epclust/R/computeWerDists.R

   1 #' computeWerDists
   2 #'
   3 #' Compute the WER distances between the synchrones curves (in columns), which are
   4 #' returned (e.g.) by \code{computeSynchrones()}
   5 #'
   6 #' @param indices Range of series indices to cluster
   7 #' @inheritParams claws
   8 #' @inheritParams computeSynchrones
   9 #'
  10 #' @return A distances matrix of size K x K where K == length(indices)
  11 #'
  12 #' @export
  13 computeWerDists <- function(indices, getSeries, nb_series_per_chunk, smooth_lvl, nvoice,
  14     nbytes, endian, ncores_clust=1, verbose=FALSE, parll=TRUE)
  15 {
  16     n <- length(indices)
  17     L <- length(getSeries(1)) #TODO: not very neat way to get L
  18     noctave <- ceiling(log2(L)) #min power of 2 to cover serie range
  19     # Since a CWT contains noctave*nvoice complex series, we deduce the number of CWT to
  20     # retrieve/put in one chunk.
  21     nb_cwt_per_chunk <- max(1, floor(nb_series_per_chunk / (nvoice*noctave*2)))
  22
  23     # Initialize result as a square big.matrix of size 'number of medoids'
  24     Xwer_dist <- bigmemory::big.matrix(nrow=n, ncol=n, type="double")
  25
  26     cwt_file <- tempfile(pattern="epclust_cwt.bin_")
  27     # Compute the getSeries(indices) CWT, and store the results in the binary file
  28     computeSaveCWT <- function(indices)
  29     {
  30         if (parll)
  31         {
  32             require("bigmemory", quietly=TRUE)
  33             require("Rwave", quietly=TRUE)
  34             require("epclust", quietly=TRUE)
  35         }
  36
  37         # Obtain CWT as big vectors of real part + imaginary part (concatenate)
  38         ts_cwt <- sapply(indices, function(i) {
  39             ts <- scale(ts(getSeries(i)), center=TRUE, scale=FALSE)
  40             ts_cwt <- Rwave::cwt(ts, noctave, nvoice, w0=2*pi, twoD=TRUE, plot=FALSE)
  41             c( as.double(Re(ts_cwt)),as.double(Im(ts_cwt)) )
  42         })
  43
  44         # Serialization
  45         binarize(ts_cwt, cwt_file, nb_cwt_per_chunk, ",", nbytes, endian)
  46     }
  47
  48     # Function to retrieve a synchrone CWT from (binary) file
  49     getCWT <- function(index, L)
  50     {
  51         flat_cwt <- getDataInFile(index, cwt_file, nbytes, endian)
  52         cwt_length <- length(flat_cwt) / 2
  53         re_part <- as.matrix(flat_cwt[1:cwt_length], nrow=L)
  54         im_part <- as.matrix(flat_cwt[(cwt_length+1):(2*cwt_length)], nrow=L)
  55         re_part + 1i * im_part
  56     }
  57
  58     # Compute distance between columns i and j for j>i
  59     computeDistances <- function(i)
  60     {
  61         if (parll)
  62         {
  63             # parallel workers start with an empty environment
  64             require("bigmemory", quietly=TRUE)
  65             require("epclust", quietly=TRUE)
  66             Xwer_dist <- bigmemory::attach.big.matrix(Xwer_dist_desc)
  67         }
  68
  69         if (verbose && !parll)
  70             cat(paste("   Distances from ",i," to ",i+1,"...",n,"\n", sep=""))
  71
  72         # Get CWT of column i, and run computations for columns j>i
  73         cwt_i <- getCWT(i, L)
  74         WX  <- filterMA(Mod(cwt_i * Conj(cwt_i)), smooth_lvl)
  75
  76         for (j in (i+1):n)
  77         {
  78             cwt_j <- getCWT(j, L)
  79
  80             # Compute the ratio of integrals formula 5.6 for WER^2
  81             # in https://arxiv.org/abs/1101.4744v2 §5.3
  82             num <- filterMA(Mod(cwt_i * Conj(cwt_j)), smooth_lvl)
  83             WY <- filterMA(Mod(cwt_j * Conj(cwt_j)), smooth_lvl)
  84             wer2 <- sum(colSums(num)^2) / sum(colSums(WX) * colSums(WY))
  85
  86             Xwer_dist[i,j] <- sqrt(L * ncol(cwt_i) * (1 - wer2))
  87             Xwer_dist[j,i] <- Xwer_dist[i,j]
  88         }
  89         Xwer_dist[i,i] <- 0.
  90     }
  91
  92     if (parll)
  93     {
  94         # outfile=="" to see stderr/stdout on terminal
  95         cl <- parallel::makeCluster(ncores_clust, outfile="")
  96         Xwer_dist_desc <- bigmemory::describe(Xwer_dist)
  97         parallel::clusterExport(cl, varlist=c("parll","nb_cwt_per_chunk","n","L",
  98             "Xwer_dist_desc","noctave","nvoice","getCWT"), envir=environment())
  99     }
 100
 101     if (verbose)
 102         cat(paste("--- Precompute and serialize synchrones CWT\n", sep=""))
 103
 104     # Split indices by packets of length at most nb_cwt_per_chunk
 105     indices_cwt <- .splitIndices(seq_len(n), nb_cwt_per_chunk)
 106     ignored <-
 107         if (parll)
 108             parallel::parLapply(cl, indices_cwt, computeSaveCWT)
 109         else
 110             lapply(indices_cwt, computeSaveCWT)
 111
 112     if (verbose)
 113         cat(paste("--- Compute WER distances\n", sep=""))
 114
 115     ignored <-
 116         if (parll)
 117             parallel::parLapply(cl, seq_len(n-1), computeDistances)
 118         else
 119             lapply(seq_len(n-1), computeDistances)
 120     Xwer_dist[n,n] <- 0.
 121
 122     if (parll)
 123         parallel::stopCluster(cl)
 124
 125     unlink(cwt_file) #remove binary file
 126
 127     Xwer_dist[,] #~small matrix K1 x K1
 128 }