epclust/R/stage2.R

   1 library("Rwave")
   2
   3 #precondition: ( log2(s0*w0/(2*pi)) - 1 ) * nvoice + 1.5 >= 1
   4 toCWT  <- function(X, tw=0, swabs=0, nvoice=12, noctave=5, s0=2, w0=2*pi,
   5     spectra=FALSE, smooth=TRUE, scaled=FALSE, scalevector)
   6 {
   7     if(missing(scalevector))
   8         scalevector  <- 2^(0:(noctave * nvoice) / nvoice) * s0
   9     s0log=as.integer((log2( s0*w0/(2*pi) )-1)*nvoice+1.5)
  10     totnoct=noctave+as.integer(s0log/nvoice)+1
  11     res <- lapply(1:nrow(X), function(n) {
  12         ts <- scale(ts( X[n,] ), center=TRUE, scale=scaled)
  13         totts.cwt = Rwave::cwt(ts,totnoct,nvoice,w0,plot=0)
  14         ts.cwt=totts.cwt[,s0log:(s0log+noctave*nvoice)]
  15         #Normalization
  16         sqs <- sqrt(2^(0:(noctave*nvoice)/nvoice)*s0)
  17         smat <- matrix(rep(sqs,length(t)),nrow=length(t),byrow=TRUE)
  18         ts.cwt*smat
  19     })
  20     if( spectra )
  21         res <- lapply(res, function(l) Mod(l)^2 )
  22     if( smooth  )
  23         res <- lapply(res, smCWT, swabs = swabs, tw = tw, scalevector = scalevector)
  24     resArray <- array(NA, c(nrow(res[[1]]), ncol(res[[1]]), length(res)))
  25     for( l in 1:length(res) )
  26         resArray[ , , l] <- res[[l]]
  27     resArray
  28 }
  29
  30 #smooth cwt result
  31 smCWT <- function(CWT, tw=  0, swabs= 0, nvoice= 12, noctave= 2, s0= 2, w0= 2*pi,
  32     lt= 24, scalevector )
  33 {
  34     wsp     <- Mod(CWT)
  35     smwsp   <- smooth.matrix(wsp, swabs)
  36     smsmwsp <- smooth.time(smwsp, tw, scalevector)
  37     smsmwsp
  38 }
  39
  40 #dans sowas.R (...donc on ne lisse pas à ce niveau ?)
  41 smooth.matrix <- function(wt,swabs)
  42 {
  43     if (swabs != 0)
  44     {
  45         smwt <- t(filter(t(wt),rep(1,2*swabs+1)/(2*swabs+1)))
  46     } else
  47     {
  48         smwt <- wt
  49     }
  50     smwt
  51 }
  52
  53 smooth.time <- function(wt,tw,scalevector)
  54 {
  55     smwt <- wt
  56     if (tw != 0)
  57     {
  58         for (i in 1:length(scalevector))
  59         {
  60             twi <- as.integer(scalevector[i]*tw)
  61             smwt[,i] <- filter(wt[,i],rep(1,2*twi+1)/(2*twi+1))
  62         }
  63     }
  64     smwt
  65 }
  66
  67 #Entrée : courbes synchrones, soit après étape 1 itérée, soit après chaqure étape 1
  68 step2 = function(conso)
  69 {
  70     n     <- nrow(conso)
  71     m <- ncol(conso)
  72
  73     #TODO: automatic tune of these parameters ? (for other users)
  74     nvoice   <- 4
  75     # noctave = 2^13 = 8192 half hours ~ 180 days ; ~log2(ncol(conso))
  76     noctave = 13
  77     # 4 here represent 2^5 = 32 half-hours ~ 1 day
  78     scalevector4  <- 2^(4:(noctave * nvoice) / nvoice) * 2
  79     lscvect4      <- length(scalevector4)
  80     lscvect <- lscvect4  # i should clean my code: werFam demands a lscvect
  81
  82     # observations node with CWT
  83     Xcwt4   <- toCWT(conso, noctave = noctave, scalevector = scalevector4,
  84         smooth = FALSE, nvoice = nvoice)
  85
  86     #matrix:
  87     Xcwt2 <- matrix(NA_complex_, nrow= n, ncol= 2 + length((c(Xcwt4[,,1]))))
  88
  89     for(i in 1:n)
  90         Xcwt2[i,] <- c(m, lscvect, Xcwt4[,,i] / max(Mod(Xcwt4[,,i])) )
  91
  92     rm(conso, Xcwt4) ; gc()
  93
  94     lscvect = dim(Xcwt4)[2]
  95
  96     Xwer_dist    <- matrix(0.0, n, n)
  97     for(i in 1:(n - 1))
  98     {
  99         mat1   <- matrix(as.vector(Xcwt2[i,])[-(1:2)], m, lscvect)
 100
 101         for(j in (i + 1):n)
 102         {
 103             mat2 <- matrix(as.vector(Xcwt2[j,])[-(1:2)], m, lscvect)
 104             num     <- Mod(mat1 * Conj(mat2))
 105             WX      <- Mod(mat1 * Conj(mat1))
 106             WY      <- Mod(mat2 * Conj(mat2))
 107             smsmnum <- smCWT(num, scalevector = scalevector4)
 108             smsmWX  <- smCWT(WX,  scalevector = scalevector4)
 109             smsmWY  <- smCWT(WY,  scalevector = scalevector4)
 110             wer2    <- sum(colSums(smsmnum)^2)  /
 111                 sum( sum(colSums(smsmWX) * colSums(smsmWY)) )
 112             Xwer_dist[i, j] <- sqrt(m * lscvect * (1 - wer2))
 113             Xwer_dist[j, i] <- Xwer_dist[i, j]
 114         }
 115     }
 116     diag(Xwer_dist) <- numeric(n)
 117     Xwer_dist
 118 }