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). \code{computeClusters1()} and
  10 #'   \code{computeClusters2()} correspond to the atomic clustering procedures respectively
  11 #'   for stage 1 and 2. The former applies the clustering algorithm (PAM) on a
  12 #'   contributions matrix, while the latter clusters a chunk of series inside one task
  13 #'   (~max nb_series_per_chunk)
  14 #'
  15 #' @param indices Range of series indices to cluster in parallel (initial data)
  16 #' @param getContribs Function to retrieve contributions from initial series indices:
  17 #'   \code{getContribs(indices)} outpus a contributions matrix
  18 #' @param contribs matrix of contributions (e.g. output of \code{curvesToContribs()})
  19 #' @inheritParams computeSynchrones
  20 #' @inheritParams claws
  21 #'
  22 #' @return For \code{clusteringTask1()} and \code{computeClusters1()}, the indices of the
  23 #'   computed (K1) medoids. Indices are irrelevant for stage 2 clustering, thus
  24 #'   \code{computeClusters2()} outputs a matrix of medoids
  25 #'   (of size limited by nb_series_per_chunk)
  26 NULL
  27
  28 #' @rdname clustering
  29 #' @export
  30 clusteringTask1 = function(
  31     indices, getContribs, K1, nb_series_per_chunk, ncores_clust=1, verbose=FALSE, parll=TRUE)
  32 {
  33     if (verbose)
  34         cat(paste("*** Clustering task on ",length(indices)," lines\n", sep=""))
  35
  36     wrapComputeClusters1 = function(inds) {
  37         if (parll)
  38             require("epclust", quietly=TRUE)
  39         if (verbose)
  40             cat(paste("   computeClusters1() on ",length(inds)," lines\n", sep=""))
  41         inds[ computeClusters1(getContribs(inds), K1) ]
  42     }
  43
  44     if (parll)
  45     {
  46         cl = parallel::makeCluster(ncores_clust)
  47         parallel::clusterExport(cl, varlist=c("getContribs","K1","verbose"), envir=environment())
  48     }
  49     while (length(indices) > K1)
  50     {
  51         indices_workers = .spreadIndices(indices, nb_series_per_chunk)
  52         if (parll)
  53             indices = unlist( parallel::parLapply(cl, indices_workers, wrapComputeClusters1) )
  54         else
  55             indices = unlist( lapply(indices_workers, wrapComputeClusters1) )
  56     }
  57     if (parll)
  58         parallel::stopCluster(cl)
  59
  60     indices #medoids
  61 }
  62
  63 #' @rdname clustering
  64 #' @export
  65 computeClusters1 = function(contribs, K1)
  66     cluster::pam(contribs, K1, diss=FALSE)$id.med
  67
  68 #' @rdname clustering
  69 #' @export
  70 computeClusters2 = function(medoids, K2,
  71     getRefSeries, nb_ref_curves, nb_series_per_chunk, ncores_clust=1,verbose=FALSE,parll=TRUE)
  72 {
  73     synchrones = computeSynchrones(medoids,
  74         getRefSeries, nb_ref_curves, nb_series_per_chunk, ncores_clust, verbose, parll)
  75     distances = computeWerDists(synchrones, ncores_clust, verbose, parll)
  76     medoids[ cluster::pam(distances, K2, diss=TRUE)$medoids , ]
  77 }
  78
  79 #' computeSynchrones
  80 #'
  81 #' Compute the synchrones curves (sum of clusters elements) from a matrix of medoids,
  82 #' using L2 distances.
  83 #'
  84 #' @param medoids Matrix of medoids (curves of same legnth as initial series)
  85 #' @param getRefSeries Function to retrieve initial series (e.g. in stage 2 after series
  86 #'   have been replaced by stage-1 medoids)
  87 #' @param nb_ref_curves How many reference series? (This number is known at this stage)
  88 #' @inheritParams claws
  89 #'
  90 #' @export
  91 computeSynchrones = function(medoids, getRefSeries,
  92     nb_ref_curves, nb_series_per_chunk, ncores_clust=1,verbose=FALSE,parll=TRUE)
  93 {
  94     computeSynchronesChunk = function(indices)
  95     {
  96         ref_series = getRefSeries(indices)
  97         #get medoids indices for this chunk of series
  98         for (i in seq_len(nrow(ref_series)))
  99         {
 100             j = which.min( rowSums( sweep(medoids, 2, ref_series[i,], '-')^2 ) )
 101             if (parll)
 102                 synchronicity::lock(m)
 103             synchrones[j,] = synchrones[j,] + ref_series[i,]
 104             counts[j,1] = counts[j,1] + 1
 105             if (parll)
 106                 synchronicity::unlock(m)
 107         }
 108     }
 109
 110     K = nrow(medoids)
 111     # Use bigmemory (shared==TRUE by default) + synchronicity to fill synchrones in //
 112     synchrones = bigmemory::big.matrix(nrow=K,ncol=ncol(medoids),type="double",init=0.)
 113     counts = bigmemory::big.matrix(nrow=K,ncol=1,type="double",init=0)
 114     # Fork (// run) only on Linux & MacOS; on Windows: run sequentially
 115     parll = (requireNamespace("synchronicity",quietly=TRUE)
 116         && parll && Sys.info()['sysname'] != "Windows")
 117     if (parll)
 118         m <- synchronicity::boost.mutex()
 119
 120     indices_workers = .spreadIndices(seq_len(nb_ref_curves), nb_series_per_chunk)
 121     for (inds in indices_workers)
 122     {
 123         if (verbose)
 124         {
 125             cat(paste("--- Compute synchrones for indices range ",
 126                 min(inds)," -> ",max(inds),"\n", sep=""))
 127         }
 128         if (parll)
 129             ignored <- parallel::mcparallel(computeSynchronesChunk(inds))
 130         else
 131             computeSynchronesChunk(inds)
 132     }
 133     if (parll)
 134         parallel::mccollect()
 135
 136     mat_syncs = matrix(nrow=K, ncol=ncol(medoids))
 137     vec_count = rep(NA, K)
 138     #TODO: can we avoid this loop?
 139     for (i in seq_len(K))
 140     {
 141         mat_syncs[i,] = synchrones[i,]
 142         vec_count[i] = counts[i,1]
 143     }
 144     #NOTE: odds for some clusters to be empty? (when series already come from stage 2)
 145     #      ...maybe; but let's hope resulting K1' be still quite bigger than K2
 146     mat_syncs = sweep(mat_syncs, 1, vec_count, '/')
 147     mat_syncs[ sapply(seq_len(K), function(i) all(!is.nan(mat_syncs[i,]))) , ]
 148 }
 149
 150 #' computeWerDists
 151 #'
 152 #' Compute the WER distances between the synchrones curves (in rows), which are
 153 #' returned (e.g.) by \code{computeSynchrones()}
 154 #'
 155 #' @param synchrones A matrix of synchrones, in rows. The series have same length as the
 156 #'   series in the initial dataset
 157 #' @inheritParams claws
 158 #'
 159 #' @export
 160 computeWerDists = function(synchrones, ncores_clust=1,verbose=FALSE,parll=TRUE)
 161 {
 162     n <- nrow(synchrones)
 163     delta <- ncol(synchrones)
 164     #TODO: automatic tune of all these parameters ? (for other users)
 165     nvoice   <- 4
 166     # noctave = 2^13 = 8192 half hours ~ 180 days ; ~log2(ncol(synchrones))
 167     noctave = 13
 168     # 4 here represent 2^5 = 32 half-hours ~ 1 day
 169     #NOTE: default scalevector == 2^(0:(noctave * nvoice) / nvoice) * s0 (?)
 170     scalevector  <- 2^(4:(noctave * nvoice) / nvoice) * 2
 171     #condition: ( log2(s0*w0/(2*pi)) - 1 ) * nvoice + 1.5 >= 1
 172     s0=2
 173     w0=2*pi
 174     scaled=FALSE
 175     s0log = as.integer( (log2( s0*w0/(2*pi) ) - 1) * nvoice + 1.5 )
 176     totnoct = noctave + as.integer(s0log/nvoice) + 1
 177
 178     computeCWT = function(i)
 179     {
 180         if (verbose)
 181             cat(paste("+++ Compute Rwave::cwt() on serie ",i,"\n", sep=""))
 182         ts <- scale(ts(synchrones[i,]), center=TRUE, scale=scaled)
 183         totts.cwt = Rwave::cwt(ts,totnoct,nvoice,w0,plot=0)
 184         ts.cwt = totts.cwt[,s0log:(s0log+noctave*nvoice)]
 185         #Normalization
 186         sqs <- sqrt(2^(0:(noctave*nvoice)/nvoice)*s0)
 187         sqres <- sweep(ts.cwt,MARGIN=2,sqs,'*')
 188         sqres / max(Mod(sqres))
 189     }
 190
 191     if (parll)
 192     {
 193         cl = parallel::makeCluster(ncores_clust)
 194         parallel::clusterExport(cl, varlist=c("getContribs","K1","verbose"), envir=environment())
 195     }
 196
 197     # (normalized) observations node with CWT
 198     Xcwt4 <-
 199         if (parll)
 200             parallel::parLapply(cl, seq_len(n), computeCWT)
 201         else
 202             lapply(seq_len(n), computeCWT)
 203
 204     if (parll)
 205         parallel::stopCluster(cl)
 206
 207     Xwer_dist <- bigmemory::big.matrix(nrow=n, ncol=n, type="double")
 208     fcoefs = rep(1/3, 3) #moving average on 3 values (TODO: very slow! correct?!)
 209     if (verbose)
 210         cat("*** Compute WER distances from CWT\n")
 211
 212     computeDistancesLineI = function(i)
 213     {
 214         if (verbose)
 215             cat(paste("   Line ",i,"\n", sep=""))
 216         for (j in (i+1):n)
 217         {
 218             #TODO: 'circular=TRUE' is wrong, should just take values on the sides; to rewrite in C
 219             num <- filter(Mod(Xcwt4[[i]] * Conj(Xcwt4[[j]])), fcoefs, circular=TRUE)
 220             WX <- filter(Mod(Xcwt4[[i]] * Conj(Xcwt4[[i]])), fcoefs, circular=TRUE)
 221             WY <- filter(Mod(Xcwt4[[j]] * Conj(Xcwt4[[j]])), fcoefs, circular=TRUE)
 222             wer2    <- sum(colSums(num)^2) / sum( sum(colSums(WX) * colSums(WY)) )
 223             if (parll)
 224                 synchronicity::lock(m)
 225             Xwer_dist[i,j] <- sqrt(delta * ncol(Xcwt4[[1]]) * (1 - wer2))
 226             Xwer_dist[j,i] <- Xwer_dist[i,j]
 227             if (parll)
 228                 synchronicity::unlock(m)
 229         }
 230         Xwer_dist[i,i] = 0.
 231     }
 232
 233     parll = (requireNamespace("synchronicity",quietly=TRUE)
 234         && parll && Sys.info()['sysname'] != "Windows")
 235     if (parll)
 236         m <- synchronicity::boost.mutex()
 237
 238     for (i in 1:(n-1))
 239     {
 240         if (parll)
 241             ignored <- parallel::mcparallel(computeDistancesLineI(i))
 242         else
 243             computeDistancesLineI(i)
 244     }
 245     Xwer_dist[n,n] = 0.
 246
 247     if (parll)
 248         parallel::mccollect()
 249
 250     mat_dists = matrix(nrow=n, ncol=n)
 251     #TODO: avoid this loop?
 252     for (i in 1:n)
 253         mat_dists[i,] = Xwer_dist[i,]
 254     mat_dists
 255 }
 256
 257 # Helper function to divide indices into balanced sets
 258 .spreadIndices = function(indices, nb_per_chunk)
 259 {
 260     L = length(indices)
 261     nb_workers = floor( L / nb_per_chunk )
 262     if (nb_workers == 0)
 263     {
 264         # L < nb_series_per_chunk, simple case
 265         indices_workers = list(indices)
 266     }
 267     else
 268     {
 269         indices_workers = lapply( seq_len(nb_workers), function(i)
 270             indices[(nb_per_chunk*(i-1)+1):(nb_per_chunk*i)] )
 271         # Spread the remaining load among the workers
 272         rem = L %% nb_per_chunk
 273         while (rem > 0)
 274         {
 275             index = rem%%nb_workers + 1
 276             indices_workers[[index]] = c(indices_workers[[index]], indices[L-rem+1])
 277             rem = rem - 1
 278         }
 279     }
 280     indices_workers
 281 }