1 ## File : 05_cluster2stepWER.r
6 setwd("~/ownCloud/projects/2014_EDF-Orsay-Lyon2/codes/")
10 #library(flexclust) # kcca
11 source("aux.r") # auxiliary clustering functions
12 source("sowas-superseded.r") # auxiliary CWT functions
14 ## 1. Read auxiliar data files ####
16 identifiants <- read.table("identifs.txt")[ ,1]
17 dates0 <- read.table("datesall.txt")[, 1]
18 dates <- as.character(dates0[grep("2009", dates0)])
21 n <- length(identifiants)
22 p <- delta <- length(dates)
24 synchros09 <- t(as.matrix(read.table("~/tmp/2009_synchros200RC.txt")))
25 #synchros09 <- t(as.matrix(read.table("~/tmp/2009_synchros200-random.txt")))
27 nas <- which(is.na(synchros09)[, 1]) # some 1/1/2009 are missing
28 synchros09[nas, 1] <- rowMeans(synchros09[nas, 2:4])
30 imput09 <- synchros09[, 4180:4181] %*% matrix(c(2/3, 1/3, 1/3, 2/3), 2)
31 synchros09 <- cbind(synchros09[, 1:4180], imput09, synchros09[, 4181:17518])
33 conso <- synchros09[-201, ]; # series must be on rows
39 ## 2. Compute WER distance matrix ####
41 ## _.a CWT -- Filtering the lowest freqs (>6m) ####
43 # # noctave4 = 2^13 = 8192 half hours ~ 180 days
44 # noctave4 <- adjust.noctave(N = delta, dt = 1, s0 = 2,
45 # tw = 0, noctave = 13)
46 # # 4 here represent 2^5 = 32 half-hours ~ 1 day
47 # scalevector4 <- 2^(4:(noctave4 * nvoice) / nvoice) * 2
48 # lscvect4 <- length(scalevector4)
49 # lscvect <- lscvect4 # i should clean my code: werFam demands a lscvect
50 # Xcwt4 <- toCWT(conso, noctave = noctave4, dt = 1,
51 # scalevector = scalevector4,
52 # lt = delta, smooth = FALSE,
53 # nvoice = nvoice) # observations node with CWT
55 # #Xcwt2 <- matrix(0.0, nrow= n, ncol= 2 + delta * lscvect)
56 # #Xcwt2 <- matrix(NA_complex_, nrow= n, ncol= 2 + length((c(Xcwt4[,,1]))))
59 # Xcwt2[i,] <- c(delta, lscvect, Xcwt4[,,i] / max(Mod(Xcwt4[,,i])) )
61 # #rm(conso, Xcwt4); gc()
63 # ## _.b WER^2 distances ########
64 # Xwer_dist <- matrix(0.0, n, n)
65 # for(i in 1:(n - 1)){
66 # mat1 <- vect2mat(Xcwt2[i,])
67 # for(j in (i + 1):n){
68 # mat2 <- vect2mat(Xcwt2[j,])
69 # num <- Mod(mat1 * Conj(mat2))
70 # WX <- Mod(mat1 * Conj(mat1))
71 # WY <- Mod(mat2 * Conj(mat2))
72 # smsmnum <- smCWT(num, scalevector = scalevector4)
73 # smsmWX <- smCWT(WX, scalevector = scalevector4)
74 # smsmWY <- smCWT(WY, scalevector = scalevector4)
75 # wer2 <- sum(colSums(smsmnum)^2) /
76 # sum( sum(colSums(smsmWX) * colSums(smsmWY)) )
77 # Xwer_dist[i, j] <- sqrt(delta * lscvect * (1 - wer2))
78 # Xwer_dist[j, i] <- Xwer_dist[i, j]
81 # diag(Xwer_dist) <- numeric(n)
83 # save(Xwer_dist, file = "../res/2009_synchros200WER.Rdata")
84 # save(Xwer_dist, file = "../res/2009_synchros200-randomWER.Rdata")
86 load("../res/2009_synchros200WER.Rdata")
87 #load("../res/2009_synchros200-randomWER.Rdata")
89 ## 3. Cluster using WER distance matrix ####
91 #hc <- hclust(as.dist(Xwer_dist), method = "ward.D")
94 # #clust <- cutree(hc, 2)
98 #pamfit <- pam(tdata[-201, ci$selectv], k = K)
99 pamfit <- pam(as.dist(Xwer_dist), k = K, diss = TRUE)
101 #table(pamfit$clustering)
103 SC <- matrix(0, ncol = p, nrow = K)
105 clustfactor <- pamfit$clustering
107 # clustk <- which(clustfactor == k)
108 # if(length(clustk) > 0) {
109 # if(length(clustk) > 1) {
110 # SCk <- colSums(synchros09[which(clustfactor == k), ])
112 # SCk <- synchros09[which(clustfactor == k), ]
114 # SC[k, ] <- SC[k, ] + SCk
119 #write.table(clustfactor, file = paste0("~/tmp/clustfactorRC", K, ".txt"))
120 #write.table(clustfactor, file = "~/tmp/clustfactor3.txt")
121 #write.table(clustfactor, file = paste0("~/tmp/clustfactorWER", K, ".txt"))
122 write.table(clustfactor, file = paste0("~/tmp/clustfactor-randomWER", K, ".txt"))
127 # matplot(t(SC)[48*10 + 1:(48*30), ], type = 'l', ylab = '',col = 1:3, lty = 1)
128 # matplot(t(SC)[48*100 + 1:(48*30), ], type = 'l', ylab = '', col = 1:3, lty = 1)