epclust/R/clustering.R

   1 #' @name clustering
   2 #' @rdname clustering
   3 #' @aliases clusteringTask1 clusteringTask2 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 #'   first clustering algorithm on a contributions matrix, while the latter clusters
  15 #'   a set of series inside one task (~nb_items_clust1)
  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 #' @inheritParams computeSynchrones
  21 #' @inheritParams claws
  22 #'
  23 #' @return For \code{clusteringTask1()}, the indices of the computed (K1) medoids.
  24 #'   Indices are irrelevant for stage 2 clustering, thus \code{clusteringTask2()}
  25 #'   outputs a big.matrix of medoids (of size LxK2, K2 = final number of clusters)
  26 NULL
  27
  28 #' @rdname clustering
  29 #' @export
  30 clusteringTask1 = function(indices, getContribs, K1, algoClust1, nb_items_clust1,
  31     ncores_clust=1, verbose=FALSE, parll=TRUE)
  32 {
  33     if (parll)
  34     {
  35         cl = parallel::makeCluster(ncores_clust, outfile = "")
  36         parallel::clusterExport(cl, c("getContribs","K1","verbose"), envir=environment())
  37     }
  38     # Iterate clustering algorithm 1 until K1 medoids are found
  39     while (length(indices) > K1)
  40     {
  41         # Balance tasks by splitting the indices set - as evenly as possible
  42         indices_workers = .spreadIndices(indices, nb_items_clust1)
  43         if (verbose)
  44             cat(paste("*** [iterated] Clustering task 1 on ",length(indices)," series\n", sep=""))
  45         indices <-
  46             if (parll)
  47             {
  48                 unlist( parallel::parLapply(cl, indices_workers, function(inds) {
  49                     require("epclust", quietly=TRUE)
  50                     inds[ algoClust1(getContribs(inds), K1) ]
  51                 }) )
  52             }
  53             else
  54             {
  55                 unlist( lapply(indices_workers, function(inds)
  56                     inds[ algoClust1(getContribs(inds), K1) ]
  57                 ) )
  58             }
  59     }
  60     if (parll)
  61         parallel::stopCluster(cl)
  62
  63     indices #medoids
  64 }
  65
  66 #' @rdname clustering
  67 #' @export
  68 clusteringTask2 = function(medoids, K2, algoClust2, getRefSeries, nb_ref_curves,
  69     nb_series_per_chunk, nvoice, nbytes,endian,ncores_clust=1,verbose=FALSE,parll=TRUE)
  70 {
  71     if (verbose)
  72         cat(paste("*** Clustering task 2 on ",ncol(medoids)," synchrones\n", sep=""))
  73
  74     if (ncol(medoids) <= K2)
  75         return (medoids)
  76
  77     # A) Obtain synchrones, that is to say the cumulated power consumptions
  78     #    for each of the K1 initial groups
  79     synchrones = computeSynchrones(medoids, getRefSeries, nb_ref_curves,
  80         nb_series_per_chunk, ncores_clust, verbose, parll)
  81
  82     # B) Compute the WER distances (Wavelets Extended coefficient of deteRmination)
  83     distances = computeWerDists(
  84         synchrones, nvoice, nbytes, endian, ncores_clust, verbose, parll)
  85
  86     # C) Apply clustering algorithm 2 on the WER distances matrix
  87     if (verbose)
  88         cat(paste("*** algoClust2() on ",nrow(distances)," items\n", sep=""))
  89     medoids[ ,algoClust2(distances,K2) ]
  90 }
  91
  92 #' computeSynchrones
  93 #'
  94 #' Compute the synchrones curves (sum of clusters elements) from a matrix of medoids,
  95 #' using euclidian distance.
  96 #'
  97 #' @param medoids big.matrix of medoids (curves of same length as initial series)
  98 #' @param getRefSeries Function to retrieve initial series (e.g. in stage 2 after series
  99 #'   have been replaced by stage-1 medoids)
 100 #' @param nb_ref_curves How many reference series? (This number is known at this stage)
 101 #' @inheritParams claws
 102 #'
 103 #' @return A big.matrix of size L x K1 where L = length of a serie
 104 #'
 105 #' @export
 106 computeSynchrones = function(medoids, getRefSeries, nb_ref_curves,
 107     nb_series_per_chunk, ncores_clust=1,verbose=FALSE,parll=TRUE)
 108 {
 109     # Synchrones computation is embarassingly parallel: compute it by chunks of series
 110     computeSynchronesChunk = function(indices)
 111     {
 112         if (parll)
 113         {
 114             require("bigmemory", quietly=TRUE)
 115             requireNamespace("synchronicity", quietly=TRUE)
 116             require("epclust", quietly=TRUE)
 117             # The big.matrix objects need to be attached to be usable on the workers
 118             synchrones <- bigmemory::attach.big.matrix(synchrones_desc)
 119             medoids <- bigmemory::attach.big.matrix(medoids_desc)
 120             m <- synchronicity::attach.mutex(m_desc)
 121         }
 122
 123         # Obtain a chunk of reference series
 124         ref_series = getRefSeries(indices)
 125         nb_series = ncol(ref_series)
 126
 127         # Get medoids indices for this chunk of series
 128         mi = computeMedoidsIndices(medoids@address, ref_series)
 129
 130         # Update synchrones using mi above
 131         for (i in seq_len(nb_series))
 132         {
 133             if (parll)
 134                 synchronicity::lock(m) #locking required because several writes at the same time
 135             synchrones[, mi[i] ] = synchrones[, mi[i] ] + ref_series[,i]
 136             if (parll)
 137                 synchronicity::unlock(m)
 138         }
 139         NULL
 140     }
 141
 142     K = ncol(medoids) ; L = nrow(medoids)
 143     # Use bigmemory (shared==TRUE by default) + synchronicity to fill synchrones in //
 144     synchrones = bigmemory::big.matrix(nrow=L, ncol=K, type="double", init=0.)
 145     # NOTE: synchronicity is only for Linux & MacOS; on Windows: run sequentially
 146     parll = (parll && requireNamespace("synchronicity",quietly=TRUE)
 147         && Sys.info()['sysname'] != "Windows")
 148     if (parll)
 149     {
 150         m <- synchronicity::boost.mutex() #for lock/unlock, see computeSynchronesChunk
 151         # mutex and big.matrix objects cannot be passed directly:
 152         # they will be accessed from their description
 153         m_desc <- synchronicity::describe(m)
 154         synchrones_desc = bigmemory::describe(synchrones)
 155         medoids_desc = bigmemory::describe(medoids)
 156         cl = parallel::makeCluster(ncores_clust)
 157         parallel::clusterExport(cl, envir=environment(),
 158             varlist=c("synchrones_desc","m_desc","medoids_desc","getRefSeries"))
 159     }
 160
 161     if (verbose)
 162         cat(paste("--- Compute ",K," synchrones with ",nb_ref_curves," series\n", sep=""))
 163
 164     # Balance tasks by splitting the indices set - maybe not so evenly, but
 165     # max==TRUE in next call ensures that no set has more than nb_series_per_chunk items.
 166     indices_workers = .spreadIndices(seq_len(nb_ref_curves), nb_series_per_chunk, max=TRUE)
 167     ignored <-
 168         if (parll)
 169             parallel::parLapply(cl, indices_workers, computeSynchronesChunk)
 170         else
 171             lapply(indices_workers, computeSynchronesChunk)
 172
 173     if (parll)
 174         parallel::stopCluster(cl)
 175
 176     return (synchrones)
 177 }
 178
 179 #' computeWerDists
 180 #'
 181 #' Compute the WER distances between the synchrones curves (in columns), which are
 182 #' returned (e.g.) by \code{computeSynchrones()}
 183 #'
 184 #' @param synchrones A big.matrix of synchrones, in columns. The series have same
 185 #'   length as the series in the initial dataset
 186 #' @inheritParams claws
 187 #'
 188 #' @return A distances matrix of size K1 x K1
 189 #'
 190 #' @export
 191 computeWerDists = function(synchrones, nvoice, nbytes,endian,ncores_clust=1,
 192     verbose=FALSE,parll=TRUE)
 193 {
 194     n <- ncol(synchrones)
 195     L <- nrow(synchrones)
 196     noctave = ceiling(log2(L)) #min power of 2 to cover serie range
 197
 198     # Initialize result as a square big.matrix of size 'number of synchrones'
 199     Xwer_dist <- bigmemory::big.matrix(nrow=n, ncol=n, type="double")
 200
 201     # Generate n(n-1)/2 pairs for WER distances computations
 202     pairs = list()
 203     V = seq_len(n)
 204     for (i in 1:n)
 205     {
 206         V = V[-1]
 207         pairs = c(pairs, lapply(V, function(v) c(i,v)))
 208     }
 209
 210     cwt_file = ".cwt.bin"
 211     # Compute the synchrones[,index] CWT, and store it in the binary file above
 212     computeSaveCWT = function(index)
 213     {
 214         if (parll && !exists(synchrones)) #avoid going here after first call on a worker
 215         {
 216             require("bigmemory", quietly=TRUE)
 217             require("Rwave", quietly=TRUE)
 218             require("epclust", quietly=TRUE)
 219             synchrones <- bigmemory::attach.big.matrix(synchrones_desc)
 220         }
 221         ts <- scale(ts(synchrones[,index]), center=TRUE, scale=FALSE)
 222         ts_cwt = Rwave::cwt(ts, noctave, nvoice, w0=2*pi, twoD=TRUE, plot=FALSE)
 223
 224         # Serialization
 225         binarize(as.matrix(c(as.double(Re(ts_cwt)),as.double(Im(ts_cwt)))), cwt_file, 1,
 226             ",", nbytes, endian)
 227     }
 228
 229     if (parll)
 230     {
 231         cl = parallel::makeCluster(ncores_clust)
 232         synchrones_desc <- bigmemory::describe(synchrones)
 233         Xwer_dist_desc <- bigmemory::describe(Xwer_dist)
 234         parallel::clusterExport(cl, varlist=c("parll","synchrones_desc","Xwer_dist_desc",
 235             "noctave","nvoice","verbose","getCWT"), envir=environment())
 236     }
 237
 238     if (verbose)
 239         cat(paste("--- Precompute and serialize synchrones CWT\n", sep=""))
 240
 241     ignored <-
 242         if (parll)
 243             parallel::parLapply(cl, 1:n, computeSaveCWT)
 244         else
 245             lapply(1:n, computeSaveCWT)
 246
 247     # Function to retrieve a synchrone CWT from (binary) file
 248     getSynchroneCWT = function(index, L)
 249     {
 250         flat_cwt <- getDataInFile(index, cwt_file, nbytes, endian)
 251         cwt_length = length(flat_cwt) / 2
 252         re_part = as.matrix(flat_cwt[1:cwt_length], nrow=L)
 253         im_part = as.matrix(flat_cwt[(cwt_length+1):(2*cwt_length)], nrow=L)
 254         re_part + 1i * im_part
 255     }
 256
 257     # Compute distance between columns i and j in synchrones
 258     computeDistanceIJ = function(pair)
 259     {
 260         if (parll)
 261         {
 262             # parallel workers start with an empty environment
 263             require("bigmemory", quietly=TRUE)
 264             require("epclust", quietly=TRUE)
 265             synchrones <- bigmemory::attach.big.matrix(synchrones_desc)
 266             Xwer_dist <- bigmemory::attach.big.matrix(Xwer_dist_desc)
 267         }
 268
 269         i = pair[1] ; j = pair[2]
 270         if (verbose && j==i+1 && !parll)
 271             cat(paste("   Distances (",i,",",j,"), (",i,",",j+1,") ...\n", sep=""))
 272
 273         # Compute CWT of columns i and j in synchrones
 274         L = nrow(synchrones)
 275         cwt_i <- getSynchroneCWT(i, L)
 276         cwt_j <- getSynchroneCWT(j, L)
 277
 278         # Compute the ratio of integrals formula 5.6 for WER^2
 279         # in https://arxiv.org/abs/1101.4744v2 §5.3
 280         num <- filterMA(Mod(cwt_i * Conj(cwt_j)))
 281         WX  <- filterMA(Mod(cwt_i * Conj(cwt_i)))
 282         WY <- filterMA(Mod(cwt_j * Conj(cwt_j)))
 283         wer2 <- sum(colSums(num)^2) / sum(colSums(WX) * colSums(WY))
 284
 285         Xwer_dist[i,j] <- sqrt(L * ncol(cwt_i) * (1 - wer2))
 286         Xwer_dist[j,i] <- Xwer_dist[i,j]
 287         Xwer_dist[i,i] <- 0.
 288     }
 289
 290     if (verbose)
 291         cat(paste("--- Compute WER distances\n", sep=""))
 292
 293     ignored <-
 294         if (parll)
 295             parallel::parLapply(cl, pairs, computeDistanceIJ)
 296         else
 297             lapply(pairs, computeDistanceIJ)
 298
 299     if (parll)
 300         parallel::stopCluster(cl)
 301
 302     unlink(cwt_file)
 303
 304     Xwer_dist[n,n] = 0.
 305     Xwer_dist[,] #~small matrix K1 x K1
 306 }
 307
 308 # Helper function to divide indices into balanced sets
 309 # If max == TRUE, sets sizes cannot exceed nb_per_set
 310 .spreadIndices = function(indices, nb_per_set, max=FALSE)
 311 {
 312     L = length(indices)
 313     nb_workers = floor( L / nb_per_set )
 314     rem = L %% nb_per_set
 315     if (nb_workers == 0 || (nb_workers==1 && rem==0))
 316     {
 317         # L <= nb_per_set, simple case
 318         indices_workers = list(indices)
 319     }
 320     else
 321     {
 322         indices_workers = lapply( seq_len(nb_workers), function(i)
 323             indices[(nb_per_set*(i-1)+1):(nb_per_set*i)] )
 324
 325         if (max)
 326         {
 327             # Sets are not so well balanced, but size is supposed to be critical
 328             return ( c( indices_workers, if (rem>0) list((L-rem+1):L) else NULL ) )
 329         }
 330
 331         # Spread the remaining load among the workers
 332         rem = L %% nb_per_set
 333         while (rem > 0)
 334         {
 335             index = rem%%nb_workers + 1
 336             indices_workers[[index]] = c(indices_workers[[index]], indices[L-rem+1])
 337             rem = rem - 1
 338         }
 339     }
 340     indices_workers
 341 }