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 algo_clust1 Clustering algorithm for stage 1. A function which takes (data, K)
  45 #'   as argument where data is a matrix in columns and K the desired number of clusters,
  46 #'   and outputs K medoids ranks. Default: PAM
  47 #' @param algo_clust2 Clustering algorithm for stage 2. A function which takes (dists, K)
  48 #'   as argument where dists is a matrix of distances and K the desired number of clusters,
  49 #'   and outputs K clusters representatives (curves). Default: k-means
  50 #' @param nb_items_clust1 (Maximum) number of items in input of the clustering algorithm
  51 #'   for stage 1
  52 #' @param wav_filt Wavelet transform filter; see ?wavelets::wt.filter
  53 #' @param contrib_type Type of contribution: "relative", "logit" or "absolute" (any prefix)
  54 #' @param WER "end" to apply stage 2 after stage 1 has fully iterated, or "mix" to apply
  55 #'   stage 2 at the end of each task
  56 #' @param random TRUE (default) for random chunks repartition
  57 #' @param ntasks Number of tasks (parallel iterations to obtain K1 [if WER=="end"]
  58 #'   or K2 [if WER=="mix"] medoids); default: 1.
  59 #'   Note: ntasks << N (number of series), so that N is "roughly divisible" by ntasks
  60 #' @param ncores_tasks Number of parallel tasks (1 to disable: sequential tasks)
  61 #' @param ncores_clust Number of parallel clusterings in one task (4 should be a minimum)
  62 #' @param sep Separator in CSV input file (if any provided)
  63 #' @param nbytes Number of bytes to serialize a floating-point number; 4 or 8
  64 #' @param endian Endianness for (de)serialization ("little" or "big")
  65 #' @param verbose Level of verbosity (0/FALSE for nothing or 1/TRUE for all; devel stage)
  66 #' @param parll TRUE to fully parallelize; otherwise run sequentially (debug, comparison)
  67 #'
  68 #' @return A matrix of the final K2 medoids curves, in columns
  69 #'
  70 #' @references Clustering functional data using Wavelets [2013];
  71 #'   A. Antoniadis, X. Brossat, J. Cugliari & J.-M. Poggi.
  72 #'   Inter. J. of Wavelets, Multiresolution and Information Procesing,
  73 #'   vol. 11, No 1, pp.1-30. doi:10.1142/S0219691313500033
  74 #'
  75 #' @examples
  76 #' \dontrun{
  77 #' # WER distances computations are too long for CRAN (for now)
  78 #'
  79 #' # Random series around cos(x,2x,3x)/sin(x,2x,3x)
  80 #' x = seq(0,500,0.05)
  81 #' L = length(x) #10001
  82 #' ref_series = matrix( c(cos(x),cos(2*x),cos(3*x),sin(x),sin(2*x),sin(3*x)), ncol=6 )
  83 #' library(wmtsa)
  84 #' series = do.call( cbind, lapply( 1:6, function(i)
  85 #'   do.call(cbind, wmtsa::wavBootstrap(ref_series[i,], n.realization=400)) ) )
  86 #' #dim(series) #c(2400,10001)
  87 #' medoids_ascii = claws(series, K1=60, K2=6, nb_per_chunk=c(200,500), verbose=TRUE)
  88 #'
  89 #' # Same example, from CSV file
  90 #' csv_file = "/tmp/epclust_series.csv"
  91 #' write.table(series, csv_file, sep=",", row.names=FALSE, col.names=FALSE)
  92 #' medoids_csv = claws(csv_file, K1=60, K2=6, nb_per_chunk=c(200,500))
  93 #'
  94 #' # Same example, from binary file
  95 #' bin_file <- "/tmp/epclust_series.bin"
  96 #' nbytes <- 8
  97 #' endian <- "little"
  98 #' binarize(csv_file, bin_file, 500, nbytes, endian)
  99 #' getSeries <- function(indices) getDataInFile(indices, bin_file, nbytes, endian)
 100 #' medoids_bin <- claws(getSeries, K1=60, K2=6, nb_per_chunk=c(200,500))
 101 #' unlink(csv_file)
 102 #' unlink(bin_file)
 103 #'
 104 #' # Same example, from SQLite database
 105 #' library(DBI)
 106 #' series_db <- dbConnect(RSQLite::SQLite(), "file::memory:")
 107 #' # Prepare data.frame in DB-format
 108 #' n <- nrow(series)
 109 #' time_values <- data.frame(
 110 #'   id = rep(1:n,each=L),
 111 #'   time = rep( as.POSIXct(1800*(0:n),"GMT",origin="2001-01-01"), L ),
 112 #'   value = as.double(t(series)) )
 113 #' dbWriteTable(series_db, "times_values", times_values)
 114 #' # Fill associative array, map index to identifier
 115 #' indexToID_inDB <- as.character(
 116 #'   dbGetQuery(series_db, 'SELECT DISTINCT id FROM time_values')[,"id"] )
 117 #' serie_length <- as.integer( dbGetQuery(series_db,
 118 #'   paste("SELECT COUNT * FROM time_values WHERE id == ",indexToID_inDB[1],sep="")) )
 119 #' getSeries <- function(indices) {
 120 #'   request <- "SELECT id,value FROM times_values WHERE id in ("
 121 #'   for (i in indices)
 122 #'     request <- paste(request, indexToID_inDB[i], ",", sep="")
 123 #'   request <- paste(request, ")", sep="")
 124 #'   df_series <- dbGetQuery(series_db, request)
 125 #'   as.matrix(df_series[,"value"], nrow=serie_length)
 126 #' }
 127 #' medoids_db = claws(getSeries, K1=60, K2=6, nb_per_chunk=c(200,500))
 128 #' dbDisconnect(series_db)
 129 #'
 130 #' # All computed medoids should be the same:
 131 #' digest::sha1(medoids_ascii)
 132 #' digest::sha1(medoids_csv)
 133 #' digest::sha1(medoids_bin)
 134 #' digest::sha1(medoids_db)
 135 #' }
 136 #' @export
 137 claws <- function(getSeries, K1, K2, nb_per_chunk,
 138     nb_items_clust1=7*K1,
 139     algo_clust1=function(data,K) cluster::pam(data,K,diss=FALSE),
 140     algo_clust2=function(dists,K) stats::kmeans(dists,K,iter.max=50,nstart=3),
 141     wav_filt="d8",contrib_type="absolute",
 142     WER="end",
 143     random=TRUE,
 144     ntasks=1, ncores_tasks=1, ncores_clust=4,
 145     sep=",",
 146     nbytes=4, endian=.Platform$endian,
 147     verbose=FALSE, parll=TRUE)
 148 {
 149     # Check/transform arguments
 150     if (!is.matrix(getSeries) && !bigmemory::is.big.matrix(getSeries)
 151         && !is.function(getSeries)
 152         && !methods::is(getSeries,"connection") && !is.character(getSeries))
 153     {
 154         stop("'getSeries': [big]matrix, function, file or valid connection (no NA)")
 155     }
 156     K1 <- .toInteger(K1, function(x) x>=2)
 157     K2 <- .toInteger(K2, function(x) x>=2)
 158     if (!is.numeric(nb_per_chunk) || length(nb_per_chunk)!=2)
 159         stop("'nb_per_chunk': numeric, size 2")
 160     nb_per_chunk[1] <- .toInteger(nb_per_chunk[1], function(x) x>=1)
 161     # A batch of contributions should have at least as many elements as a batch of series,
 162     # because it always contains much less values
 163     nb_per_chunk[2] <- max(.toInteger(nb_per_chunk[2],function(x) x>=1), nb_per_chunk[1])
 164     nb_items_clust1 <- .toInteger(nb_items_clust1, function(x) x>K1)
 165     random <- .toLogical(random)
 166     tryCatch
 167     (
 168         {ignored <- wavelets::wt.filter(wav_filt)},
 169         error = function(e) stop("Invalid wavelet filter; see ?wavelets::wt.filter")
 170     )
 171     ctypes = c("relative","absolute","logit")
 172     contrib_type = ctypes[ pmatch(contrib_type,ctypes) ]
 173     if (is.na(contrib_type))
 174         stop("'contrib_type' in {'relative','absolute','logit'}")
 175     if (WER!="end" && WER!="mix")
 176         stop("'WER': in {'end','mix'}")
 177     random <- .toLogical(random)
 178     ntasks <- .toInteger(ntasks, function(x) x>=1)
 179     ncores_tasks <- .toInteger(ncores_tasks, function(x) x>=1)
 180     ncores_clust <- .toInteger(ncores_clust, function(x) x>=1)
 181     if (!is.character(sep))
 182         stop("'sep': character")
 183     nbytes <- .toInteger(nbytes, function(x) x==4 || x==8)
 184     verbose <- .toLogical(verbose)
 185     parll <- .toLogical(parll)
 186
 187     # Since we don't make assumptions on initial data, there is a possibility that even
 188     # when serialized, contributions or synchrones do not fit in RAM. For example,
 189     # 30e6 series of length 100,000 would lead to a +4Go contribution matrix. Therefore,
 190     # it's safer to place these in (binary) files, located in the following folder.
 191     bin_dir <- ".epclust_bin/"
 192     dir.create(bin_dir, showWarnings=FALSE, mode="0755")
 193
 194     # Binarize series if getSeries is not a function; the aim is to always use a function,
 195     # to uniformize treatments. An equally good alternative would be to use a file-backed
 196     # bigmemory::big.matrix, but it would break the uniformity.
 197     if (!is.function(getSeries))
 198     {
 199         if (verbose)
 200             cat("...Serialize time-series\n")
 201         series_file = paste(bin_dir,"data",sep="") ; unlink(series_file)
 202         binarize(getSeries, series_file, nb_series_per_chunk, sep, nbytes, endian)
 203         getSeries = function(inds) getDataInFile(inds, series_file, nbytes, endian)
 204     }
 205
 206     # Serialize all computed wavelets contributions into a file
 207     contribs_file = paste(bin_dir,"contribs",sep="") ; unlink(contribs_file)
 208     index = 1
 209     nb_curves = 0
 210     if (verbose)
 211         cat("...Compute contributions and serialize them\n")
 212     nb_curves = binarizeTransform(getSeries,
 213         function(series) curvesToContribs(series, wf, ctype),
 214         contribs_file, nb_series_per_chunk, nbytes, endian)
 215     getContribs = function(indices) getDataInFile(indices, contribs_file, nbytes, endian)
 216
 217     # A few sanity checks: do not continue if too few data available.
 218     if (nb_curves < K2)
 219         stop("Not enough data: less series than final number of clusters")
 220     nb_series_per_task = round(nb_curves / ntasks)
 221     if (nb_series_per_task < K2)
 222         stop("Too many tasks: less series in one task than final number of clusters")
 223
 224     # Generate a random permutation of 1:N (if random==TRUE); otherwise just use arrival
 225     # (storage) order.
 226     indices_all = if (random) sample(nb_curves) else seq_len(nb_curves)
 227     # Split (all) indices into ntasks groups of ~same size
 228     indices_tasks = lapply(seq_len(ntasks), function(i) {
 229         upper_bound = ifelse( i<ntasks, min(nb_series_per_task*i,nb_curves), nb_curves )
 230         indices_all[((i-1)*nb_series_per_task+1):upper_bound]
 231     })
 232
 233     if (parll && ntasks>1)
 234     {
 235         # Initialize parallel runs: outfile="" allow to output verbose traces in the console
 236         # under Linux. All necessary variables are passed to the workers.
 237         cl = parallel::makeCluster(ncores_tasks, outfile="")
 238         varlist = c("getSeries","getContribs","K1","K2","algo_clust1","algo_clust2",
 239             "nb_per_chunk","nb_items_clust","ncores_clust","sep","nbytes","endian",
 240             "verbose","parll")
 241         if (WER=="mix")
 242             varlist = c(varlist, "medoids_file")
 243         parallel::clusterExport(cl, varlist, envir = environment())
 244     }
 245
 246     # This function achieves one complete clustering task, divided in stage 1 + stage 2.
 247     # stage 1: n indices  --> clusteringTask1(...) --> K1 medoids
 248     # stage 2: K1 medoids --> clusteringTask2(...) --> K2 medoids,
 249     # where n = N / ntasks, N being the total number of curves.
 250     runTwoStepClustering = function(inds)
 251     {
 252         # When running in parallel, the environment is blank: we need to load required
 253         # packages, and pass useful variables.
 254         if (parll && ntasks>1)
 255             require("epclust", quietly=TRUE)
 256         indices_medoids = clusteringTask1(
 257             inds, getContribs, K1, nb_series_per_chunk, ncores_clust, verbose, parll)
 258         if (WER=="mix")
 259         {
 260             if (parll && ntasks>1)
 261                 require("bigmemory", quietly=TRUE)
 262             medoids1 = bigmemory::as.big.matrix( getSeries(indices_medoids) )
 263             medoids2 = clusteringTask2(medoids1, K2, getSeries, nb_curves, nb_series_per_chunk,
 264                 nbytes, endian, ncores_clust, verbose, parll)
 265             binarize(medoids2, medoids_file, nb_series_per_chunk, sep, nbytes, endian)
 266             return (vector("integer",0))
 267         }
 268         indices_medoids
 269     }
 270
 271     # Synchrones (medoids) need to be stored only if WER=="mix"; indeed in this case, every
 272     # task output is a set of new (medoids) curves. If WER=="end" however, output is just a
 273     # set of indices, representing some initial series.
 274     if (WER=="mix")
 275         {medoids_file = paste(bin_dir,"medoids",sep="") ; unlink(medoids_file)}
 276
 277     if (verbose)
 278     {
 279         message = paste("...Run ",ntasks," x stage 1", sep="")
 280         if (WER=="mix")
 281             message = paste(message," + stage 2", sep="")
 282         cat(paste(message,"\n", sep=""))
 283     }
 284
 285     # As explained above, indices will be assigned to ntasks*K1 medoids indices [if WER=="end"],
 286     # or nothing (empty vector) if WER=="mix"; in this case, medoids (synchrones) are stored
 287     # in a file.
 288     indices <-
 289         if (parll && ntasks>1)
 290             unlist( parallel::parLapply(cl, indices_tasks, runTwoStepClustering) )
 291         else
 292             unlist( lapply(indices_tasks, runTwoStepClustering) )
 293     if (parll && ntasks>1)
 294         parallel::stopCluster(cl)
 295
 296     # Right before the final stage, two situations are possible:
 297     #  a. data to be processed now sit in binary format in medoids_file (if WER=="mix")
 298     #  b. data still is the initial set of curves, referenced by the ntasks*K1 indices
 299     # So, the function getSeries() will potentially change. However, computeSynchrones()
 300     # requires a function retrieving the initial series. Thus, the next line saves future
 301     # conditional instructions.
 302     getRefSeries = getSeries
 303
 304     if (WER=="mix")
 305     {
 306         indices = seq_len(ntasks*K2)
 307         # Now series must be retrieved from synchrones_file
 308         getSeries = function(inds) getDataInFile(inds, synchrones_file, nbytes, endian)
 309         # Contributions must be re-computed
 310         unlink(contribs_file)
 311         index = 1
 312         if (verbose)
 313             cat("...Serialize contributions computed on synchrones\n")
 314         ignored = binarizeTransform(getSeries,
 315             function(series) curvesToContribs(series, wf, ctype),
 316             contribs_file, nb_series_per_chunk, nbytes, endian)
 317     }
 318
 319     # Run step2 on resulting indices or series (from file)
 320     if (verbose)
 321         cat("...Run final // stage 1 + stage 2\n")
 322     indices_medoids = clusteringTask1(
 323         indices, getContribs, K1, nb_series_per_chunk, ncores_tasks*ncores_clust, verbose, parll)
 324     medoids1 = bigmemory::as.big.matrix( getSeries(indices_medoids) )
 325     medoids2 = clusteringTask2(medoids1, K2, getRefSeries, nb_curves, nb_series_per_chunk,
 326         nbytes, endian, ncores_tasks*ncores_clust, verbose, parll)
 327
 328     # Cleanup: remove temporary binary files and their folder
 329     unlink(bin_dir, recursive=TRUE)
 330
 331     # Return medoids as a standard matrix, since K2 series have to fit in RAM
 332     # (clustering algorithm 1 takes K1 > K2 of them as input)
 333     medoids2[,]
 334 }
 335
 336 #' curvesToContribs
 337 #'
 338 #' Compute the discrete wavelet coefficients for each series, and aggregate them in
 339 #' energy contribution across scales as described in https://arxiv.org/abs/1101.4744v2
 340 #'
 341 #' @param series [big.]matrix of series (in columns), of size L x n
 342 #' @inheritParams claws
 343 #'
 344 #' @return A [big.]matrix of size log(L) x n containing contributions in columns
 345 #'
 346 #' @export
 347 curvesToContribs = function(series, wav_filt, contrib_type)
 348 {
 349     L = nrow(series)
 350     D = ceiling( log2(L) )
 351     nb_sample_points = 2^D
 352     apply(series, 2, function(x) {
 353         interpolated_curve = spline(1:L, x, n=nb_sample_points)$y
 354         W = wavelets::dwt(interpolated_curve, filter=wf, D)@W
 355         nrj = rev( sapply( W, function(v) ( sqrt( sum(v^2) ) ) ) )
 356         if (contrib_type!="absolute")
 357             nrj = nrj / sum(nrj)
 358         if (contrib_type=="logit")
 359             nrj = - log(1 - nrj)
 360         nrj
 361     })
 362 }
 363
 364 # Check integer arguments with functional conditions
 365 .toInteger <- function(x, condition)
 366 {
 367     errWarn <- function(ignored)
 368         paste("Cannot convert argument' ",substitute(x),"' to integer", sep="")
 369     if (!is.integer(x))
 370         tryCatch({x = as.integer(x)[1]; if (is.na(x)) stop()},
 371             warning = errWarn, error = errWarn)
 372     if (!condition(x))
 373     {
 374         stop(paste("Argument '",substitute(x),
 375             "' does not verify condition ",body(condition), sep=""))
 376     }
 377     x
 378 }
 379
 380 # Check logical arguments
 381 .toLogical <- function(x)
 382 {
 383     errWarn <- function(ignored)
 384         paste("Cannot convert argument' ",substitute(x),"' to logical", sep="")
 385     if (!is.logical(x))
 386         tryCatch({x = as.logical(x)[1]; if (is.na(x)) stop()},
 387             warning = errWarn, error = errWarn)
 388     x
 389 }