dohts.R

library(data.tree)
library(xts)
f <- read.csv("https://gist.githubusercontent.com/gluc/3969bcfb2ddce61d8bdec2f52c133674/raw/16b577a6842492153db8e95ede0062a6187e60ab/hts-data.csv", stringsAsFactors = FALSE)

# construct tree
# ==============

# you may chose a different hierarchy order
dfs <- data.frame(pathString = paste("rt", df$u_loc, df$u_code, df$c_code, df$citizenship, df$grade, sep = "/"), stringsAsFactors = FALSE)

dfs <- data.frame(pathString = unique(dfs[,c("pathString")]), stringsAsFactors = FALSE)

# set the index for future reference / debugging
dfs$idx <- rownames(dfs)

#convert to a tree
tree <- FromDataFrameTable(dfs)


# set timeseries
# ==============

Extractor <- function(node) {
  #you could be more generic and create the hierarchic timeseries for intermediate levels as well
  path <- node$path
  u_loc <- path[[2]]
  u_code <- path[[3]]
  c_code <- path[[4]]
  citizenship <- path[[5]]
  grade <- path[[6]]
  
  ts <- df[df$c_code == c_code & df$u_code == u_code & df$u_loc == u_loc & df$grade == grade & df$citizenship == citizenship, c("year", "eftsl")]
  xts <- xts(ts$eftsl, order.by = as.Date(paste(ts$year, "01", "01", sep = "-")))
  #not such a good idea, as more nodes than letters in the alphabet
  #colnames(xts) <- paste( letters[ rev( node$Get("position", traversal = "ancestor") ) ], collapse = "")
  colnames(xts) <- node$idx
  node$timeseries <- xts
  
}

Do(tree$leaves, Extractor)


# extract hts matrix
# ==================

library(hts)
y <- do.call(cbind, Get(tree$leaves, function(n) n$timeseries, simplify = FALSE))

GetCount <- function(node) {
  ch <- NULL
  #browser()
  for (chld in node$children) {
    if (isNotLeaf(chld)) ch <- append(ch, GetCount(chld))
  }


  if (length(ch) > 0) res <- list(node$count, ch)
  else res <- node$count
  return (res)
}

nodes <- GetCount(tree)

myhts <- hts(y, nodes)

hts-data.csv