progress in main.R
authorBenjamin Auder <benjamin.auder@somewhere>
Mon, 9 Jan 2017 11:06:39 +0000 (12:06 +0100)
committerBenjamin Auder <benjamin.auder@somewhere>
Mon, 9 Jan 2017 11:06:39 +0000 (12:06 +0100)
code/draft_R_pkg/DESCRIPTION
code/draft_R_pkg/R/main.R

index ce9129b..669e8c0 100644 (file)
@@ -10,12 +10,11 @@ Author:
        Jairo Cugliari <Jairo.Cugliari@univ-lyon2.fr> [aut]
 Maintainer: Benjamin Auder <Benjamin.Auder@math.u-psud.fr>
 Depends:
-    R (>= 3.0.0)
-Imports:
-               MASS
+    R (>= 3.0.0),
+               parallel,
+               cluster
 Suggests:
     testthat,
-    parallel,
     knitr
 License: MIT + file LICENSE
 VignetteBuilder: knitr
index 695b928..19729ed 100644 (file)
@@ -30,7 +30,17 @@ epclust = function(data, K, nbSeriesPerChunk, writeTmp=ref_writeTmp, readTmp=ref
 
        #0) check arguments
        if (!is.data.frame(data) && !is.function(data))
-               tryCatch({dataCon = open(data)},
+               tryCatch(
+                       {
+                               if (is.character(data))
+                               {
+                                       dataCon = file(data, open="r")
+                               } else if (!isOpen(data))
+                               {
+                                       open(data)
+                                       dataCon = data
+                               }
+                       },
                        error="data should be a data.frame, a function or a valid connection")
        if (!is.integer(K) || K < 2)
                stop("K should be an integer greater or equal to 2")
@@ -44,17 +54,17 @@ epclust = function(data, K, nbSeriesPerChunk, writeTmp=ref_writeTmp, readTmp=ref
 
        #1) acquire data (process curves, get as coeffs)
        index = 1
-       nbCurves = nrow(data)
-       while (index < nbCurves)
+       nbCurves = 0
+       repeat
        {
                if (is.data.frame(data))
                {
                        #full data matrix
-                       writeTmp( getCoeffs( data[index:(min(index+nbSeriesPerChunk-1,nbCurves)),] ) )
+                       error = writeTmp( getCoeffs( data[index:(min(index+nbSeriesPerChunk-1,nrow(data))),] ) )
                } else if (is.function(data))
                {
                        #custom user function to retrieve next n curves, probably to read from DB
-                       writeTmp( getCoeffs( data(index, nbSeriesPerChunk) ) )
+                       error = writeTmp( getCoeffs( data(index, nbSeriesPerChunk) ) )
                } else
                {
                        #incremental connection
@@ -62,7 +72,7 @@ epclust = function(data, K, nbSeriesPerChunk, writeTmp=ref_writeTmp, readTmp=ref
                        ascii_lines = readLines(dataCon, nbSeriesPerChunk)
                        seriesChunkFile = ".tmp/seriesChunk"
                        writeLines(ascii_lines, seriesChunkFile)
-                       writeTmp( getCoeffs( read.csv(seriesChunkFile) ) )
+                       error = writeTmp( getCoeffs( read.csv(seriesChunkFile) ) )
                }
                index = index + nbSeriesPerChunk
        }
@@ -73,9 +83,11 @@ epclust = function(data, K, nbSeriesPerChunk, writeTmp=ref_writeTmp, readTmp=ref
        ncores = ifelse(is.integer(ncores), ncores, parallel::detectCores())
        cl = parallel::makeCluster(ncores)
        parallel::clusterExport(cl=cl, varlist=c("X", "Y", "K", "p"), envir=environment())
+       library(cluster)
        li = parallel::parLapply(cl, 1:B, getParamsAtIndex)
 
-       #2) process coeffs (by nbSeriesPerChunk) and cluster in parallel (just launch async task, wait for them to complete, and re-do if necessary)
+       #2) process coeffs (by nbSeriesPerChunk) and cluster them in parallel
+       #TODO: be careful of writing to a new temp file, then flush initial one, then re-use it...
        repeat
        {
                completed = rep(FALSE, ............)
@@ -85,7 +97,7 @@ epclust = function(data, K, nbSeriesPerChunk, writeTmp=ref_writeTmp, readTmp=ref
                #C) flush tmp file (current parallel processes will write in it)
                #always check "complete" flag (array, as I did in MPI) to know if "slaves" finished
        }
-
+pam(x, k
        parallel::stopCluster(cl)
 
        #3) readTmp last results, apply PAM on it, and return medoids + identifiers