1 # Cluster one full task (nb_curves / ntasks series)
2 clusteringTask = function(indices,getSeries,getSeriesForSynchrones,synchrones_file,
3 getCoefs,K1,K2,nb_series_per_chunk,ncores,to_file)
5 cl = parallel::makeCluster(ncores)
8 nb_workers = max( 1, round( length(indices) / nb_series_per_chunk ) )
9 indices_workers = lapply(seq_len(nb_workers), function(i) {
10 upper_bound = ifelse( i<nb_workers,
11 min(nb_series_per_chunk*i,length(indices)), length(indices) )
12 indices[(nb_series_per_chunk*(i-1)+1):upper_bound]
14 indices = unlist( parallel::parLapply(cl, indices_workers, function(inds)
15 computeClusters1(inds, getCoefs, K1)) )
16 if (length(indices_clust) == K1)
19 parallel::stopCluster(cl)
22 computeClusters2(indices, K2, getSeries, getSeriesForSynchrones, to_file)
26 # Apply the clustering algorithm (PAM) on a coeffs or distances matrix
27 computeClusters1 = function(indices, getCoefs, K1)
29 coefs = getCoefs(indices)
30 indices[ cluster::pam(coefs, K1, diss=FALSE)$id.med ]
33 # Cluster a chunk of series inside one task (~max nb_series_per_chunk)
34 computeClusters2 = function(indices, K2, getSeries, getSeriesForSynchrones, to_file)
36 curves = computeSynchrones(indices, getSeries, getSeriesForSynchrones)
37 dists = computeWerDists(curves)
38 medoids = cluster::pam(dists, K2, diss=TRUE)$medoids
41 serialize(medoids, synchrones_file)
47 # Compute the synchrones curves (sum of clusters elements) from a clustering result
48 computeSynchrones = function(indices, getSeries, getSeriesForSynchrones)
50 #les getSeries(indices) sont les medoides --> init vect nul pour chacun, puis incr avec les
51 #courbes (getSeriesForSynchrones) les plus proches... --> au sens de la norme L2 ?
52 series = getSeries(indices)
54 #sapply(seq_along(inds), colMeans(getSeries(inds[[i]]$indices,inds[[i]]$ids)))
57 # Compute the WER distance between the synchrones curves (in rows)
58 computeWerDist = function(curves)
60 if (!require("Rwave", quietly=TRUE))
61 stop("Unable to load Rwave library")
64 #TODO: automatic tune of all these parameters ? (for other users)
66 # noctave = 2^13 = 8192 half hours ~ 180 days ; ~log2(ncol(curves))
68 # 4 here represent 2^5 = 32 half-hours ~ 1 day
69 #NOTE: default scalevector == 2^(0:(noctave * nvoice) / nvoice) * s0 (?)
70 scalevector <- 2^(4:(noctave * nvoice) / nvoice) * 2
71 #condition: ( log2(s0*w0/(2*pi)) - 1 ) * nvoice + 1.5 >= 1
75 s0log = as.integer( (log2( s0*w0/(2*pi) ) - 1) * nvoice + 1.5 )
76 totnoct = noctave + as.integer(s0log/nvoice) + 1
78 # (normalized) observations node with CWT
79 Xcwt4 <- lapply(seq_len(n), function(i) {
80 ts <- scale(ts(curves[i,]), center=TRUE, scale=scaled)
81 totts.cwt = Rwave::cwt(ts,totnoct,nvoice,w0,plot=0)
82 ts.cwt = totts.cwt[,s0log:(s0log+noctave*nvoice)]
84 sqs <- sqrt(2^(0:(noctave*nvoice)/nvoice)*s0)
85 sqres <- sweep(ts.cwt,MARGIN=2,sqs,'*')
86 sqres / max(Mod(sqres))
89 Xwer_dist <- matrix(0., n, n)
90 fcoefs = rep(1/3, 3) #moving average on 3 values (TODO: very slow! correct?!)
95 #TODO: later, compute CWT here (because not enough storage space for 200k series)
96 # 'circular=TRUE' is wrong, should just take values on the sides; to rewrite in C
97 num <- filter(Mod(Xcwt4[[i]] * Conj(Xcwt4[[j]])), fcoefs, circular=TRUE)
98 WX <- filter(Mod(Xcwt4[[i]] * Conj(Xcwt4[[i]])), fcoefs, circular=TRUE)
99 WY <- filter(Mod(Xcwt4[[j]] * Conj(Xcwt4[[j]])), fcoefs, circular=TRUE)
100 wer2 <- sum(colSums(num)^2) / sum( sum(colSums(WX) * colSums(WY)) )
101 Xwer_dist[i,j] <- sqrt(delta * ncol(Xcwt4[[1]]) * (1 - wer2))
102 Xwer_dist[j,i] <- Xwer_dist[i,j]
105 diag(Xwer_dist) <- numeric(n)