epclust/R/main.R

   1 #' CLAWS: CLustering with wAvelets and Wer distanceS
   2 #'
   3 #' Cluster electricity power curves (or any series of similar nature) by applying a
   4 #' two stage procedure in parallel (see details).
   5 #' Input series must be sampled on the same time grid, no missing values.
   6 #'
   7 #' @details Summary of the function execution flow:
   8 #'   \enumerate{
   9 #'     \item Compute and serialize all contributions, obtained through discrete wavelet
  10 #'       decomposition (see Antoniadis & al. [2013])
  11 #'     \item Divide series into \code{ntasks} groups to process in parallel. In each task:
  12 #'     \enumerate{
  13 #'       \item iterate the first clustering algorithm on its aggregated outputs,
  14 #'         on inputs of size \code{nb_items_clust}
  15 #'       \item optionally, if WER=="mix":
  16 #'         a) compute the K1 synchrones curves,
  17 #'         b) compute WER distances (K1xK1 matrix) between synchrones and
  18 #'         c) apply the second clustering algorithm
  19 #'     }
  20 #'     \item Launch a final task on the aggregated outputs of all previous tasks:
  21 #'       in the case WER=="end" this task takes indices in input, otherwise
  22 #'       (medoid) curves
  23 #'   }
  24 #'   The main argument -- \code{getSeries} -- has a quite misleading name, since it can be
  25 #'   either a [big.]matrix, a CSV file, a connection or a user function to retrieve
  26 #'   series; the name was chosen because all types of arguments are converted to a function.
  27 #'   When \code{getSeries} is given as a function, it must take a single argument,
  28 #'   'indices', integer vector equal to the indices of the curves to retrieve;
  29 #'   see SQLite example. The nature and role of other arguments should be clear
  30 #'
  31 #' @param getSeries Access to the (time-)series, which can be of one of the three
  32 #'   following types:
  33 #'   \itemize{
  34 #'     \item [big.]matrix: each column contains the (time-ordered) values of one time-serie
  35 #'     \item connection: any R connection object providing lines as described above
  36 #'     \item character: name of a CSV file containing series in rows (no header)
  37 #'     \item function: a custom way to retrieve the curves; it has only one argument:
  38 #'       the indices of the series to be retrieved. See SQLite example
  39 #'   }
  40 #' @param K1 Number of clusters to be found after stage 1 (K1 << N [number of series])
  41 #' @param K2 Number of clusters to be found after stage 2 (K2 << K1)
  42 #' @param nb_per_chunk (Maximum) number of items to retrieve in one batch, for both types of
  43 #'   retrieval: resp. series and contribution; in a vector of size 2
  44 #' @param nb_items_clust1 (Maximum) number of items in input of the clustering algorithm
  45 #'   for stage 1
  46 #' @param wav_filt Wavelet transform filter; see ?wavelets::wt.filter
  47 #' @param contrib_type Type of contribution: "relative", "logit" or "absolute" (any prefix)
  48 #' @param WER "end" to apply stage 2 after stage 1 has fully iterated, or "mix" to apply
  49 #'   stage 2 at the end of each task
  50 #' @param random TRUE (default) for random chunks repartition
  51 #' @param ntasks Number of tasks (parallel iterations to obtain K1 [if WER=="end"]
  52 #'   or K2 [if WER=="mix"] medoids); default: 1.
  53 #'   Note: ntasks << N (number of series), so that N is "roughly divisible" by ntasks
  54 #' @param ncores_tasks Number of parallel tasks (1 to disable: sequential tasks)
  55 #' @param ncores_clust Number of parallel clusterings in one task (4 should be a minimum)
  56 #' @param sep Separator in CSV input file (if any provided)
  57 #' @param nbytes Number of bytes to serialize a floating-point number; 4 or 8
  58 #' @param endian Endianness for (de)serialization ("little" or "big")
  59 #' @param verbose Level of verbosity (0/FALSE for nothing or 1/TRUE for all; devel stage)
  60 #' @param parll TRUE to fully parallelize; otherwise run sequentially (debug, comparison)
  61 #'
  62 #' @return A matrix of the final K2 medoids curves, in columns
  63 #'
  64 #' @references Clustering functional data using Wavelets [2013];
  65 #'   A. Antoniadis, X. Brossat, J. Cugliari & J.-M. Poggi.
  66 #'   Inter. J. of Wavelets, Multiresolution and Information Procesing,
  67 #'   vol. 11, No 1, pp.1-30. doi:10.1142/S0219691313500033
  68 #'
  69 #' @examples
  70 #' \dontrun{
  71 #' # WER distances computations are too long for CRAN (for now)
  72 #'
  73 #' # Random series around cos(x,2x,3x)/sin(x,2x,3x)
  74 #' x = seq(0,500,0.05)
  75 #' L = length(x) #10001
  76 #' ref_series = matrix( c(cos(x),cos(2*x),cos(3*x),sin(x),sin(2*x),sin(3*x)), ncol=6 )
  77 #' library(wmtsa)
  78 #' series = do.call( cbind, lapply( 1:6, function(i)
  79 #'   do.call(cbind, wmtsa::wavBootstrap(ref_series[i,], n.realization=400)) ) )
  80 #' #dim(series) #c(2400,10001)
  81 #' medoids_ascii = claws(series, K1=60, K2=6, nb_per_chunk=c(200,500), verbose=TRUE)
  82 #'
  83 #' # Same example, from CSV file
  84 #' csv_file = "/tmp/epclust_series.csv"
  85 #' write.table(series, csv_file, sep=",", row.names=FALSE, col.names=FALSE)
  86 #' medoids_csv = claws(csv_file, K1=60, K2=6, nb_per_chunk=c(200,500))
  87 #'
  88 #' # Same example, from binary file
  89 #' bin_file <- "/tmp/epclust_series.bin"
  90 #' nbytes <- 8
  91 #' endian <- "little"
  92 #' binarize(csv_file, bin_file, 500, nbytes, endian)
  93 #' getSeries <- function(indices) getDataInFile(indices, bin_file, nbytes, endian)
  94 #' medoids_bin <- claws(getSeries, K1=60, K2=6, nb_per_chunk=c(200,500))
  95 #' unlink(csv_file)
  96 #' unlink(bin_file)
  97 #'
  98 #' # Same example, from SQLite database
  99 #' library(DBI)
 100 #' series_db <- dbConnect(RSQLite::SQLite(), "file::memory:")
 101 #' # Prepare data.frame in DB-format
 102 #' n <- nrow(series)
 103 #' time_values <- data.frame(
 104 #'   id = rep(1:n,each=L),
 105 #'   time = rep( as.POSIXct(1800*(0:n),"GMT",origin="2001-01-01"), L ),
 106 #'   value = as.double(t(series)) )
 107 #' dbWriteTable(series_db, "times_values", times_values)
 108 #' # Fill associative array, map index to identifier
 109 #' indexToID_inDB <- as.character(
 110 #'   dbGetQuery(series_db, 'SELECT DISTINCT id FROM time_values')[,"id"] )
 111 #' serie_length <- as.integer( dbGetQuery(series_db,
 112 #'   paste("SELECT COUNT * FROM time_values WHERE id == ",indexToID_inDB[1],sep="")) )
 113 #' getSeries <- function(indices) {
 114 #'   request <- "SELECT id,value FROM times_values WHERE id in ("
 115 #'   for (i in indices)
 116 #'     request <- paste(request, indexToID_inDB[i], ",", sep="")
 117 #'   request <- paste(request, ")", sep="")
 118 #'   df_series <- dbGetQuery(series_db, request)
 119 #'   as.matrix(df_series[,"value"], nrow=serie_length)
 120 #' }
 121 #' medoids_db = claws(getSeries, K1=60, K2=6, nb_per_chunk=c(200,500))
 122 #' dbDisconnect(series_db)
 123 #'
 124 #' # All computed medoids should be the same:
 125 #' digest::sha1(medoids_ascii)
 126 #' digest::sha1(medoids_csv)
 127 #' digest::sha1(medoids_bin)
 128 #' digest::sha1(medoids_db)
 129 #' }
 130 #' @export
 131 claws <- function(getSeries, K1, K2,
 132     nb_per_chunk,nb_items_clust1=7*K1 #volumes of data
 133     wav_filt="d8",contrib_type="absolute", #stage 1
 134     WER="end", #stage 2
 135     random=TRUE, #randomize series order?
 136     ntasks=1, ncores_tasks=1, ncores_clust=4, #parallelism
 137     sep=",", #ASCII input separator
 138     nbytes=4, endian=.Platform$endian, #serialization (write,read)
 139     verbose=FALSE, parll=TRUE)
 140 {
 141     # Check/transform arguments
 142     if (!is.matrix(getSeries) && !bigmemory::is.big.matrix(getSeries)
 143         && !is.function(getSeries)
 144         && !methods::is(getSeries,"connection") && !is.character(getSeries))
 145     {
 146         stop("'getSeries': [big]matrix, function, file or valid connection (no NA)")
 147     }
 148     K1 <- .toInteger(K1, function(x) x>=2)
 149     K2 <- .toInteger(K2, function(x) x>=2)
 150     if (!is.numeric(nb_per_chunk) || length(nb_per_chunk)!=2)
 151         stop("'nb_per_chunk': numeric, size 2")
 152     nb_per_chunk[1] <- .toInteger(nb_per_chunk[1], function(x) x>=1)
 153     # A batch of contributions should have at least as many elements as a batch of series,
 154     # because it always contains much less values
 155     nb_per_chunk[2] <- max(.toInteger(nb_per_chunk[2],function(x) x>=1), nb_per_chunk[1])
 156     nb_items_clust1 <- .toInteger(nb_items_clust1, function(x) x>K1)
 157     random <- .toLogical(random)
 158     tryCatch
 159     (
 160         {ignored <- wavelets::wt.filter(wav_filt)},
 161         error = function(e) stop("Invalid wavelet filter; see ?wavelets::wt.filter")
 162     )
 163     ctypes = c("relative","absolute","logit")
 164     contrib_type = ctypes[ pmatch(contrib_type,ctypes) ]
 165     if (is.na(contrib_type))
 166         stop("'contrib_type' in {'relative','absolute','logit'}")
 167     if (WER!="end" && WER!="mix")
 168         stop("'WER': in {'end','mix'}")
 169     random <- .toLogical(random)
 170     ntasks <- .toInteger(ntasks, function(x) x>=1)
 171     ncores_tasks <- .toInteger(ncores_tasks, function(x) x>=1)
 172     ncores_clust <- .toInteger(ncores_clust, function(x) x>=1)
 173     if (!is.character(sep))
 174         stop("'sep': character")
 175     nbytes <- .toInteger(nbytes, function(x) x==4 || x==8)
 176     verbose <- .toLogical(verbose)
 177     parll <- .toLogical(parll)
 178
 179     # Serialize series if required, to always use a function
 180     bin_dir <- ".epclust_bin/"
 181     dir.create(bin_dir, showWarnings=FALSE, mode="0755")
 182     if (!is.function(getSeries))
 183     {
 184         if (verbose)
 185             cat("...Serialize time-series\n")
 186         series_file = paste(bin_dir,"data",sep="") ; unlink(series_file)
 187         binarize(getSeries, series_file, nb_series_per_chunk, sep, nbytes, endian)
 188         getSeries = function(inds) getDataInFile(inds, series_file, nbytes, endian)
 189     }
 190
 191     # Serialize all computed wavelets contributions into a file
 192     contribs_file = paste(bin_dir,"contribs",sep="") ; unlink(contribs_file)
 193     index = 1
 194     nb_curves = 0
 195     if (verbose)
 196         cat("...Compute contributions and serialize them\n")
 197     nb_curves = binarizeTransform(getSeries,
 198         function(series) curvesToContribs(series, wf, ctype),
 199         contribs_file, nb_series_per_chunk, nbytes, endian)
 200     getContribs = function(indices) getDataInFile(indices, contribs_file, nbytes, endian)
 201
 202     if (nb_curves < K2)
 203         stop("Not enough data: less series than final number of clusters")
 204     nb_series_per_task = round(nb_curves / ntasks)
 205     if (nb_series_per_task < K2)
 206         stop("Too many tasks: less series in one task than final number of clusters")
 207
 208     runTwoStepClustering = function(inds)
 209     {
 210         if (parll && ntasks>1)
 211             require("epclust", quietly=TRUE)
 212         indices_medoids = clusteringTask1(
 213             inds, getContribs, K1, nb_series_per_chunk, ncores_clust, verbose, parll)
 214         if (WER=="mix")
 215         {
 216             if (parll && ntasks>1)
 217                 require("bigmemory", quietly=TRUE)
 218             medoids1 = bigmemory::as.big.matrix( getSeries(indices_medoids) )
 219             medoids2 = clusteringTask2(medoids1, K2, getSeries, nb_curves, nb_series_per_chunk,
 220                 nbytes, endian, ncores_clust, verbose, parll)
 221             binarize(medoids2, synchrones_file, nb_series_per_chunk, sep, nbytes, endian)
 222             return (vector("integer",0))
 223         }
 224         indices_medoids
 225     }
 226
 227     # Cluster contributions in parallel (by nb_series_per_chunk)
 228     indices_all = if (random) sample(nb_curves) else seq_len(nb_curves)
 229     indices_tasks = lapply(seq_len(ntasks), function(i) {
 230         upper_bound = ifelse( i<ntasks, min(nb_series_per_task*i,nb_curves), nb_curves )
 231         indices_all[((i-1)*nb_series_per_task+1):upper_bound]
 232     })
 233     if (verbose)
 234     {
 235         message = paste("...Run ",ntasks," x stage 1", sep="")
 236         if (WER=="mix")
 237             message = paste(message," + stage 2", sep="")
 238         cat(paste(message,"\n", sep=""))
 239     }
 240     if (WER=="mix")
 241         {synchrones_file = paste(bin_dir,"synchrones",sep="") ; unlink(synchrones_file)}
 242     if (parll && ntasks>1)
 243     {
 244         cl = parallel::makeCluster(ncores_tasks, outfile="")
 245         varlist = c("getSeries","getContribs","K1","K2","verbose","parll",
 246             "nb_series_per_chunk","ntasks","ncores_clust","sep","nbytes","endian")
 247         if (WER=="mix")
 248             varlist = c(varlist, "synchrones_file")
 249         parallel::clusterExport(cl, varlist=varlist, envir = environment())
 250     }
 251
 252     # 1000*K1 indices [if WER=="end"], or empty vector [if WER=="mix"] --> series on file
 253     indices <-
 254         if (parll && ntasks>1)
 255             unlist( parallel::parLapply(cl, indices_tasks, runTwoStepClustering) )
 256         else
 257             unlist( lapply(indices_tasks, runTwoStepClustering) )
 258     if (parll && ntasks>1)
 259         parallel::stopCluster(cl)
 260
 261     getRefSeries = getSeries
 262     if (WER=="mix")
 263     {
 264         indices = seq_len(ntasks*K2)
 265         #Now series must be retrieved from synchrones_file
 266         getSeries = function(inds) getDataInFile(inds, synchrones_file, nbytes, endian)
 267         #Contributions must be re-computed
 268         unlink(contribs_file)
 269         index = 1
 270         if (verbose)
 271             cat("...Serialize contributions computed on synchrones\n")
 272         ignored = binarizeTransform(getSeries,
 273             function(series) curvesToContribs(series, wf, ctype),
 274             contribs_file, nb_series_per_chunk, nbytes, endian)
 275     }
 276
 277     # Run step2 on resulting indices or series (from file)
 278     if (verbose)
 279         cat("...Run final // stage 1 + stage 2\n")
 280     indices_medoids = clusteringTask1(
 281         indices, getContribs, K1, nb_series_per_chunk, ncores_tasks*ncores_clust, verbose, parll)
 282     medoids1 = bigmemory::as.big.matrix( getSeries(indices_medoids) )
 283     medoids2 = clusteringTask2(medoids1, K2, getRefSeries, nb_curves, nb_series_per_chunk,
 284         nbytes, endian, ncores_tasks*ncores_clust, verbose, parll)
 285
 286     # Cleanup
 287     unlink(bin_dir, recursive=TRUE)
 288
 289     medoids2[,]
 290 }
 291
 292 #' curvesToContribs
 293 #'
 294 #' Compute the discrete wavelet coefficients for each series, and aggregate them in
 295 #' energy contribution across scales as described in https://arxiv.org/abs/1101.4744v2
 296 #'
 297 #' @param series [big.]matrix of series (in columns), of size L x n
 298 #' @inheritParams claws
 299 #'
 300 #' @return A [big.]matrix of size log(L) x n containing contributions in columns
 301 #'
 302 #' @export
 303 curvesToContribs = function(series, wav_filt, contrib_type)
 304 {
 305     L = nrow(series)
 306     D = ceiling( log2(L) )
 307     nb_sample_points = 2^D
 308     apply(series, 2, function(x) {
 309         interpolated_curve = spline(1:L, x, n=nb_sample_points)$y
 310         W = wavelets::dwt(interpolated_curve, filter=wf, D)@W
 311         nrj = rev( sapply( W, function(v) ( sqrt( sum(v^2) ) ) ) )
 312         if (contrib_type!="absolute")
 313             nrj = nrj / sum(nrj)
 314         if (contrib_type=="logit")
 315             nrj = - log(1 - nrj)
 316         nrj
 317     })
 318 }
 319
 320 # Check integer arguments with functional conditions
 321 .toInteger <- function(x, condition)
 322 {
 323     errWarn <- function(ignored)
 324         paste("Cannot convert argument' ",substitute(x),"' to integer", sep="")
 325     if (!is.integer(x))
 326         tryCatch({x = as.integer(x)[1]; if (is.na(x)) stop()},
 327             warning = errWarn, error = errWarn)
 328     if (!condition(x))
 329     {
 330         stop(paste("Argument '",substitute(x),
 331             "' does not verify condition ",body(condition), sep=""))
 332     }
 333     x
 334 }
 335
 336 # Check logical arguments
 337 .toLogical <- function(x)
 338 {
 339     errWarn <- function(ignored)
 340         paste("Cannot convert argument' ",substitute(x),"' to logical", sep="")
 341     if (!is.logical(x))
 342         tryCatch({x = as.logical(x)[1]; if (is.na(x)) stop()},
 343             warning = errWarn, error = errWarn)
 344     x
 345 }