| 1 | ## File : 00_plots-energycon.r |
| 2 | ## Description : Using the full data matrix, we extract handy features to |
| 3 | ## cluster. |
| 4 | |
| 5 | rm(list = ls()) |
| 6 | |
| 7 | library(Rwave) # CWT |
| 8 | #library(cluster) # pam |
| 9 | #library(flexclust) # kcca |
| 10 | source("aux.r") # auxiliary clustering functions |
| 11 | source("sowas-superseded.r") # auxiliary CWT functions |
| 12 | source('http://eric.univ-lyon2.fr/~jcugliari/codes/functional-clustering.r') |
| 13 | setwd("~/recherche/03_projects/2014_EDF-Orsay-Lyon2/codes/") |
| 14 | |
| 15 | |
| 16 | ## 1. Read auxiliar data files #### |
| 17 | |
| 18 | #identifiants <- read.table("identifs.txt")[ ,1] |
| 19 | dates0 <- read.table("datesall.txt")[, 1] |
| 20 | dates <- dates0[grep("2009", dates0)] |
| 21 | #rm(dates0) |
| 22 | |
| 23 | #n <- length(identifiants) |
| 24 | p <- length(dates) |
| 25 | |
| 26 | #blocks <- c(rep(6500, 3), 5511) |
| 27 | |
| 28 | # table( substr(dates, 11, 15) ) # Sunlight time saving produces an |
| 29 | # unbalanced number of time points |
| 30 | # per time stepa across the year |
| 31 | |
| 32 | |
| 33 | ## 2. Process the large file #### |
| 34 | |
| 35 | # if(exists("con")) close(con) |
| 36 | # con <- file("~/tmp/2009_full.txt") # Establish a connection to the file |
| 37 | # open(con, "r") # Open the connection |
| 38 | # |
| 39 | # nb <- 4 |
| 40 | # actual <- readLines(con = con, n = nb )[-3] |
| 41 | # auxmat <- matrix(unlist(strsplit(actual, " ")), ncol = p + 1, byrow = TRUE) |
| 42 | # rm(actual) |
| 43 | # |
| 44 | # datamat <- t(apply(auxmat[, -1], 1, as.numeric)) |
| 45 | # rownames(datamat) <- substr(auxmat[, 1], 2, 7) |
| 46 | # rm(auxmat) |
| 47 | # |
| 48 | # datamat <- datamat[, 1:(48 * 14)] |
| 49 | # p <- ncol(datamat) |
| 50 | # |
| 51 | # auxDWT <- t(apply(datamat, 1, toDWT)) |
| 52 | # matcontrib <- t(apply(auxDWT, 1, contrib)) |
| 53 | # rm(auxDWT) |
| 54 | # |
| 55 | # close(con); rm(con) # close connection to the file |
| 56 | |
| 57 | |
| 58 | load("~/data/Irlande/Data_CER_clean/SME.RData") |
| 59 | |
| 60 | smp <- c(4, 79, 126) |
| 61 | |
| 62 | datamat <- t(SME[1:(48 * 7), smp]) |
| 63 | p <- ncol(datamat) |
| 64 | |
| 65 | auxDWT <- t(apply(datamat, 1, toDWT)) |
| 66 | matcontrib <- t(apply(auxDWT, 1, contrib)) |
| 67 | rm(auxDWT) |
| 68 | |
| 69 | matplot(t(datamat), type = "l", lty = 1, |
| 70 | col = 1:8, lwd = 2) |
| 71 | |
| 72 | pdf("~/courbes.pdf") |
| 73 | #par(mai = c(1, 1, 0.8, 0.6), mfcol = c(4, 2), cex = 2) |
| 74 | for(courbe in 1:ncol(SME)) { |
| 75 | plot(SME[1:(48 * 7), courbe], main = paste(courbe), |
| 76 | xlab = "", ylab = "", type = "l") |
| 77 | } |
| 78 | dev.off() |
| 79 | |
| 80 | |
| 81 | |
| 82 | |
| 83 | ## Plots for ENERGYCON full article #### |
| 84 | op <- par() |
| 85 | |
| 86 | col <- c("grey", "black", "black") |
| 87 | ## Curves |
| 88 | pdf('~/curves.pdf', width = 12) |
| 89 | par(mai = c(1.3, 1.3, 0.1, 0.1), cex = 1.2) |
| 90 | matplot(t(datamat), type = "l", lty = c(1, 1, 2), |
| 91 | lwd = 2, col = col, # c(1, 2, 4), |
| 92 | ylab = "Load", xlab = "Time (1/2 hours)") |
| 93 | legend("top", c("Cust. A", "Cust. B", "Cust. C"), |
| 94 | col = col, lty = c(1, 1, 2), ncol = 3, lwd = 2) |
| 95 | # matplot(scale(t(datamat), scale = F), |
| 96 | # type = "l", lty = 1, col = c(1, 2, 4), lwd = 2, |
| 97 | # ylab = "Load", xlab = "Time (1/2 hours)") |
| 98 | dev.off() |
| 99 | |
| 100 | normi <- function(x) x / max(x) |
| 101 | normi2 <- function(x) (x - min(x)) / (max(x) - min(x)) |
| 102 | |
| 103 | matplot(apply(datamat, 1, normi2), |
| 104 | type = "l", lty = 1, col = c(1, 2, 4), lwd = 2, |
| 105 | ylab = "Load", xlab = "Time (1/2 hours)") |
| 106 | |
| 107 | |
| 108 | cont_dist <- dist(scale(matcontrib)) |
| 109 | cont_dist <- cont_dist / max(cont_dist) |
| 110 | |
| 111 | |
| 112 | delta <- p |
| 113 | n <- nrow(datamat) |
| 114 | |
| 115 | ## _.a CWT -- Filtering the lowest freqs (>6m) #### |
| 116 | nvoice <- 4 |
| 117 | # noctave4 = 2^13 = 8192 half hours ~ 180 days |
| 118 | noctave4 <- adjust.noctave(N = delta, dt = 1, s0 = 2, |
| 119 | tw = 0, noctave = 10) |
| 120 | # 10 here represent 2^10 = 1024 half-hours ~ 2 weeks day |
| 121 | scalevector4 <- 2^(4:(noctave4 * nvoice) / nvoice) * 2 |
| 122 | lscvect4 <- length(scalevector4) |
| 123 | lscvect <- lscvect4 # i should clean my code: werFam demands a lscvect |
| 124 | Xcwt4 <- toCWT(datamat, noctave = noctave4, dt = 1, |
| 125 | scalevector = scalevector4, |
| 126 | lt = delta, smooth = FALSE, |
| 127 | nvoice = nvoice) # observations node with CWT |
| 128 | |
| 129 | Xcwt2 <- matrix(NA_complex_, nrow= n, ncol= 2 + length(c(Xcwt4[, ,1]))) |
| 130 | |
| 131 | |
| 132 | for(i in 1:n) |
| 133 | Xcwt2[i,] <- c(delta, lscvect, Xcwt4[,,i] / max(Mod(Xcwt4[,,i])) ) |
| 134 | |
| 135 | rm(conso, Xcwt4); gc() |
| 136 | |
| 137 | lscvect <- 41 ## very very nasty: toCWT changes scalevector to 41 (instead of |
| 138 | ## the original length--37--) |
| 139 | ## _.b WER^2 distances ######## |
| 140 | Xwer_dist <- matrix(0.0, n, n) |
| 141 | for(i in 1:(n - 1)){ |
| 142 | mat1 <- vect2mat(Xcwt2[i,]) |
| 143 | for(j in (i + 1):n){ |
| 144 | mat2 <- vect2mat(Xcwt2[j,]) |
| 145 | num <- Mod(mat1 * Conj(mat2)) |
| 146 | WX <- Mod(mat1 * Conj(mat1)) |
| 147 | WY <- Mod(mat2 * Conj(mat2)) |
| 148 | smsmnum <- smCWT(num, scalevector = scalevector4) |
| 149 | smsmWX <- smCWT(WX, scalevector = scalevector4) |
| 150 | smsmWY <- smCWT(WY, scalevector = scalevector4) |
| 151 | wer2 <- sum(colSums(smsmnum)^2) / |
| 152 | sum( sum(colSums(smsmWX) * colSums(smsmWY)) ) |
| 153 | Xwer_dist[i, j] <- sqrt(delta * lscvect * (1 - wer2)) |
| 154 | Xwer_dist[j, i] <- Xwer_dist[i, j] |
| 155 | } |
| 156 | } |
| 157 | diag(Xwer_dist) <- numeric(n) |
| 158 | |
| 159 | |
| 160 | Xwer_dist <- Xwer_dist / max(Xwer_dist) |
| 161 | |
| 162 | |
| 163 | pdf("~/cmdscale.pdf", width = 14) |
| 164 | #layout(matrix(1:2, 1, 2)) |
| 165 | par(mai = c(1, 1, 0.8, 0.6), mfcol = c(1, 2), cex = 2) |
| 166 | plot(cmdscale(cont_dist), pch = c("A", "B", "C"),#c(15, 17, 19), |
| 167 | main = "RC based distance", |
| 168 | #col = c(1, 2, 4), |
| 169 | xlim = c(-1, 1), ylim = c(-1, 1), |
| 170 | xlab = "", ylab = "", asp = 1) |
| 171 | abline(h= 0, lwd = 2); abline(v = 0, lwd = 2); grid(lty = 1) |
| 172 | |
| 173 | plot(cmdscale(Xwer_dist), pch = c("A", "B", "C"),#c(15, 17, 19), |
| 174 | main = "WER distance", |
| 175 | #col = c(1, 2, 4), |
| 176 | xlim = c(-1, 1), ylim = c(-1, 1), |
| 177 | xlab = "", ylab = "", asp = 1) |
| 178 | abline(h= 0, lwd = 2); abline(v = 0, lwd = 2); grid(lty = 1) |
| 179 | dev.off() |
| 180 | |
| 181 | |
| 182 | |
| 183 | ## Effectives distribution |
| 184 | #load('../res/clfit500.Rdata') |
| 185 | pdf("~/distro500.pdf", width = 12) |
| 186 | plot(sort(table(clfit$clustering), decreasing = TRUE), |
| 187 | type ="s", ylab = "Effectives", xlab = "Class", lwd = 2) |
| 188 | abline(v = 200) |
| 189 | #dev.off() |
| 190 | |