# Cluster one full task (nb_curves / ntasks series)
-clusteringTask = function(indices_clust)
+clusteringTask = function(indices, ncores)
{
- cl_clust = parallel::makeCluster(ncores_clust)
- parallel::clusterExport(cl_clust,
- varlist=c("K1","K2","WER"),
+ cl = parallel::makeCluster(ncores)
+ parallel::clusterExport(cl,
+ varlist=c("K1","getCoefs"),
envir=environment())
repeat
{
min(nb_series_per_chunk*i,length(indices_clust)), length(indices_clust) )
indices_clust[(nb_series_per_chunk*(i-1)+1):upper_bound]
})
- indices_clust = parallel::parLapply(cl, indices_workers, clusterChunk)
- # TODO: soft condition between K2 and K1, before applying final WER step
- if ((WER=="end" && length(indices_clust)==K1) || (WER=="mix" && length(indices_clust)==K2))
+ indices_clust = unlist( parallel::parLapply(cl, indices_workers, function(indices)
+ computeClusters1(indices, getCoefs, K1)) )
+ if (length(indices_clust) == K1)
break
}
parallel::stopCluster(cl_clust)
- unlist(indices_clust)
+ if (WER == "end")
+ return (indices_clust)
+ #WER=="mix"
+ computeClusters2(indices_clust, K2, getSeries, to_file=TRUE)
}
+# Apply the clustering algorithm (PAM) on a coeffs or distances matrix
+computeClusters1 = function(indices, getCoefs, K1)
+ indices[ cluster::pam(getCoefs(indices), K1, diss=FALSE)$id.med ]
+
# Cluster a chunk of series inside one task (~max nb_series_per_chunk)
-clusterChunk = function(indices_chunk)
+computeClusters2 = function(indices, K2, getSeries, to_file)
{
- coeffs = readCoeffs(indices_chunk)
- cl = computeClusters(as.matrix(coeffs[,2:ncol(coeffs)]), K1, diss=FALSE)
+ if (is.null(indices))
+ {
+ #get series from file
+ }
+#Puis K-means après WER...
if (WER=="mix" > 0)
{
- curves = computeSynchrones(cl)
+ curves = computeSynchrones(indices)
dists = computeWerDists(curves)
- cl = computeClusters(dists, K2, diss=TRUE)
+ indices = computeClusters(dists, K2, diss=TRUE)
}
- indices_chunk[cl]
-}
-
-# Apply the clustering algorithm (PAM) on a coeffs or distances matrix
-computeClusters = function(md, K, diss)
-{
- if (!require(cluster, quietly=TRUE))
- stop("Unable to load cluster library")
- cluster::pam(md, K, diss=diss)$id.med
+ if (to_file)
+ #write results to file (JUST series ; no possible ID here)
}
# Compute the synchrones curves (sum of clusters elements) from a clustering result
-computeSynchrones = function(indices)
-{
- colSums( getData(indices) )
-}
+computeSynchrones = function(inds)
+ sapply(seq_along(inds), colMeans(getSeries(inds[[i]]$indices,inds[[i]]$ids)))
-# Compute the WER distance between the synchrones curves
+# Compute the WER distance between the synchrones curves (in columns)
computeWerDist = function(curves)
{
if (!require("Rwave", quietly=TRUE))
# (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
{
for (j in (i+1):n)
{
- #TODO: later, compute CWT here (because not enough storage space for 32M series)
+ #TODO: later, compute CWT here (because not enough storage space for 200k series)
# 'circular=TRUE' is wrong, should just take values on the sides; to rewrite in C
num <- filter(Mod(Xcwt4[[i]] * Conj(Xcwt4[[j]])), fcoefs, circular=TRUE)
WX <- filter(Mod(Xcwt4[[i]] * Conj(Xcwt4[[i]])), fcoefs, circular=TRUE)
+++ /dev/null
-computeCoeffs = function(data, index, nb_series_per_chunk, wf)
-{
- coeffs_chunk = NULL
- if (is.data.frame(data) && index < nrow(data))
- {
- #full data matrix
- coeffs_chunk = curvesToCoeffs(
- data[index:(min(index+nb_series_per_chunk-1,nrow(data))),], wf)
- }
- else if (is.function(data))
- {
- #custom user function to retrieve next n curves, probably to read from DB
- coeffs_chunk = curvesToCoeffs( data(rank=(index-1)+seq_len(nb_series_per_chunk)), wf )
- }
- else if (exists(data_con))
- {
- #incremental connection ; TODO: more efficient way to parse than using a temp file
- ascii_lines = readLines(data_con, nb_series_per_chunk)
- if (length(ascii_lines > 0))
- {
- series_chunk_file = ".series_chunk"
- writeLines(ascii_lines, series_chunk_file)
- coeffs_chunk = curvesToCoeffs( read.csv(series_chunk_file), wf )
- unlink(series_chunk_file)
- }
- }
- coeffs_chunk
-}
-
-curvesToCoeffs = function(series, wf)
-{
- if (!require(wavelets, quietly=TRUE))
- stop("Couldn't load wavelets library")
- L = length(series[1,])
- D = ceiling( log2(L) )
- nb_sample_points = 2^D
- #TODO: parallel::parApply() ?!
- as.data.frame( apply(series, 1, function(x) {
- interpolated_curve = spline(1:L, x, n=nb_sample_points)$y
- W = wavelets::dwt(interpolated_curve, filter=wf, D)@W
- rev( sapply( W, function(v) ( sqrt( sum(v^2) ) ) ) )
- }) )
-}
#' @param ncores_tasks "MPI" number of parallel tasks (1 to disable: sequential tasks)
#' @param ncores_clust "OpenMP" number of parallel clusterings in one task
#' @param random Randomize chunks repartition
+#' @param ... Other arguments to be passed to \code{data} function
#'
#' @return A data.frame of the final medoids curves (identifiers + values)
#'
#' + sampleCurves : wavBootstrap de package wmtsa
#' cl = epclust(getData, K1=200, K2=15, ntasks=1000, nb_series_per_chunk=5000, WER="mix")
#' @export
-epclust = function(data, K1, K2, ntasks=1, nb_series_per_chunk=50*K1, min_series_per_chunk=5*K1,
- wf="haar", WER="end", ncores_tasks=1, ncores_clust=4, random=TRUE)
+epclust = function(series,K1,K2,ntasks=1,nb_series_per_chunk=50*K1,min_series_per_chunk=5*K1,
+ wf="haar",WER="end",ncores_tasks=1,ncores_clust=4,random=TRUE,...)
{
- # Check arguments
- if (!is.data.frame(data) && !is.function(data))
+ # Check/transform arguments
+ bin_dir = "epclust.bin/"
+ dir.create(bin_dir, showWarnings=FALSE, mode="0755")
+ if (!is.function(series))
+ {
+ series_file = paste(bin_dir,"data",sep="")
+ unlink(series_file)
+ }
+ if (is.matrix(series))
+ serialize(series, series_file)
+ else if (!is.function(series))
{
tryCatch(
{
- if (is.character(data))
- data_con = file(data, open="r")
- else if (!isOpen(data))
+ if (is.character(series))
+ series_con = file(series, open="r")
+ else if (!isOpen(series))
{
- open(data)
- data_con = data
+ open(series)
+ series_con = series
}
+ serialize(series_con, series_file)
+ close(series_con)
},
- error=function(e) "data should be a data.frame, a function or a valid connection"
+ error=function(e) "series should be a data.frame, a function or a valid connection"
)
}
+ if (!is.function(series))
+ series = function(indices) getDataInFile(indices, series_file)
+ getSeries = series
+
K1 = toInteger(K1, function(x) x>=2)
K2 = toInteger(K2, function(x) x>=2)
ntasks = toInteger(ntasks, function(x) x>=1)
stop("WER takes values in {'end','mix'}")
# Serialize all wavelets coefficients (+ IDs) onto a file
- unlink(".coeffs")
+ coefs_file = paste(bin_dir,"coefs",sep="")
+ unlink(coefs_file)
index = 1
nb_curves = 0
- nb_coeffs = NA
repeat
{
- coeffs_chunk = computeCoeffs(data, index, nb_series_per_chunk, wf)
- if (is.null(coeffs_chunk))
+ series = getSeries((index-1)+seq_len(nb_series_per_chunk))
+ if (is.null(series))
break
- writeCoeffs(coeffs_chunk)
+ coeffs_chunk = curvesToCoeffs(series, wf)
+ serialize(coeffs_chunk, coefs_file)
index = index + nb_series_per_chunk
nb_curves = nb_curves + nrow(coeffs_chunk)
- if (is.na(nb_coeffs))
- nb_coeffs = ncol(coeffs_chunk)-1
}
+ 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!")
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]
})
- library(parallel, quietly=TRUE)
cl_tasks = parallel::makeCluster(ncores_tasks)
parallel::clusterExport(cl_tasks,
- varlist=c("K1","K2","WER","nb_series_per_chunk","ncores_clust"),#TODO: pass also
- #nb_coeffs...and filename (in a list... ?)
+ varlist=c("getSeries","getCoefs","K1","K2","WER","nb_series_per_chunk","ncores_clust"),
envir=environment())
+ #1000*K1 (or K2) indices (or NOTHING--> series on file)
indices = parallel::parLapply(cl_tasks, indices_tasks, clusteringTask)
parallel::stopCluster(cl_tasks)
- # Run step1+2 step on resulting ranks
- indices = clusterChunk(indices, K1, K2)
- return (list("indices"=indices, "medoids"=getSeries(data, indices)))
+ #Now series must be retrieved from synchrones_file, and have no ID
+ getSeries = function(indices, ids) getDataInFile(indices, synchrones_file)
+
+ # Run step2 on resulting indices or series (from file)
+ computeClusters2(indices=if (WER=="end") indices else NULL, K2, to_file=FALSE)
}
{
#TODO:
}
+
+curvesToCoeffs = function(series, wf)
+{
+ L = length(series[1,])
+ D = ceiling( log2(L) )
+ nb_sample_points = 2^D
+ apply(series, 1, function(x) {
+ interpolated_curve = spline(1:L, x, n=nb_sample_points)$y
+ W = wavelets::dwt(interpolated_curve, filter=wf, D)@W
+ rev( sapply( W, function(v) ( sqrt( sum(v^2) ) ) ) )
+ })
+}
projects / epclust.git / commitdiff