epclust/R/main.R

   1 #' CLAWS: CLustering with wAvelets and Wer distanceS
   2 #'
   3 #' Groups electricity power curves (or any series of similar nature) by applying PAM
   4 #' algorithm in parallel to chunks of size \code{nb_series_per_chunk}. Input series
   5 #' must be sampled on the same time grid, no missing values.
   6 #'
   7 #' @param getSeries Access to the (time-)series, which can be of one of the three
   8 #'   following types:
   9 #'   \itemize{
  10 #'     \item matrix: each line contains all the values for one time-serie, ordered by time
  11 #'     \item connection: any R connection object (e.g. a file) providing lines as described above
  12 #'     \item function: a custom way to retrieve the curves; it has only one argument:
  13 #'       the indices of the series to be retrieved. See examples
  14 #'   }
  15 #' @inheritParams clustering
  16 #' @param K1 Number of super-consumers to be found after stage 1 (K1 << N)
  17 #' @param K2 Number of clusters to be found after stage 2 (K2 << K1)
  18 #' @param wf Wavelet transform filter; see ?wavelets::wt.filter
  19 #' @param ctype Type of contribution: "relative" or "absolute" (or any prefix)
  20 #' @param WER "end" to apply stage 2 after stage 1 has fully iterated, or "mix" to apply stage 2
  21 #'   at the end of each task
  22 #' @param random TRUE (default) for random chunks repartition
  23 #' @param ntasks Number of tasks (parallel iterations to obtain K1 medoids); default: 1.
  24 #'   Note: ntasks << N, so that N is "roughly divisible" by N (number of series)
  25 #' @param ncores_tasks "MPI" number of parallel tasks (1 to disable: sequential tasks)
  26 #' @param ncores_clust "OpenMP" number of parallel clusterings in one task
  27 #' @param nb_series_per_chunk (~Maximum) number of series in each group, inside a task
  28 #' @param min_series_per_chunk Minimum number of series in each group
  29 #' @param sep Separator in CSV input file (if any provided)
  30 #' @param nbytes Number of bytes to serialize a floating-point number; 4 or 8
  31 #' @param endian Endianness to use for (de)serialization. Use "little" or "big" for portability
  32 #' @param verbose Level of verbosity (0/FALSE for nothing or 1/TRUE for all; devel stage)
  33 #'
  34 #' @return A matrix of the final medoids curves (K2) in rows
  35 #'
  36 #' @examples
  37 #' \dontrun{
  38 #' # WER distances computations are a bit too long for CRAN (for now)
  39 #'
  40 #' # Random series around cos(x,2x,3x)/sin(x,2x,3x)
  41 #' x = seq(0,500,0.05)
  42 #' L = length(x) #10001
  43 #' ref_series = matrix( c(cos(x), cos(2*x), cos(3*x), sin(x), sin(2*x), sin(3*x)),
  44 #'   byrow=TRUE, ncol=L )
  45 #' library(wmtsa)
  46 #' series = do.call( rbind, lapply( 1:6, function(i)
  47 #'   do.call(rbind, wmtsa::wavBootstrap(ref_series[i,], n.realization=400)) ) )
  48 #' #dim(series) #c(2400,10001)
  49 #' medoids_ascii = claws(series, K1=60, K2=6, "d8", "rel", nb_series_per_chunk=500)
  50 #'
  51 #' # Same example, from CSV file
  52 #' csv_file = "/tmp/epclust_series.csv"
  53 #' write.table(series, csv_file, sep=",", row.names=FALSE, col.names=FALSE)
  54 #' medoids_csv = claws(csv_file, K1=60, K2=6, "d8", "rel", nb_series_per_chunk=500)
  55 #'
  56 #' # Same example, from binary file
  57 #' bin_file = "/tmp/epclust_series.bin"
  58 #' nbytes = 8
  59 #' endian = "little"
  60 #' epclust::binarize(csv_file, bin_file, 500, nbytes, endian)
  61 #' getSeries = function(indices) getDataInFile(indices, bin_file, nbytes, endian)
  62 #' medoids_bin = claws(getSeries, K1=60, K2=6, "d8", "rel", nb_series_per_chunk=500)
  63 #' unlink(csv_file)
  64 #' unlink(bin_file)
  65 #'
  66 #' # Same example, from SQLite database
  67 #' library(DBI)
  68 #' series_db <- dbConnect(RSQLite::SQLite(), "file::memory:")
  69 #' # Prepare data.frame in DB-format
  70 #' n = nrow(series)
  71 #' time_values = data.frame(
  72 #'   id = rep(1:n,each=L),
  73 #'   time = rep( as.POSIXct(1800*(0:n),"GMT",origin="2001-01-01"), L ),
  74 #'   value = as.double(t(series)) )
  75 #' dbWriteTable(series_db, "times_values", times_values)
  76 #' # Fill associative array, map index to identifier
  77 #' indexToID_inDB <- as.character(
  78 #'   dbGetQuery(series_db, 'SELECT DISTINCT id FROM time_values')[,"id"] )
  79 #' getSeries = function(indices) {
  80 #'   request = "SELECT id,value FROM times_values WHERE id in ("
  81 #'   for (i in indices)
  82 #'     request = paste(request, i, ",", sep="")
  83 #'   request = paste(request, ")", sep="")
  84 #'   df_series = dbGetQuery(series_db, request)
  85 #'   # Assume that all series share same length at this stage
  86 #'   ts_length = sum(df_series[,"id"] == df_series[1,"id"])
  87 #'   t( as.matrix(df_series[,"value"], nrow=ts_length) )
  88 #' }
  89 #' medoids_db = claws(getSeries, K1=60, K2=6, "d8", "rel", nb_series_per_chunk=500)
  90 #' dbDisconnect(series_db)
  91 #'
  92 #' # All computed medoids should be the same:
  93 #' digest::sha1(medoids_ascii)
  94 #' digest::sha1(medoids_csv)
  95 #' digest::sha1(medoids_bin)
  96 #' digest::sha1(medoids_db)
  97 #' }
  98 #' @export
  99 claws = function(getSeries, K1, K2,
 100     wf,ctype, #stage 1
 101     WER="end", #stage 2
 102     random=TRUE, #randomize series order?
 103     ntasks=1, ncores_tasks=1, ncores_clust=4, #control parallelism
 104     nb_series_per_chunk=50*K1, min_series_per_chunk=5*K1, #chunk size
 105     sep=",", #ASCII input separator
 106     nbytes=4, endian=.Platform$endian, #serialization (write,read)
 107     verbose=FALSE)
 108 {
 109     # Check/transform arguments
 110     if (!is.matrix(getSeries) && !is.function(getSeries) &&
 111         !methods::is(getSeries, "connection" && !is.character(getSeries)))
 112     {
 113         stop("'getSeries': matrix, function, file or valid connection (no NA)")
 114     }
 115     K1 = .toInteger(K1, function(x) x>=2)
 116     K2 = .toInteger(K2, function(x) x>=2)
 117     if (!is.logical(random))
 118         stop("'random': logical")
 119     tryCatch(
 120         {ignored <- wavelets::wt.filter(wf)},
 121         error = function(e) stop("Invalid wavelet filter; see ?wavelets::wt.filter"))
 122     if (WER!="end" && WER!="mix")
 123         stop("WER takes values in {'end','mix'}")
 124     ntasks = .toInteger(ntasks, function(x) x>=1)
 125     ncores_tasks = .toInteger(ncores_tasks, function(x) x>=1)
 126     ncores_clust = .toInteger(ncores_clust, function(x) x>=1)
 127     nb_series_per_chunk = .toInteger(nb_series_per_chunk, function(x) x>=K1)
 128     min_series_per_chunk = .toInteger(K1, function(x) x>=K1 && x<=nb_series_per_chunk)
 129     if (!is.character(sep))
 130         stop("'sep': character")
 131     nbytes = .toInteger(nbytes, function(x) x==4 || x==8)
 132
 133     # Serialize series if required, to always use a function
 134     bin_dir = ".epclust.bin/"
 135     dir.create(bin_dir, showWarnings=FALSE, mode="0755")
 136     if (!is.function(getSeries))
 137     {
 138         if (verbose)
 139             cat("...Serialize time-series\n")
 140         series_file = paste(bin_dir,"data",sep="") ; unlink(series_file)
 141         binarize(getSeries, series_file, nb_series_per_chunk, sep, nbytes, endian)
 142         getSeries = function(inds) getDataInFile(inds, series_file, nbytes, endian)
 143     }
 144
 145     # Serialize all computed wavelets contributions onto a file
 146     contribs_file = paste(bin_dir,"contribs",sep="") ; unlink(contribs_file)
 147     index = 1
 148     nb_curves = 0
 149     if (verbose)
 150         cat("...Compute contributions and serialize them\n")
 151     repeat
 152     {
 153         series = getSeries((index-1)+seq_len(nb_series_per_chunk))
 154         if (is.null(series))
 155             break
 156         contribs_chunk = curvesToContribs(series, wf, ctype)
 157         binarize(contribs_chunk, contribs_file, nb_series_per_chunk, sep, nbytes, endian)
 158         index = index + nb_series_per_chunk
 159         nb_curves = nb_curves + nrow(contribs_chunk)
 160     }
 161     getContribs = function(indices) getDataInFile(indices, contribs_file, nbytes, endian)
 162
 163     if (nb_curves < min_series_per_chunk)
 164         stop("Not enough data: less rows than min_series_per_chunk!")
 165     nb_series_per_task = round(nb_curves / ntasks)
 166     if (nb_series_per_task < min_series_per_chunk)
 167         stop("Too many tasks: less series in one task than min_series_per_chunk!")
 168
 169     # Cluster contributions in parallel (by nb_series_per_chunk)
 170     indices_all = if (random) sample(nb_curves) else seq_len(nb_curves)
 171     indices_tasks = lapply(seq_len(ntasks), function(i) {
 172         upper_bound = ifelse( i<ntasks, min(nb_series_per_task*i,nb_curves), nb_curves )
 173         indices_all[((i-1)*nb_series_per_task+1):upper_bound]
 174     })
 175     if (verbose)
 176         cat(paste("...Run ",ntasks," x stage 1 in parallel\n",sep=""))
 177 #   cl = parallel::makeCluster(ncores_tasks)
 178 #   parallel::clusterExport(cl, varlist=c("getSeries","getContribs","K1","K2",
 179 #       "nb_series_per_chunk","ncores_clust","synchrones_file","sep","nbytes","endian"),
 180 #       envir = environment())
 181     # 1000*K1 indices [if WER=="end"], or empty vector [if WER=="mix"] --> series on file
 182 #   indices = unlist( parallel::parLapply(cl, indices_tasks, function(inds) {
 183     indices = unlist( lapply(indices_tasks, function(inds) {
 184 #       require("epclust", quietly=TRUE)
 185
 186         browser() #TODO: CONTINUE DEBUG HERE
 187
 188         indices_medoids = clusteringTask(inds,getContribs,K1,nb_series_per_chunk,ncores_clust)
 189         if (WER=="mix")
 190         {
 191             medoids2 = computeClusters2(
 192                 getSeries(indices_medoids), K2, getSeries, nb_series_per_chunk)
 193             binarize(medoids2, synchrones_file, nb_series_per_chunk, sep, nbytes, endian)
 194             return (vector("integer",0))
 195         }
 196         indices_medoids
 197     }) )
 198 #   parallel::stopCluster(cl)
 199
 200     getRefSeries = getSeries
 201     synchrones_file = paste(bin_dir,"synchrones",sep="") ; unlink(synchrones_file)
 202     if (WER=="mix")
 203     {
 204         indices = seq_len(ntasks*K2)
 205         #Now series must be retrieved from synchrones_file
 206         getSeries = function(inds) getDataInFile(inds, synchrones_file, nbytes, endian)
 207         #Contributions must be re-computed
 208         unlink(contribs_file)
 209         index = 1
 210         if (verbose)
 211             cat("...Serialize contributions computed on synchrones\n")
 212         repeat
 213         {
 214             series = getSeries((index-1)+seq_len(nb_series_per_chunk))
 215             if (is.null(series))
 216                 break
 217             contribs_chunk = curvesToContribs(series, wf, ctype)
 218             binarize(contribs_chunk, contribs_file, nb_series_per_chunk, sep, nbytes, endian)
 219             index = index + nb_series_per_chunk
 220         }
 221     }
 222
 223     # Run step2 on resulting indices or series (from file)
 224     if (verbose)
 225         cat("...Run final // stage 1 + stage 2\n")
 226     indices_medoids = clusteringTask(
 227         indices, getContribs, K1, nb_series_per_chunk, ncores_tasks*ncores_clust)
 228     medoids = computeClusters2(getSeries(indices_medoids),K2,getRefSeries,nb_series_per_chunk)
 229
 230     # Cleanup
 231     unlink(bin_dir, recursive=TRUE)
 232
 233     medoids
 234 }
 235
 236 #' curvesToContribs
 237 #'
 238 #' Compute the discrete wavelet coefficients for each series, and aggregate them in
 239 #' energy contribution across scales as described in https://arxiv.org/abs/1101.4744v2
 240 #'
 241 #' @param series Matrix of series (in rows), of size n x L
 242 #' @inheritParams claws
 243 #'
 244 #' @return A matrix of size n x log(L) containing contributions in rows
 245 #'
 246 #' @export
 247 curvesToContribs = function(series, wf, ctype)
 248 {
 249     L = length(series[1,])
 250     D = ceiling( log2(L) )
 251     nb_sample_points = 2^D
 252     cont_types = c("relative","absolute")
 253     ctype = cont_types[ pmatch(ctype,cont_types) ]
 254     t( apply(series, 1, function(x) {
 255         interpolated_curve = spline(1:L, x, n=nb_sample_points)$y
 256         W = wavelets::dwt(interpolated_curve, filter=wf, D)@W
 257         nrj = rev( sapply( W, function(v) ( sqrt( sum(v^2) ) ) ) )
 258         if (ctype=="relative") nrj / sum(nrj) else nrj
 259     }) )
 260 }
 261
 262 # Helper for main function: check integer arguments with functiional conditions
 263 .toInteger <- function(x, condition)
 264 {
 265     if (!is.integer(x))
 266         tryCatch(
 267             {x = as.integer(x)[1]},
 268             error = function(e) paste("Cannot convert argument",substitute(x),"to integer")
 269         )
 270     if (!condition(x))
 271         stop(paste("Argument",substitute(x),"does not verify condition",body(condition)))
 272     x
 273 }