epclust/R/main.R

   1 #' @include defaults.R
   2
   3 #' @title Cluster power curves with PAM in parallel
   4 #'
   5 #' @description Groups electricity power curves (or any series of similar nature) by applying PAM
   6 #' algorithm in parallel to chunks of size \code{nb_series_per_chunk}
   7 #'
   8 #' @param data Access to the data, which can be of one of the three following types:
   9 #' \itemize{
  10 #'   \item data.frame: each line contains its ID in the first cell, and all values after
  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 two arguments: the start index
  13 #'     (start) and number of curves (n); see example in package vignette.
  14 #' }
  15 #' @param K1 Number of super-consumers to be found after stage 1 (K1 << N)
  16 #' @param K2 Number of clusters to be found after stage 2 (K2 << K1)
  17 #' @param ntasks Number of tasks (parallel iterations to obtain K1 medoids); default: 1.
  18 #'   Note: ntasks << N, so that N is "roughly divisible" by N (number of series)
  19 #' @param nb_series_per_chunk (Maximum) number of series in each group, inside a task
  20 #' @param min_series_per_chunk Minimum number of series in each group
  21 #' @param writeTmp Function to write temporary wavelets coefficients (+ identifiers);
  22 #'   see defaults in defaults.R
  23 #' @param readTmp Function to read temporary wavelets coefficients (see defaults.R)
  24 #' @param wf Wavelet transform filter; see ?wt.filter. Default: haar
  25 #' @param WER "end" to apply stage 2 after stage 1 has iterated and finished, or "mix"
  26 #'   to apply it after every stage 1
  27 #' @param ncores_tasks number of parallel tasks (1 to disable: sequential tasks)
  28 #' @param ncores_clust number of parallel clusterings in one task
  29 #'
  30 #' @return A data.frame of the final medoids curves (identifiers + values)
  31 #'
  32 #' @examples
  33 #' getData = function(start, n) {
  34 #'   con = dbConnect(drv = RSQLite::SQLite(), dbname = "mydata.sqlite")
  35 #'   df = dbGetQuery(con, paste(
  36 #'     "SELECT * FROM times_values GROUP BY id OFFSET ",start,
  37 #'     "LIMIT ", n, " ORDER BY date", sep=""))
  38 #'   return (df)
  39 #' }
  40 #' cl = epclust(getData, K1=200, K2=15, ntasks=1000, nb_series_per_chunk=5000, WER="mix")
  41 #' @export
  42 epclust = function(data, K1, K2,
  43     ntasks=1, nb_series_per_chunk=50*K1, min_series_per_chunk=5*K1,
  44     writeTmp=defaultWriteTmp, readTmp=defaultReadTmp, wf="haar", WER="end",
  45     ncores_tasks=1, ncores_clust=4)
  46 {
  47     #TODO: setRefClass(...) to avoid copy data:
  48     #http://stackoverflow.com/questions/2603184/r-pass-by-reference
  49
  50     #0) check arguments
  51     if (!is.data.frame(data) && !is.function(data))
  52         tryCatch(
  53             {
  54                 if (is.character(data))
  55                 {
  56                     data_con = file(data, open="r")
  57                 } else if (!isOpen(data))
  58                 {
  59                     open(data)
  60                     data_con = data
  61                 }
  62             },
  63             error="data should be a data.frame, a function or a valid connection")
  64     if (!is.integer(K) || K < 2)
  65         stop("K should be an integer greater or equal to 2")
  66     if (!is.integer(nb_series_per_chunk) || nb_series_per_chunk < K)
  67         stop("nb_series_per_chunk should be an integer greater or equal to K")
  68     if (!is.function(writeTmp) || !is.function(readTmp))
  69         stop("read/writeTmp should be functional (see defaults.R)")
  70     if (WER!="end" && WER!="mix")
  71         stop("WER takes values in {'end','mix'}")
  72     #concerning ncores, any non-integer type will be treated as "use parallel:detectCores()/4"
  73
  74     #1) acquire data (process curves, get as coeffs)
  75     #TODO: for data.frame and custom function, run in parallel (connections are sequential[?!])
  76     index = 1
  77     nb_curves = 0
  78     repeat
  79     {
  80         coeffs_chunk = NULL
  81         if (is.data.frame(data))
  82         {
  83             #full data matrix
  84             if (index < nrow(data))
  85             {
  86                 coeffs_chunk = curvesToCoeffs(
  87                     data[index:(min(index+nb_series_per_chunk-1,nrow(data))),], wf)
  88             }
  89         } else if (is.function(data))
  90         {
  91             #custom user function to retrieve next n curves, probably to read from DB
  92             coeffs_chunk = curvesToCoeffs( data(index, nb_series_per_chunk), wf )
  93         } else
  94         {
  95             #incremental connection
  96             #TODO: find a better way to parse than using a temp file
  97             ascii_lines = readLines(data_con, nb_series_per_chunk)
  98             if (length(ascii_lines > 0))
  99             {
 100                 series_chunk_file = ".tmp/series_chunk"
 101                 writeLines(ascii_lines, series_chunk_file)
 102                 coeffs_chunk = curvesToCoeffs( read.csv(series_chunk_file), wf )
 103             }
 104         }
 105         if (is.null(coeffs_chunk))
 106             break
 107         writeTmp(coeffs_chunk)
 108         nb_curves = nb_curves + nrow(coeffs_chunk)
 109         index = index + nb_series_per_chunk
 110     }
 111     if (exists(data_con))
 112         close(data_con)
 113     if (nb_curves < min_series_per_chunk)
 114         stop("Not enough data: less rows than min_series_per_chunk!")
 115
 116     #2) process coeffs (by nb_series_per_chunk) and cluster them in parallel
 117     library(parallel)
 118     cl_tasks = parallel::makeCluster(ncores_tasks)
 119     #Nothing to export because each worker retrieve and put data from/on files (or DB)
 120     #parallel::clusterExport(cl=cl, varlist=c("nothing","to","export"), envir=environment())
 121     #TODO: be careful of writing to a new temp file, then flush initial one, then re-use it...
 122     res_tasks = parallel::parSapply(cl_tasks, 1:ntasks, function() {
 123         cl_clust = parallel::makeCluster(ncores_clust)
 124         repeat
 125         {
 126             #while there are jobs to do
 127             #(i.e. size of tmp "file" is greater than ntasks * nb_series_per_chunk)
 128             nb_workers = nb_curves %/% nb_series_per_chunk
 129             indices = list()
 130             #indices[[i]] == (start_index,number_of_elements)
 131             for (i in 1:nb_workers)
 132                 indices[[i]] = c(nb_series_per_chunk*(i-1)+1, nb_series_per_chunk)
 133             remainder = nb_curves %% nb_series_per_chunk
 134             if (remainder >= min_series_per_chunk)
 135             {
 136                 nb_workers = nb_workers + 1
 137                 indices[[nb_workers]] = c(nb_curves-remainder+1, nb_curves)
 138             } else if (remainder > 0)
 139             {
 140                 #spread the load among other workers
 141                 #...
 142             }
 143             res_clust = parallel::parSapply(cl, indices, processChunk, K, WER=="mix")
 144             #C) flush tmp file (current parallel processes will write in it)
 145         }
 146         parallel:stopCluster(cl_clust)
 147     })
 148     parallel::stopCluster(cl_tasks)
 149
 150     #3) readTmp last results, apply PAM on it, and return medoids + identifiers
 151     final_coeffs = readTmp(1, nb_series_per_chunk)
 152     if (nrow(final_coeffs) == K)
 153     {
 154         return ( list( medoids=coeffsToCurves(final_coeffs[,2:ncol(final_coeffs)]),
 155             ids=final_coeffs[,1] ) )
 156     }
 157     pam_output = getClusters(as.matrix(final_coeffs[,2:ncol(final_coeffs)]), K)
 158     medoids = coeffsToCurves(pam_output$medoids, wf)
 159     ids = final_coeffs[,1] [pam_output$ranks]
 160
 161     #4) apply stage 2 (in parallel ? inside task 2) ?)
 162     if (WER == "end")
 163     {
 164         #from center curves, apply stage 2...
 165         #TODO:
 166     }
 167
 168     return (list(medoids=medoids, ids=ids))
 169 }
 170
 171 processChunk = function(indice, K, WER)
 172 {
 173     #1) retrieve data
 174     coeffs = readTmp(indice[1], indice[2])
 175     #2) cluster
 176     cl = getClusters(as.matrix(coeffs[,2:ncol(coeffs)]), K)
 177     #3) WER (optional)
 178     #TODO:
 179 }
 180
 181 #TODO: difficulté : retrouver courbe à partir de l'identifiant (DB ok mais le reste ?)
 182 #aussi : que passe-t-on aux noeuds ? curvesToCoeffs en // ?
 183 #enfin : WER ?!