-epclust = function(data=NULL, con=NULL, raw=FALSE, K, nbPerChunk, ...)
-{
+#TODO: setRefClass... to avoid copy data !!
+#http://stackoverflow.com/questions/2603184/r-pass-by-reference
+#fields: data (can be NULL or provided by user), coeffs (will be computed
+#con can be a character string naming a file; see readLines()
+#data can be in DB format, on one column : TODO: guess (from header, or col. length...)
-#TODO: just a wrapper which calls ppam.exe (system("...")) and reads output (binary) file to retrieve medoids + IDs
- #on input: can be data or con; data handled by writing it to file (ascii or bin ?!),
- #con handled
+writeTmp(curves [uncompressed coeffs, limited number - nbSeriesPerChunk], last=FALSE) #if last=TRUE, close the conn
+readTmp(..., from index, n curves) #careful: connection must remain open
+#TODO: write read/write tmp reference ( on file in .tmp/ folder ... )
- #options for tmp files: in RAM, on disk, on DB (can be distributed)
+epclust = function(data=NULL, K, nbPerChunk, ..., writeTmp=ref_writeTmp, readTmp=ref_readTmp) #where to put/retrieve intermediate results; if not provided, use file on disk
+{
+ #on input: can be data or con; data handled by writing it to file (ascii or bin ?!),
+#data: con or matrix or DB
- if (!is.null(data))
+ #1) acquire data (process curves, get as coeffs)
+ if (is.numeric(data))
{
#full data matrix
-
- } else if (!is.null(con))
+ index = 1
+ n = nrow(data)
+ while (index < n)
+ {
+ writeTmp( getCoeffs(data) )
+ index = index + nbSeriesPerChunk
+ }
+ } else if (is.function(data))
+ {
+ #custom user function to retrieve next n curves, probably to read from DB
+ writeTmp( getCoeffs( data(nbPerChunk) ) )
+ } else
{
#incremental connection
#read it one by one and get coeffs until nbSeriesPerChunk
#then launch a clustering task............
+ ascii_lines = readLines(data, nbSeriesPerChunk)
+ seriesChunkFile = ".tmp/seriesChunk" #TODO: find a better way
+ writeLines(ascii_lines, seriesChunkFile)
+ writeTmp( getCoeffs( read.csv(seriesChunkFile) ) )
} else
- stop("at least 'data' or 'con' argument must be present")
+ stop("Unrecognizable 'data' argument (must be numeric, functional or connection)")
+ #2) process coeffs (by nbSeriesPerChunk) and cluster in parallel (just launch async task, wait for them to complete, and re-do if necessary)
+
+
+ #3) apply stage 2 (in parallel ? inside task 2) ?)
+}
+
+getCoeffs = function(series)
+{
+ #... return wavelets coeffs : compute in parallel !
}