jpfcst/meterdata.R at master · jayesh291/jpfcst · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58

#transpose to have dates as columns
mtr.dtscol <- cast(mtr.f, id ~ ts, value = "val")
write.csv(mtr.dtscol, file = "./outs/transposed.csv")

#transpose to have meters as columns
mtr.mtrcol <- t(mtr.dtscol)
C1 <- cor(mtr.mtrcol, use = "pairwise.complete.obs", method = "kendall")

write.csv(C1, "./outs/correlations.csv")

# rownames(C1) <- c(1:nrow(C1))
# colnames(C1) <- c(1:ncol(C1))

#cluster and plot
hc <- hclust(dist(C1))

pdf("./outs/dendogram.pdf")
plot(hc, cex = 0.2, hang = -1)
rhc <- rect.hclust(hc, k = 15, border = 2)
dev.off()

#get representative timeseries for each cluster
grpno <- NULL
rep.ids <- NULL
for (i in 1:length(rhc)) {
  # subset data
  grpno <-
    c(grpno, unlist(
      x = rhc[[i]],
      recursive = T,
      use.names = T
    ))
  id <- names(rhc[[i]])
  if (length(id) > 1) {
    mtr.clstr.i <- mtr.mtrcol[, c(id)]

    #get group wise correlations
    C2 <-
      cor(mtr.clstr.i, use = "pairwise.complete.obs", method = "kendall")
    C3 <- abs(C2)
    # C3$avg <- rowMeans(C3)
    write.csv(C3, paste0("./outs/correlations", i , ".csv"))

    #get representative timeseries id for each group
    cl.mns <- colMeans(C3)
    rep.mtr.id <- names(cl.mns)[match(max(cl.mns), cl.mns)]
    #rep.ids <- c(rep.ids, rep.mtr.id)
    rep.ids[i] <- rep.mtr.id

    #plot groups
    filter.cluster <- (mtr.f$id %in% c(id))
    plotmeterswithrep(mtr.data = mtr.f[which(filter.cluster),],
               filename = paste0("./outs/correlations", i , "-", rep.mtr.id, ".pdf"), rep.mtr.id = rep.mtr.id)
  }
}

print(rep.ids)