'update'
[epclust.git] / epclust / R / clustering.R
1 #' @name clustering
2 #' @rdname clustering
3 #' @aliases clusteringTask1 computeClusters1 computeClusters2
4 #'
5 #' @title Two-stage clustering, withing one task (see \code{claws()})
6 #'
7 #' @description \code{clusteringTask1()} runs one full stage-1 task, which consists in
8 #' iterated stage 1 clustering (on nb_curves / ntasks energy contributions, computed
9 #' through discrete wavelets coefficients).
10 #' \code{clusteringTask2()} runs a full stage-2 task, which consists in synchrones
11 #' and then WER distances computations, before applying the clustering algorithm.
12 #' \code{computeClusters1()} and \code{computeClusters2()} correspond to the atomic
13 #' clustering procedures respectively for stage 1 and 2. The former applies the
14 #' clustering algorithm (PAM) on a contributions matrix, while the latter clusters
15 #' a chunk of series inside one task (~max nb_series_per_chunk)
16 #'
17 #' @param indices Range of series indices to cluster in parallel (initial data)
18 #' @param getContribs Function to retrieve contributions from initial series indices:
19 #' \code{getContribs(indices)} outpus a contributions matrix
20 #' @param contribs matrix of contributions (e.g. output of \code{curvesToContribs()})
21 #' @inheritParams computeSynchrones
22 #' @inheritParams claws
23 #'
24 #' @return For \code{clusteringTask1()} and \code{computeClusters1()}, the indices of the
25 #' computed (K1) medoids. Indices are irrelevant for stage 2 clustering, thus
26 #' \code{computeClusters2()} outputs a big.matrix of medoids
27 #' (of size limited by nb_series_per_chunk)
28 NULL
29
30 #' @rdname clustering
31 #' @export
32 clusteringTask1 = function(
33 indices, getContribs, K1, nb_series_per_chunk, ncores_clust=1, verbose=FALSE, parll=TRUE)
34 {
35 if (verbose)
36 cat(paste("*** Clustering task on ",length(indices)," lines\n", sep=""))
37
38 wrapComputeClusters1 = function(inds) {
39 if (parll)
40 require("epclust", quietly=TRUE)
41 if (verbose)
42 cat(paste(" computeClusters1() on ",length(inds)," lines\n", sep=""))
43 inds[ computeClusters1(getContribs(inds), K1) ]
44 }
45
46 if (parll)
47 {
48 cl = parallel::makeCluster(ncores_clust)
49 parallel::clusterExport(cl, varlist=c("getContribs","K1","verbose"), envir=environment())
50 }
51 while (length(indices) > K1)
52 {
53 indices_workers = .spreadIndices(indices, nb_series_per_chunk)
54 if (parll)
55 indices = unlist( parallel::parLapply(cl, indices_workers, wrapComputeClusters1) )
56 else
57 indices = unlist( lapply(indices_workers, wrapComputeClusters1) )
58 }
59 if (parll)
60 parallel::stopCluster(cl)
61
62 indices #medoids
63 }
64
65 #' @rdname clustering
66 #' @export
67 clusteringTask2 = function(medoids, K2,
68 getRefSeries, nb_ref_curves, nb_series_per_chunk, ncores_clust=1,verbose=FALSE,parll=TRUE)
69 {
70 if (nrow(medoids) <= K2)
71 return (medoids)
72 synchrones = computeSynchrones(medoids,
73 getRefSeries, nb_ref_curves, nb_series_per_chunk, ncores_clust, verbose, parll)
74 distances = computeWerDists(synchrones, ncores_clust, verbose, parll)
75 # PAM in package 'cluster' cannot take big.matrix in input: need to cast it
76 mat_dists = matrix(nrow=K1, ncol=K1)
77 for (i in seq_len(K1))
78 mat_dists[i,] = distances[i,]
79 medoids[ computeClusters2(mat_dists,K2), ]
80 }
81
82 #' @rdname clustering
83 #' @export
84 computeClusters1 = function(contribs, K1)
85 cluster::pam(contribs, K1, diss=FALSE)$id.med
86
87 #' @rdname clustering
88 #' @export
89 computeClusters2 = function(distances, K2)
90 cluster::pam(distances, K2, diss=TRUE)$id.med
91
92 #' computeSynchrones
93 #'
94 #' Compute the synchrones curves (sum of clusters elements) from a matrix of medoids,
95 #' using L2 distances.
96 #'
97 #' @param medoids big.matrix of medoids (curves of same length as initial series)
98 #' @param getRefSeries Function to retrieve initial series (e.g. in stage 2 after series
99 #' have been replaced by stage-1 medoids)
100 #' @param nb_ref_curves How many reference series? (This number is known at this stage)
101 #' @inheritParams claws
102 #'
103 #' @return A big.matrix of size K1 x L where L = data_length
104 #'
105 #' @export
106 computeSynchrones = function(medoids, getRefSeries,
107 nb_ref_curves, nb_series_per_chunk, ncores_clust=1,verbose=FALSE,parll=TRUE)
108 {
109 computeSynchronesChunk = function(indices)
110 {
111 if (verbose)
112 cat(paste("--- Compute synchrones for ",length(indices)," lines\n", sep=""))
113 ref_series = getRefSeries(indices)
114 #get medoids indices for this chunk of series
115 for (i in seq_len(nrow(ref_series)))
116 {
117 j = which.min( rowSums( sweep(medoids, 2, ref_series[i,], '-')^2 ) )
118 if (parll)
119 synchronicity::lock(m)
120 synchrones[j,] = synchrones[j,] + ref_series[i,]
121 counts[j,1] = counts[j,1] + 1
122 if (parll)
123 synchronicity::unlock(m)
124 }
125 }
126
127 K = nrow(medoids)
128 # Use bigmemory (shared==TRUE by default) + synchronicity to fill synchrones in //
129 # TODO: if size > RAM (not our case), use file-backed big.matrix
130 synchrones = bigmemory::big.matrix(nrow=K,ncol=ncol(medoids),type="double",init=0.)
131 counts = bigmemory::big.matrix(nrow=K,ncol=1,type="double",init=0)
132 # synchronicity is only for Linux & MacOS; on Windows: run sequentially
133 parll = (requireNamespace("synchronicity",quietly=TRUE)
134 && parll && Sys.info()['sysname'] != "Windows")
135 if (parll)
136 m <- synchronicity::boost.mutex()
137
138 if (parll)
139 {
140 cl = parallel::makeCluster(ncores_clust)
141 parallel::clusterExport(cl,
142 varlist=c("synchrones","counts","verbose","medoids","getRefSeries"),
143 envir=environment())
144 }
145
146 indices_workers = .spreadIndices(seq_len(nb_ref_curves), nb_series_per_chunk)
147 ignored <-
148 if (parll)
149 parallel::parLapply(indices_workers, computeSynchronesChunk)
150 else
151 lapply(indices_workers, computeSynchronesChunk)
152
153 if (parll)
154 parallel::stopCluster(cl)
155
156 #TODO: can we avoid this loop? ( synchrones = sweep(synchrones, 1, counts, '/') )
157 for (i in seq_len(K))
158 synchrones[i,] = synchrones[i,] / counts[i,1]
159 #NOTE: odds for some clusters to be empty? (when series already come from stage 2)
160 # ...maybe; but let's hope resulting K1' be still quite bigger than K2
161 noNA_rows = sapply(seq_len(K), function(i) all(!is.nan(synchrones[i,])))
162 if (all(noNA_rows))
163 return (synchrones)
164 # Else: some clusters are empty, need to slice synchrones
165 synchrones[noNA_rows,]
166 }
167
168 #' computeWerDists
169 #'
170 #' Compute the WER distances between the synchrones curves (in rows), which are
171 #' returned (e.g.) by \code{computeSynchrones()}
172 #'
173 #' @param synchrones A big.matrix of synchrones, in rows. The series have same length
174 #' as the series in the initial dataset
175 #' @inheritParams claws
176 #'
177 #' @return A big.matrix of size K1 x K1
178 #'
179 #' @export
180 computeWerDists = function(synchrones, ncores_clust=1,verbose=FALSE,parll=TRUE)
181 {
182
183
184
185 #TODO: re-organize to call computeWerDist(x,y) [C] (in //?) from two indices + big.matrix
186
187
188 n <- nrow(synchrones)
189 delta <- ncol(synchrones)
190 #TODO: automatic tune of all these parameters ? (for other users)
191 nvoice <- 4
192 # noctave = 2^13 = 8192 half hours ~ 180 days ; ~log2(ncol(synchrones))
193 noctave = 13
194 # 4 here represent 2^5 = 32 half-hours ~ 1 day
195 #NOTE: default scalevector == 2^(0:(noctave * nvoice) / nvoice) * s0 (?)
196 scalevector <- 2^(4:(noctave * nvoice) / nvoice + 1)
197 #condition: ( log2(s0*w0/(2*pi)) - 1 ) * nvoice + 1.5 >= 1
198 s0=2
199 w0=2*pi
200 scaled=FALSE
201 s0log = as.integer( (log2( s0*w0/(2*pi) ) - 1) * nvoice + 1.5 )
202 totnoct = noctave + as.integer(s0log/nvoice) + 1
203
204 computeCWT = function(i)
205 {
206 if (verbose)
207 cat(paste("+++ Compute Rwave::cwt() on serie ",i,"\n", sep=""))
208 ts <- scale(ts(synchrones[i,]), center=TRUE, scale=scaled)
209 totts.cwt = Rwave::cwt(ts, totnoct, nvoice, w0, plot=FALSE)
210 ts.cwt = totts.cwt[,s0log:(s0log+noctave*nvoice)]
211 #Normalization
212 sqs <- sqrt(2^(0:(noctave*nvoice)/nvoice)*s0)
213 sqres <- sweep(ts.cwt,2,sqs,'*')
214 sqres / max(Mod(sqres))
215 }
216
217 if (parll)
218 {
219 cl = parallel::makeCluster(ncores_clust)
220 parallel::clusterExport(cl,
221 varlist=c("synchrones","totnoct","nvoice","w0","s0log","noctave","s0","verbose"),
222 envir=environment())
223 }
224
225 # list of CWT from synchrones
226 # TODO: fit in RAM, OK? If not, 2 options: serialize, compute individual distances
227 Xcwt4 <-
228 if (parll)
229 parallel::parLapply(cl, seq_len(n), computeCWT)
230 else
231 lapply(seq_len(n), computeCWT)
232
233 if (parll)
234 parallel::stopCluster(cl)
235
236 Xwer_dist <- bigmemory::big.matrix(nrow=n, ncol=n, type="double")
237 fcoefs = rep(1/3, 3) #moving average on 3 values (TODO: very slow! correct?!)
238 if (verbose)
239 cat("*** Compute WER distances from CWT\n")
240
241 #TODO: computeDistances(i,j), et répartir les n(n-1)/2 couples d'indices
242 #là c'est trop déséquilibré
243
244 computeDistancesLineI = function(i)
245 {
246 if (verbose)
247 cat(paste(" Line ",i,"\n", sep=""))
248 for (j in (i+1):n)
249 {
250 #TODO: 'circular=TRUE' is wrong, should just take values on the sides; to rewrite in C
251 num <- filter(Mod(Xcwt4[[i]] * Conj(Xcwt4[[j]])), fcoefs, circular=TRUE)
252 WX <- filter(Mod(Xcwt4[[i]] * Conj(Xcwt4[[i]])), fcoefs, circular=TRUE)
253 WY <- filter(Mod(Xcwt4[[j]] * Conj(Xcwt4[[j]])), fcoefs, circular=TRUE)
254 wer2 <- sum(colSums(num)^2) / sum( sum(colSums(WX) * colSums(WY)) )
255 if (parll)
256 synchronicity::lock(m)
257 Xwer_dist[i,j] <- sqrt(delta * ncol(Xcwt4[[1]]) * (1 - wer2))
258 Xwer_dist[j,i] <- Xwer_dist[i,j]
259 if (parll)
260 synchronicity::unlock(m)
261 }
262 Xwer_dist[i,i] = 0.
263 }
264
265 parll = (requireNamespace("synchronicity",quietly=TRUE)
266 && parll && Sys.info()['sysname'] != "Windows")
267 if (parll)
268 m <- synchronicity::boost.mutex()
269
270 ignored <-
271 if (parll)
272 {
273 parallel::mclapply(seq_len(n-1), computeDistancesLineI,
274 mc.cores=ncores_clust, mc.allow.recursive=FALSE)
275 }
276 else
277 lapply(seq_len(n-1), computeDistancesLineI)
278 Xwer_dist[n,n] = 0.
279 Xwer_dist
280 }
281
282 # Helper function to divide indices into balanced sets
283 .spreadIndices = function(indices, nb_per_chunk)
284 {
285 L = length(indices)
286 nb_workers = floor( L / nb_per_chunk )
287 if (nb_workers == 0)
288 {
289 # L < nb_series_per_chunk, simple case
290 indices_workers = list(indices)
291 }
292 else
293 {
294 indices_workers = lapply( seq_len(nb_workers), function(i)
295 indices[(nb_per_chunk*(i-1)+1):(nb_per_chunk*i)] )
296 # Spread the remaining load among the workers
297 rem = L %% nb_per_chunk
298 while (rem > 0)
299 {
300 index = rem%%nb_workers + 1
301 indices_workers[[index]] = c(indices_workers[[index]], indices[L-rem+1])
302 rem = rem - 1
303 }
304 }
305 indices_workers
306 }