
 347     indices_workers
 348 }