before computeSynchrones
authorBenjamin Auder <benjamin.auder@somewhere>
Sat, 4 Mar 2017 23:03:37 +0000 (00:03 +0100)
committerBenjamin Auder <benjamin.auder@somewhere>
Sat, 4 Mar 2017 23:03:37 +0000 (00:03 +0100)
epclust/R/clustering.R
epclust/R/main.R

index 6090517..c8bad66 100644 (file)
@@ -1,57 +1,60 @@
 # Cluster one full task (nb_curves / ntasks series)
-clusteringTask = function(indices, ncores)
+clusteringTask = function(indices,getSeries,getSeriesForSynchrones,synchrones_file,
+       getCoefs,K1,K2,nb_series_per_chunk,ncores,to_file)
 {
        cl = parallel::makeCluster(ncores)
-       parallel::clusterExport(cl,
-               varlist=c("K1","getCoefs"),
-               envir=environment())
        repeat
        {
-               nb_workers = max( 1, round( length(indices_clust) / nb_series_per_chunk ) )
+               nb_workers = max( 1, round( length(indices) / nb_series_per_chunk ) )
                indices_workers = lapply(seq_len(nb_workers), function(i) {
                        upper_bound = ifelse( i<nb_workers,
-                               min(nb_series_per_chunk*i,length(indices_clust)), length(indices_clust) )
-                       indices_clust[(nb_series_per_chunk*(i-1)+1):upper_bound]
+                               min(nb_series_per_chunk*i,length(indices)), length(indices) )
+                       indices[(nb_series_per_chunk*(i-1)+1):upper_bound]
                })
-               indices_clust = unlist( parallel::parLapply(cl, indices_workers, function(indices)
-                       computeClusters1(indices, getCoefs, K1)) )
+               indices = unlist( parallel::parLapply(cl, indices_workers, function(inds)
+                       computeClusters1(inds, getCoefs, K1)) )
                if (length(indices_clust) == K1)
                        break
        }
-       parallel::stopCluster(cl_clust)
-       if (WER == "end")
-               return (indices_clust)
-       #WER=="mix"
-       computeClusters2(indices_clust, K2, getSeries, to_file=TRUE)
+       parallel::stopCluster(cl)
+       if (K2 == 0)
+               return (indices)
+       computeClusters2(indices, K2, getSeries, getSeriesForSynchrones, to_file)
+       vector("integer",0)
 }
 
 # Apply the clustering algorithm (PAM) on a coeffs or distances matrix
 computeClusters1 = function(indices, getCoefs, K1)
-       indices[ cluster::pam(getCoefs(indices), K1, diss=FALSE)$id.med ]
+{
+       coefs = getCoefs(indices)
+       indices[ cluster::pam(coefs, K1, diss=FALSE)$id.med ]
+}
 
 # Cluster a chunk of series inside one task (~max nb_series_per_chunk)
-computeClusters2 = function(indices, K2, getSeries, to_file)
+computeClusters2 = function(indices, K2, getSeries, getSeriesForSynchrones, to_file)
 {
-       if (is.null(indices))
-       {
-               #get series from file
-       }
-#Puis K-means après WER...
-       if (WER=="mix" > 0)
+       curves = computeSynchrones(indices, getSeries, getSeriesForSynchrones)
+       dists = computeWerDists(curves)
+       medoids = cluster::pam(dists, K2, diss=TRUE)$medoids
+       if (to_file)
        {
-               curves = computeSynchrones(indices)
-               dists = computeWerDists(curves)
-               indices = computeClusters(dists, K2, diss=TRUE)
+               serialize(medoids, synchrones_file)
+               return (NULL)
        }
-       if (to_file)
-               #write results to file (JUST series ; no possible ID here)
+       medoids
 }
 
 # Compute the synchrones curves (sum of clusters elements) from a clustering result
-computeSynchrones = function(inds)
-       sapply(seq_along(inds), colMeans(getSeries(inds[[i]]$indices,inds[[i]]$ids)))
+computeSynchrones = function(indices, getSeries, getSeriesForSynchrones)
+{
+       #les getSeries(indices) sont les medoides --> init vect nul pour chacun, puis incr avec les
+       #courbes (getSeriesForSynchrones) les plus proches... --> au sens de la norme L2 ?
+       series = getSeries(indices)
+       #...........
+       #sapply(seq_along(inds), colMeans(getSeries(inds[[i]]$indices,inds[[i]]$ids)))
+}
 
-# Compute the WER distance between the synchrones curves (in columns)
+# Compute the WER distance between the synchrones curves (in rows)
 computeWerDist = function(curves)
 {
        if (!require("Rwave", quietly=TRUE))
@@ -74,7 +77,7 @@ computeWerDist = function(curves)
 
        # (normalized) observations node with CWT
        Xcwt4 <- lapply(seq_len(n), function(i) {
-               ts <- scale(ts(curves[,i]), center=TRUE, scale=scaled)
+               ts <- scale(ts(curves[i,]), center=TRUE, scale=scaled)
                totts.cwt = Rwave::cwt(ts,totnoct,nvoice,w0,plot=0)
                ts.cwt = totts.cwt[,s0log:(s0log+noctave*nvoice)]
                #Normalization
index ac4ea8d..0b59832 100644 (file)
@@ -34,6 +34,7 @@
 #'     "LIMIT ", n, " ORDER BY date", sep=""))
 #'   return (df)
 #' }
+#' #####TODO: if DB, array rank --> ID at first retrieval, when computing coeffs; so:: NO use of IDs !
 #'   #TODO: 3 examples, data.frame / binary file / DB sqLite
 #'   + sampleCurves : wavBootstrap de package wmtsa
 #' cl = epclust(getData, K1=200, K2=15, ntasks=1000, nb_series_per_chunk=5000, WER="mix")
@@ -98,7 +99,6 @@ epclust = function(series,K1,K2,ntasks=1,nb_series_per_chunk=50*K1,min_series_pe
                nb_curves = nb_curves + nrow(coeffs_chunk)
        }
        getCoefs = function(indices) getDataInFile(indices, coefs_file)
-######TODO: if DB, array rank --> ID at first retrieval, when computing coeffs; so:: NO use of IDs !
 
        if (nb_curves < min_series_per_chunk)
                stop("Not enough data: less rows than min_series_per_chunk!")
@@ -112,17 +112,36 @@ epclust = function(series,K1,K2,ntasks=1,nb_series_per_chunk=50*K1,min_series_pe
                upper_bound = ifelse( i<ntasks, min(nb_series_per_task*i,nb_curves), nb_curves )
                indices[((i-1)*nb_series_per_task+1):upper_bound]
        })
-       cl_tasks = parallel::makeCluster(ncores_tasks)
-       parallel::clusterExport(cl_tasks,
-               varlist=c("getSeries","getCoefs","K1","K2","WER","nb_series_per_chunk","ncores_clust"),
-               envir=environment())
+       cl = parallel::makeCluster(ncores_tasks)
        #1000*K1 (or K2) indices (or NOTHING--> series on file)
-       indices = parallel::parLapply(cl_tasks, indices_tasks, clusteringTask)
-       parallel::stopCluster(cl_tasks)
+       indices = unlist( parallel::parLapply(cl, indices_tasks, function(inds) {
+               clusteringTask(inds, getSeries, getSeries, getCoefs, K1, K2*(WER=="mix"),
+                       nb_series_per_chunk,ncores_clust,to_file=TRUE)
+       }) )
+       parallel::stopCluster(cl)
 
-       #Now series must be retrieved from synchrones_file, and have no ID
-       getSeries = function(indices, ids) getDataInFile(indices, synchrones_file)
+       getSeriesForSynchrones = getSeries
+       synchrones_file = paste(bin_dir,"synchrones",sep="")
+       if (WER=="mix")
+       {
+               indices = seq_len(ntasks*K2)
+               #Now series must be retrieved from synchrones_file
+               getSeries = function(inds) getDataInFile(inds, synchrones_file)
+               #Coefs must be re-computed
+               unlink(coefs_file)
+               index = 1
+               repeat
+               {
+                       series = getSeries((index-1)+seq_len(nb_series_per_chunk))
+                       if (is.null(series))
+                               break
+                       coeffs_chunk = curvesToCoeffs(series, wf)
+                       serialize(coeffs_chunk, coefs_file)
+                       index = index + nb_series_per_chunk
+               }
+       }
 
        # Run step2 on resulting indices or series (from file)
-       computeClusters2(indices=if (WER=="end") indices else NULL, K2, to_file=FALSE)
+       clusteringTask(indices, getSeries, getSeriesForSynchrones, getCoefs, K1, K2,
+               nb_series_per_chunk, ncores_tasks*ncores_clust, to_file=FALSE)
 }