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 #' 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_series_per_chunk}
  15 #'     \item optionally, if WER=="mix":
  16 #'       a) compute the K1 synchrones curves,
  17 #'       a) compute WER distances (K1xK1 matrix) between medoids and
  18 #'       b) apply the second clustering algorithm (output: K2 indices)
  19 #'   }
  20 #'   \item Launch a final task on the aggregated outputs of all previous tasks:
  21 #'     ntasks*K1 if WER=="end", ntasks*K2 otherwise
  22 #'   \item Compute synchrones (sum of series within each final group)
  23 #' }
  24 #' \cr
  25 #' The main argument -- \code{series} -- has a quite misleading name, since it can be
  26 #' either a [big.]matrix, a CSV file, a connection or a user function to retrieve series.
  27 #' When \code{series} 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.
  30 #' WARNING: the return value must be a matrix (in columns), or NULL if no matches.
  31 #' \cr
  32 #' Note: Since we don't make assumptions on initial data, there is a possibility that
  33 #' even when serialized, contributions do not fit in RAM. For example,
  34 #' 30e6 series of length 100,000 would lead to a +4Go contribution matrix. Therefore,
  35 #' it's safer to place these in (binary) files; that's what we do.
  36 #'
  37 #' @param series Access to the (time-)series, which can be of one of the three
  38 #'   following types:
  39 #'   \itemize{
  40 #'     \item [big.]matrix: each column contains the (time-ordered) values of one time-serie
  41 #'     \item connection: any R connection object providing lines as described above
  42 #'     \item character: name of a CSV file containing series in rows (no header)
  43 #'     \item function: a custom way to retrieve the curves; it has only one argument:
  44 #'       the indices of the series to be retrieved. See SQLite example
  45 #'   }
  46 #' @param K1 Number of clusters to be found after stage 1 (K1 << N [number of series])
  47 #' @param K2 Number of clusters to be found after stage 2 (K2 << K1)
  48 #' @param nb_series_per_chunk (Maximum) number of series to retrieve in one batch;
  49 #'   this value is also used for the (maximum) number of series to cluster at a time
  50 #' @param algoClust1 Clustering algorithm for stage 1. A function which takes (data, K)
  51 #'   as argument where data is a matrix in columns and K the desired number of clusters,
  52 #'   and outputs K medoids ranks. Default: PAM. In our method, this function is called
  53 #'   on iterated medoids during stage 1
  54 #' @param algoClust2 Clustering algorithm for stage 2. A function which takes (dists, K)
  55 #'   as argument where dists is a matrix of distances and K the desired number of clusters,
  56 #'   and outputs K medoids ranks. Default: PAM.  In our method, this function is called
  57 #'   on a matrix of K1 x K1 (WER) distances computed between medoids after algorithm 1
  58 #' @param wav_filt Wavelet transform filter; see ?wavelets::wt.filter
  59 #' @param contrib_type Type of contribution: "relative", "logit" or "absolute" (any prefix)
  60 #' @param WER "end" to apply stage 2 after stage 1 has fully iterated, or "mix" to apply
  61 #'   stage 2 at the end of each task
  62 #' @param nvoice Number of voices within each octave for CWT computations
  63 #' @param random TRUE (default) for random chunks repartition
  64 #' @param ntasks Number of tasks (parallel iterations to obtain K1 [if WER=="end"]
  65 #'   or K2 [if WER=="mix"] medoids); default: 1.
  66 #'   Note: ntasks << N (number of series), so that N is "roughly divisible" by ntasks
  67 #' @param ncores_tasks Number of parallel tasks (1 to disable: sequential tasks)
  68 #' @param ncores_clust Number of parallel clusterings in one task (3 should be a minimum)
  69 #' @param sep Separator in CSV input file (if any provided)
  70 #' @param nbytes Number of bytes to serialize a floating-point number; 4 or 8
  71 #' @param endian Endianness for (de)serialization ("little" or "big")
  72 #' @param verbose Level of verbosity (0/FALSE for nothing or 1/TRUE for all; devel stage)
  73 #' @param parll TRUE to fully parallelize; otherwise run sequentially (debug, comparison)
  74 #'
  75 #' @return A list with
  76 #' \itemize{
  77 #'   medoids: a matrix of the final K2 medoids curves, in columns
  78 #'   ranks: corresponding indices in the dataset
  79 #'   synchrones: a matrix of the K2 sum of series within each final group
  80 #' }
  81 #'
  82 #' @references Clustering functional data using Wavelets [2013];
  83 #'   A. Antoniadis, X. Brossat, J. Cugliari & J.-M. Poggi.
  84 #'   Inter. J. of Wavelets, Multiresolution and Information Procesing,
  85 #'   vol. 11, No 1, pp.1-30. doi:10.1142/S0219691313500033
  86 #'
  87 #' @examples
  88 #' \dontrun{
  89 #' # WER distances computations are too long for CRAN (for now)
  90 #'
  91 #' # Random series around cos(x,2x,3x)/sin(x,2x,3x)
  92 #' x = seq(0,500,0.05)
  93 #' L = length(x) #10001
  94 #' ref_series = matrix( c(cos(x),cos(2*x),cos(3*x),sin(x),sin(2*x),sin(3*x)), ncol=6 )
  95 #' library(wmtsa)
  96 #' series = do.call( cbind, lapply( 1:6, function(i)
  97 #'   do.call(cbind, wmtsa::wavBootstrap(ref_series[,i], n.realization=400)) ) )
  98 #' #dim(series) #c(2400,10001)
  99 #' res_ascii = claws(series, K1=60, K2=6, 200, verbose=TRUE)
 100 #'
 101 #' # Same example, from CSV file
 102 #' csv_file = "/tmp/epclust_series.csv"
 103 #' write.table(series, csv_file, sep=",", row.names=FALSE, col.names=FALSE)
 104 #' res_csv = claws(csv_file, K1=60, K2=6, 200)
 105 #'
 106 #' # Same example, from binary file
 107 #' bin_file <- "/tmp/epclust_series.bin"
 108 #' nbytes <- 8
 109 #' endian <- "little"
 110 #' binarize(csv_file, bin_file, 500, nbytes, endian)
 111 #' getSeries <- function(indices) getDataInFile(indices, bin_file, nbytes, endian)
 112 #' res_bin <- claws(getSeries, K1=60, K2=6, 200)
 113 #' unlink(csv_file)
 114 #' unlink(bin_file)
 115 #'
 116 #' # Same example, from SQLite database
 117 #' library(DBI)
 118 #' series_db <- dbConnect(RSQLite::SQLite(), "file::memory:")
 119 #' # Prepare data.frame in DB-format
 120 #' n <- nrow(series)
 121 #' time_values <- data.frame(
 122 #'   id = rep(1:n,each=L),
 123 #'   time = rep( as.POSIXct(1800*(0:n),"GMT",origin="2001-01-01"), L ),
 124 #'   value = as.double(t(series)) )
 125 #' dbWriteTable(series_db, "times_values", times_values)
 126 #' # Fill associative array, map index to identifier
 127 #' indexToID_inDB <- as.character(
 128 #'   dbGetQuery(series_db, 'SELECT DISTINCT id FROM time_values')[,"id"] )
 129 #' serie_length <- as.integer( dbGetQuery(series_db,
 130 #'   paste("SELECT COUNT * FROM time_values WHERE id == ",indexToID_inDB[1],sep="")) )
 131 #' getSeries <- function(indices) {
 132 #'   request <- "SELECT id,value FROM times_values WHERE id in ("
 133 #'   for (i in indices)
 134 #'     request <- paste(request, indexToID_inDB[i], ",", sep="")
 135 #'   request <- paste(request, ")", sep="")
 136 #'   df_series <- dbGetQuery(series_db, request)
 137 #'   if (length(df_series) >= 1)
 138 #'     as.matrix(df_series[,"value"], nrow=serie_length)
 139 #'   else
 140 #'     NULL
 141 #' }
 142 #' res_db = claws(getSeries, K1=60, K2=6, 200))
 143 #' dbDisconnect(series_db)
 144 #'
 145 #' # All results should be the same:
 146 #' library(digest)
 147 #' digest::sha1(res_ascii)
 148 #' digest::sha1(res_csv)
 149 #' digest::sha1(res_bin)
 150 #' digest::sha1(res_db)
 151 #' }
 152 #' @export
 153 claws <- function(getSeries, K1, K2, nb_series_per_chunk,
 154     algoClust1=function(data,K) cluster::pam(t(data),K,diss=FALSE,pamonce=1)$id.med,
 155     algoClust2=function(dists,K) cluster::pam(dists,K,diss=TRUE,pamonce=1)$id.med,
 156     wav_filt="d8", contrib_type="absolute", WER="end", nvoice=4, random=TRUE,
 157     ntasks=1, ncores_tasks=1, ncores_clust=4, sep=",", nbytes=4,
 158     endian=.Platform$endian, verbose=FALSE, parll=TRUE)
 159 {
 160     # Check/transform arguments
 161     if (!is.matrix(series) && !bigmemory::is.big.matrix(series)
 162         && !is.function(series)
 163         && !methods::is(series,"connection") && !is.character(series))
 164     {
 165         stop("'series': [big]matrix, function, file or valid connection (no NA)")
 166     }
 167     K1 <- .toInteger(K1, function(x) x>=2)
 168     K2 <- .toInteger(K2, function(x) x>=2)
 169     nb_series_per_chunk <- .toInteger(nb_series_per_chunk, function(x) x>=1)
 170     # K1 (number of clusters at step 1) cannot exceed nb_series_per_chunk, because we will need
 171     # to load K1 series in memory for clustering stage 2.
 172     if (K1 > nb_series_per_chunk)
 173         stop("'K1' cannot exceed 'nb_series_per_chunk'")
 174     random <- .toLogical(random)
 175     tryCatch( {ignored <- wavelets::wt.filter(wav_filt)},
 176         error = function(e) stop("Invalid wavelet filter; see ?wavelets::wt.filter") )
 177     ctypes = c("relative","absolute","logit")
 178     contrib_type = ctypes[ pmatch(contrib_type,ctypes) ]
 179     if (is.na(contrib_type))
 180         stop("'contrib_type' in {'relative','absolute','logit'}")
 181     if (WER!="end" && WER!="mix")
 182         stop("'WER': in {'end','mix'}")
 183     random <- .toLogical(random)
 184     ntasks <- .toInteger(ntasks, function(x) x>=1)
 185     ncores_tasks <- .toInteger(ncores_tasks, function(x) x>=1)
 186     ncores_clust <- .toInteger(ncores_clust, function(x) x>=1)
 187     if (!is.character(sep))
 188         stop("'sep': character")
 189     nbytes <- .toInteger(nbytes, function(x) x==4 || x==8)
 190     verbose <- .toLogical(verbose)
 191     parll <- .toLogical(parll)
 192
 193     # Binarize series if it is not a function; the aim is to always use a function,
 194     # to uniformize treatments. An equally good alternative would be to use a file-backed
 195     # bigmemory::big.matrix, but it would break the "all-is-function" pattern.
 196     if (!is.function(series))
 197     {
 198         if (verbose)
 199             cat("...Serialize time-series (or retrieve past binary file)\n")
 200         series_file = ".series.epclust.bin"
 201         if (!file.exists(series_file))
 202             binarize(series, series_file, nb_series_per_chunk, sep, nbytes, endian)
 203         getSeries = function(inds) getDataInFile(inds, series_file, nbytes, endian)
 204     }
 205     else
 206         getSeries = series
 207
 208     # Serialize all computed wavelets contributions into a file
 209     contribs_file = ".contribs.epclust.bin"
 210     index = 1
 211     nb_curves = 0
 212     if (verbose)
 213         cat("...Compute contributions and serialize them (or retrieve past binary file)\n")
 214     if (!file.exists(contribs_file))
 215     {
 216         nb_curves = binarizeTransform(getSeries,
 217             function(curves) curvesToContribs(curves, wav_filt, contrib_type),
 218             contribs_file, nb_series_per_chunk, nbytes, endian)
 219     }
 220     else
 221     {
 222         # TODO: duplicate from getDataInFile() in de_serialize.R
 223         contribs_size = file.info(contribs_file)$size #number of bytes in the file
 224         contrib_length = readBin(contribs_file, "integer", n=1, size=8, endian=endian)
 225         nb_curves = (contribs_size-8) / (nbytes*contrib_length)
 226     }
 227     getContribs = function(indices) getDataInFile(indices, contribs_file, nbytes, endian)
 228
 229     # A few sanity checks: do not continue if too few data available.
 230     if (nb_curves < K2)
 231         stop("Not enough data: less series than final number of clusters")
 232     nb_series_per_task = round(nb_curves / ntasks)
 233     if (nb_series_per_task < K2)
 234         stop("Too many tasks: less series in one task than final number of clusters")
 235
 236     # Generate a random permutation of 1:N (if random==TRUE);
 237     # otherwise just use arrival (storage) order.
 238     indices_all = if (random) sample(nb_curves) else seq_len(nb_curves)
 239     # Split (all) indices into ntasks groups of ~same size
 240     indices_tasks = lapply(seq_len(ntasks), function(i) {
 241         upper_bound = ifelse( i<ntasks, min(nb_series_per_task*i,nb_curves), nb_curves )
 242         indices_all[((i-1)*nb_series_per_task+1):upper_bound]
 243     })
 244
 245     if (parll && ntasks>1)
 246     {
 247         # Initialize parallel runs: outfile="" allow to output verbose traces in the console
 248         # under Linux. All necessary variables are passed to the workers.
 249         cl = parallel::makeCluster(ncores_tasks, outfile="")
 250         parallel::clusterExport(cl, envir = environment(),
 251             varlist = c("getSeries","getContribs","K1","K2","algoClust1","algoClust2",
 252             "nb_series_per_chunk","ncores_clust","nvoice","nbytes","endian","verbose","parll"))
 253     }
 254
 255     # This function achieves one complete clustering task, divided in stage 1 + stage 2.
 256     # stage 1: n indices  --> clusteringTask1(...) --> K1 medoids
 257     # stage 2: K1 medoids --> clusteringTask2(...) --> K2 medoids,
 258     # where n = N / ntasks, N being the total number of curves.
 259     runTwoStepClustering = function(inds)
 260     {
 261         # When running in parallel, the environment is blank: we need to load the required
 262         # packages, and pass useful variables.
 263         if (parll && ntasks>1)
 264             require("epclust", quietly=TRUE)
 265         indices_medoids = clusteringTask1(inds, getContribs, K1, algoClust1,
 266             nb_series_per_chunk, ncores_clust, verbose, parll)
 267         if (WER=="mix")
 268         {
 269             indices_medoids = clusteringTask2(indices_medoids, getSeries, K2, algoClust2,
 270                 nb_series_per_chunk, nvoice, nbytes, endian, ncores_clust, verbose, parll)
 271         }
 272         indices_medoids
 273     }
 274
 275     if (verbose)
 276     {
 277         message = paste("...Run ",ntasks," x stage 1", sep="")
 278         if (WER=="mix")
 279             message = paste(message," + stage 2", sep="")
 280         cat(paste(message,"\n", sep=""))
 281     }
 282
 283     # As explained above, indices will be assigned to ntasks*K1 medoids indices [if WER=="end"],
 284     # or nothing (empty vector) if WER=="mix"; in this case, synchrones are stored in a file.
 285     indices_medoids_all <-
 286         if (parll && ntasks>1)
 287             unlist( parallel::parLapply(cl, indices_tasks, runTwoStepClustering) )
 288         else
 289             unlist( lapply(indices_tasks, runTwoStepClustering) )
 290
 291     if (parll && ntasks>1)
 292         parallel::stopCluster(cl)
 293
 294     # Right before the final stage, input data still is the initial set of curves, referenced
 295     # by the ntasks*[K1 or K2] medoids indices.
 296
 297     # Run last clustering tasks to obtain only K2 medoids indices, from ntasks*[K1 or K2]
 298     # indices, depending wether WER=="end" or WER=="mix"
 299     if (verbose)
 300         cat("...Run final // stage 1 + stage 2\n")
 301     indices_medoids = clusteringTask1(indices_medoids_all, getContribs, K1, algoClust1,
 302         nb_series_per_chunk, ncores_tasks*ncores_clust, verbose, parll)
 303     indices_medoids = clusteringTask2(indices_medoids, getContribs, K2, algoClust2,
 304         nb_series_per_chunk, nvoice, nbytes, endian, ncores_tasks*ncores_clust, verbose, parll)
 305
 306     # Compute synchrones
 307     medoids = getSeries(indices_medoids)
 308     synchrones = computeSynchrones(medoids, getSeries, nb_curves, nb_series_per_chunk,
 309         ncores_tasks*ncores_clust, verbose, parll)
 310
 311     # NOTE: no need to use big.matrix here, since there are only K2 << K1 << N remaining curves
 312     list("medoids"=medoids, "ranks"=indices_medoids, "synchrones"=synchrones)
 313 }