4fdc5ae899c61aa3347e0aba1a843c12a5ecd012
[epclust.git] / 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_items_clust}\cr
15 #' -> K1 medoids indices
16 #' \item optionally, if WER=="mix":\cr
17 #' a. compute WER distances (K1xK1) between medoids\cr
18 #' b. apply the 2nd clustering algorithm\cr
19 #' -> K2 medoids indices
20 #' }
21 #' \item Launch a final task on the aggregated outputs of all previous tasks:
22 #' ntasks*K1 if WER=="end", ntasks*K2 otherwise
23 #' \item Compute synchrones (sum of series within each final group)
24 #' }
25 #'
26 #' The main argument -- \code{series} -- has a quite misleading name, since it can be
27 #' either a [big.]matrix, a CSV file, a connection or a user function to retrieve series.
28 #' When \code{series} is given as a function it must take a single argument,
29 #' 'indices': integer vector equal to the indices of the curves to retrieve;
30 #' see SQLite example.
31 #' WARNING: the return value must be a matrix (in columns), or NULL if no matches.
32 #'
33 #' Note: Since we don't make assumptions on initial data, there is a possibility that
34 #' even when serialized, contributions do not fit in RAM. For example,
35 #' 30e6 series of length 100,000 would lead to a +4Go contribution matrix. Therefore,
36 #' it's safer to place these in (binary) files; that's what we do.
37 #'
38 #' @param series Access to the N (time-)series, which can be of one of the four
39 #' following types:
40 #' \itemize{
41 #' \item [big.]matrix: each column contains the (time-ordered) values of one time-serie
42 #' \item connection: any R connection object providing lines as described above
43 #' \item character: name of a CSV file containing series in rows (no header)
44 #' \item function: a custom way to retrieve the curves; it has only one argument:
45 #' the indices of the series to be retrieved. See SQLite example
46 #' }
47 #' @param K1 Number of clusters to be found after stage 1 (K1 << N)
48 #' @param K2 Number of clusters to be found after stage 2 (K2 << K1)
49 #' @param nb_series_per_chunk Number of series to retrieve in one batch
50 #' @param nb_items_clust Number of items in 1st clustering algorithm input
51 #' @param algoClust1 Clustering algorithm for stage 1. A function which takes (data, K)
52 #' as argument where data is a matrix in columns and K the desired number of clusters,
53 #' and outputs K medoids ranks. Default: PAM.
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.
57 #' @param wav_filt Wavelet transform filter; see ?wavelets::wt.filter
58 #' @param contrib_type Type of contribution: "relative", "logit" or "absolute" (any prefix)
59 #' @param WER "end" to apply stage 2 after stage 1 has fully iterated, or "mix" to apply
60 #' stage 2 at the end of each task
61 #' @param smooth_lvl Smoothing level: odd integer, 1 == no smoothing.
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' == sequential tasks)
68 #' @param ncores_clust Number of parallel clusterings in one task
69 #' @param sep Separator in CSV input file (if any provided)
70 #' @param nbytes 4 or 8 bytes to (de)serialize a floating-point number
71 #' @param endian Endianness for (de)serialization: "little" or "big"
72 #' @param verbose FALSE: nothing printed; TRUE: some execution traces
73 #' @param parll TRUE: run in parallel. FALSE: run sequentially
74 #'
75 #' @return A list:
76 #' \itemize{
77 #' \item medoids: matrix of the final K2 medoids curves
78 #' \item ranks: corresponding indices in the dataset
79 #' \item synchrones: 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 #' parll = FALSE #on this small example, sequential run is faster
91 #'
92 #' # Random series around cos(x,2x,3x)/sin(x,2x,3x)
93 #' x <- seq(0,50,0.05)
94 #' L <- length(x) #1001
95 #' ref_series <- matrix( c(cos(x),cos(2*x),cos(3*x),sin(x),sin(2*x),sin(3*x)), ncol=6 )
96 #' library(wmtsa)
97 #' series <- do.call( cbind, lapply( 1:6, function(i)
98 #' do.call(cbind, wmtsa::wavBootstrap(ref_series[,i], n.realization=40)) ) )
99 #' # Mix series so that all groups are evenly spread
100 #' permut <- (0:239)%%6 * 40 + (0:239)%/%6 + 1
101 #' series = series[,permut]
102 #' #dim(series) #c(240,1001)
103 #' res_ascii <- claws(series, K1=30, K2=6, nb_series_per_chunk=500,
104 #' nb_items_clust=100, random=FALSE, verbose=TRUE, parll=parll)
105 #'
106 #' # Same example, from CSV file
107 #' csv_file <- tempfile(pattern="epclust_series.csv_")
108 #' write.table(t(series), csv_file, sep=",", row.names=FALSE, col.names=FALSE)
109 #' res_csv <- claws(csv_file, 30, 6, 500, 100, random=FALSE, parll=parll)
110 #'
111 #' # Same example, from binary file
112 #' bin_file <- tempfile(pattern="epclust_series.bin_")
113 #' nbytes <- 8
114 #' endian <- "little"
115 #' binarize(csv_file, bin_file, 500, ",", nbytes, endian)
116 #' getSeries <- function(indices) getDataInFile(indices, bin_file, nbytes, endian)
117 #' res_bin <- claws(getSeries, 30, 6, 500, 100, random=FALSE, parll=parll)
118 #' unlink(csv_file)
119 #' unlink(bin_file)
120 #'
121 #' # Same example, from SQLite database
122 #' library(DBI)
123 #' series_db <- dbConnect(RSQLite::SQLite(), "file::memory:")
124 #' # Prepare data.frame in DB-format
125 #' n <- ncol(series)
126 #' times_values <- data.frame(
127 #' id = rep(1:n,each=L),
128 #' time = rep( as.POSIXct(1800*(1:L),"GMT",origin="2001-01-01"), n ),
129 #' value = as.double(series) )
130 #' dbWriteTable(series_db, "times_values", times_values)
131 #' # Fill associative array, map index to identifier
132 #' indexToID_inDB <- as.character(
133 #' dbGetQuery(series_db, 'SELECT DISTINCT id FROM times_values')[,"id"] )
134 #' serie_length <- as.integer( dbGetQuery(series_db,
135 #' paste("SELECT COUNT(*) FROM times_values WHERE id == ",indexToID_inDB[1],sep="")) )
136 #' getSeries <- function(indices) {
137 #' indices = indices[ indices <= length(indexToID_inDB) ]
138 #' if (length(indices) == 0)
139 #' return (NULL)
140 #' request <- "SELECT id,value FROM times_values WHERE id in ("
141 #' for (i in seq_along(indices)) {
142 #' request <- paste(request, indexToID_inDB[ indices[i] ], sep="")
143 #' if (i < length(indices))
144 #' request <- paste(request, ",", sep="")
145 #' }
146 #' request <- paste(request, ")", sep="")
147 #' df_series <- dbGetQuery(series_db, request)
148 #' matrix(df_series[,"value"], nrow=serie_length)
149 #' }
150 #' res_db <- claws(getSeries, 30, 6, 500, 100, random=FALSE, parll=parll)
151 #' dbDisconnect(series_db)
152 #'
153 #' # All results should be equal:
154 #' all(res_ascii$ranks == res_csv$ranks
155 #' & res_ascii$ranks == res_bin$ranks
156 #' & res_ascii$ranks == res_db$ranks)
157 #' }
158 #' @export
159 claws <- function(series, K1, K2, nb_series_per_chunk, nb_items_clust=5*K1,
160 algoClust1=function(data,K) cluster::pam(t(data),K,diss=FALSE,pamonce=1)$id.med,
161 algoClust2=function(dists,K) cluster::pam(dists,K,diss=TRUE,pamonce=1)$id.med,
162 wav_filt="d8", contrib_type="absolute", WER="end", smooth_lvl=3, nvoice=4,
163 random=TRUE, ntasks=1, ncores_tasks=1, ncores_clust=3, sep=",", nbytes=4,
164 endian=.Platform$endian, verbose=FALSE, parll=TRUE)
165 {
166 # Check/transform arguments
167 if (!is.matrix(series) && !bigmemory::is.big.matrix(series)
168 && !is.function(series)
169 && !methods::is(series,"connection") && !is.character(series))
170 {
171 stop("'series': [big]matrix, function, file or valid connection (no NA)")
172 }
173 K1 <- .toInteger(K1, function(x) x>=2)
174 K2 <- .toInteger(K2, function(x) x>=2)
175 nb_series_per_chunk <- .toInteger(nb_series_per_chunk, function(x) x>=1)
176 nb_items_clust <- .toInteger(nb_items_clust, function(x) x>K1)
177 random <- .toLogical(random)
178 tryCatch({ignored <- wavelets::wt.filter(wav_filt)},
179 error=function(e) stop("Invalid wavelet filter; see ?wavelets::wt.filter") )
180 ctypes <- c("relative","absolute","logit")
181 contrib_type <- ctypes[ pmatch(contrib_type,ctypes) ]
182 if (is.na(contrib_type))
183 stop("'contrib_type' in {'relative','absolute','logit'}")
184 if (WER!="end" && WER!="mix")
185 stop("'WER': in {'end','mix'}")
186 random <- .toLogical(random)
187 ntasks <- .toInteger(ntasks, function(x) x>=1)
188 ncores_tasks <- .toInteger(ncores_tasks, function(x) x>=1)
189 ncores_clust <- .toInteger(ncores_clust, function(x) x>=1)
190 if (!is.character(sep))
191 stop("'sep': character")
192 nbytes <- .toInteger(nbytes, function(x) x==4 || x==8)
193 verbose <- .toLogical(verbose)
194 parll <- .toLogical(parll)
195
196 # Binarize series if it is not a function; the aim is to always use a function,
197 # to uniformize treatments. An equally good alternative would be to use a file-backed
198 # bigmemory::big.matrix, but it would break the "all-is-function" pattern.
199 if (!is.function(series))
200 {
201 if (verbose)
202 cat("...Serialize time-series (or retrieve past binary file)\n")
203 series_file <- ".series.epclust.bin"
204 if (!file.exists(series_file))
205 binarize(series, series_file, nb_series_per_chunk, sep, nbytes, endian)
206 getSeries <- function(inds) getDataInFile(inds, series_file, nbytes, endian)
207 }
208 else
209 getSeries <- series
210
211 # Serialize all computed wavelets contributions into a file
212 contribs_file <- ".contribs.epclust.bin"
213 if (verbose)
214 cat("...Compute contributions and serialize them (or retrieve past binary file)\n")
215 if (!file.exists(contribs_file))
216 {
217 nb_curves <- binarizeTransform(getSeries,
218 function(curves) curvesToContribs(curves, wav_filt, contrib_type),
219 contribs_file, nb_series_per_chunk, nbytes, endian)
220 }
221 else
222 {
223 # TODO: duplicate from getDataInFile() in de_serialize.R
224 contribs_size <- file.info(contribs_file)$size #number of bytes in the file
225 contrib_length <- readBin(contribs_file, "integer", n=1, size=8, endian=endian)
226 nb_curves <- (contribs_size-8) / (nbytes*contrib_length)
227 }
228 getContribs <- function(indices) getDataInFile(indices, contribs_file, nbytes, endian)
229
230 # A few sanity checks: do not continue if too few data available.
231 if (nb_curves < K2)
232 stop("Not enough data: less series than final number of clusters")
233 nb_series_per_task <- round(nb_curves / ntasks)
234 if (nb_series_per_task < K2)
235 stop("Too many tasks: less series in one task than final number of clusters")
236
237 # Generate a random permutation of 1:N (if random==TRUE);
238 # otherwise just use arrival (storage) order.
239 indices_all <- if (random) sample(nb_curves) else seq_len(nb_curves)
240 # Split (all) indices into ntasks groups of ~same size
241 indices_tasks <- lapply(seq_len(ntasks), function(i) {
242 upper_bound <- ifelse( i<ntasks, min(nb_series_per_task*i,nb_curves), nb_curves )
243 indices_all[((i-1)*nb_series_per_task+1):upper_bound]
244 })
245
246 if (parll && ntasks>1)
247 {
248 # Initialize parallel runs: outfile="" allow to output verbose traces in the console
249 # under Linux. All necessary variables are passed to the workers.
250 cl <-
251 if (verbose)
252 parallel::makeCluster(ncores_tasks, outfile="")
253 else
254 parallel::makeCluster(ncores_tasks)
255 varlist <- c("ncores_clust","verbose","parll", #task 1 & 2
256 "K1","getContribs","algoClust1","nb_items_clust") #task 1
257 if (WER=="mix")
258 {
259 # Add variables for task 2
260 varlist <- c(varlist, "K2","getSeries","algoClust2","nb_series_per_chunk",
261 "smooth_lvl","nvoice","nbytes","endian")
262 }
263 parallel::clusterExport(cl, varlist, envir <- environment())
264 }
265
266 # This function achieves one complete clustering task, divided in stage 1 + stage 2.
267 # stage 1: n indices --> clusteringTask1(...) --> K1 medoids (indices)
268 # stage 2: K1 indices --> K1xK1 WER distances --> clusteringTask2(...) --> K2 medoids,
269 # where n == N / ntasks, N being the total number of curves.
270 runTwoStepClustering <- function(inds)
271 {
272 # When running in parallel, the environment is blank: we need to load the required
273 # packages, and pass useful variables.
274 if (parll && ntasks>1)
275 require("epclust", quietly=TRUE)
276 indices_medoids <- clusteringTask1(inds, getContribs, K1, algoClust1,
277 nb_items_clust, ncores_clust, verbose, parll)
278 if (WER=="mix")
279 {
280 indices_medoids <- clusteringTask2(indices_medoids, getSeries, K2, algoClust2,
281 nb_series_per_chunk,smooth_lvl,nvoice,nbytes,endian,ncores_clust,verbose,parll)
282 }
283 indices_medoids
284 }
285
286 if (verbose)
287 {
288 message <- paste("...Run ",ntasks," x stage 1", sep="")
289 if (WER=="mix")
290 message <- paste(message," + stage 2", sep="")
291 cat(paste(message,"\n", sep=""))
292 }
293
294 # As explained above, we obtain after all runs ntasks*[K1 or K2] medoids indices,
295 # depending whether WER=="end" or "mix", respectively.
296 indices_medoids_all <-
297 if (parll && ntasks>1)
298 unlist( parallel::parLapply(cl, indices_tasks, runTwoStepClustering) )
299 else
300 unlist( lapply(indices_tasks, runTwoStepClustering) )
301
302 if (parll && ntasks>1)
303 parallel::stopCluster(cl)
304
305 # For the last stage, ncores_tasks*(ncores_clusts+1) cores should be available:
306 # - ntasks for level 1 parallelism
307 # - ntasks*ncores_clust for level 2 parallelism,
308 # but since an extension MPI <--> tasks / OpenMP <--> sub-tasks is on the way,
309 # it's better to just re-use ncores_clust
310 ncores_last_stage <- ncores_clust
311
312 # Run last clustering tasks to obtain only K2 medoids indices
313 if (verbose)
314 cat("...Run final // stage 1 + stage 2\n")
315 indices_medoids <- clusteringTask1(indices_medoids_all, getContribs, K1, algoClust1,
316 nb_items_clust, ncores_tasks*ncores_clust, verbose, parll)
317
318 indices_medoids <- clusteringTask2(indices_medoids, getSeries, K2, algoClust2,
319 nb_series_per_chunk,smooth_lvl,nvoice,nbytes,endian,ncores_last_stage,verbose,parll)
320
321 # Compute synchrones, that is to say the cumulated power consumptions for each of the K2
322 # final groups.
323 medoids <- getSeries(indices_medoids)
324 synchrones <- computeSynchrones(medoids, getSeries, nb_curves, nb_series_per_chunk,
325 ncores_last_stage, verbose, parll)
326
327 # NOTE: no need to use big.matrix here, since there are only K2 << K1 << N remaining curves
328 list("medoids"=medoids, "ranks"=indices_medoids, "synchrones"=synchrones)
329 }