Creating a dataframe of edges from node information in R?

Question

I am trying to create an edge list based on binary splits. If I have a data frame that only contains the node number and some other metric, then I can manually create an edge list for the nodes. For example, if my data frame looks like this:

dfTest <- data.frame(
  node = c(1,2,3,4,5),
  var = c("milk", NA, "coffee", NA, NA),
  isLeaf = c(F, T, F, T, T)
)

> dfTest
  node    var isLeaf
1    1   milk  FALSE
2    2      TRUE
3    3 coffee  FALSE
4    4      TRUE
5    5      TRUE

Then, based on the var or isLeaf column, I can manually create an edge list to connect the nodes. For example, As node 2 is a leaf, I know that node 1 must go to node 2. Then (as they are binary splits) I know node 1 must also connect to node 3. And as node 4 and 5 are leaf nodes, I know that they must split on node 3.

Manually creating the edge list would look like this:

edges <- data.frame(
  from = c(1, 1, 3, 3),
  to   = c(2, 3, 4, 5)
)

The to column is easy to find... it will always be c(2:length(dfTest$nodes)). In this case 2,3,4,5. But the from column is proving difficult to find.

Just for a visual aid, the resulting tree would look like this:

Is there any way to do this without having to manually work out the edges?

EDIT: In response to an answer, I'm adding a slightly larger dataset to use:

dfTest <- data.frame(
  node = c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11),
  var = c("milk", "milk", NA, NA, "coffee", "sugar", NA, NA, "milk", NA, NA),
  isLeaf = c(F, F, T, T, F, F, T, T, F, T, T)
)

A little explanation: From the var column I know that milk (the root/node 1) splits to another milk (node 2). I can then see that node 2 splits to NA (node 3) and NA (node 4). As I know they are binary splits, I know that node 2 cant split any further. So, I must go back to the previous node that only had 1 split… in this case node 1 (i.e., milk) which then splits to the right on coffee (node 5). Again, as they are binary splits, I now know that coffee (node 5) must split to sugar (node 6). Sugar (node 6) is followed by 2 NAs (node 7 & 8 ). Now, I must go back to coffee (node 5) and split to the right to get milk (node 9) which splits to 2 NAs (node 10 &11)

The desired node/edge list should look like this:

edges <- data.frame(
  from = c(1,2,2,1,5,6,6,5,9,9),
  to   = c(2,3,4,5,6,7,8,9,10,11)
)

Which would produce a tree like this:

user_12158757 · Answer

As per your update, perhaps you can try the code below

grps <- split(dfTest, ~ cumsum(!isLeaf))

edges <- do.call(
  rbind,
  lapply(
    grps,
    function(x) {
      with(x, expand.grid(from = node[!isLeaf], to = node[isLeaf]))
    }
  )
)

for (k in seq_along(grps)) {
  if (nrow(grps[[k]]) == 1) {
    lleaf <- with(grps[[k + 1]], node[!isLeaf])
    rleaf <- with(grps[[k + 2]], node[!isLeaf])
    edges <- rbind(edges, data.frame(from = grps[[k]]$node, to = c(lleaf, rleaf)))
  }
}

edges <- `row.names<-`(edges[with(edges, order(from, to)), ], NULL)

which gives

Visualization

library(igraph)
graph_from_data_frame(edges) %>%
  plot(layout = layout_as_tree)

Creating a dataframe of edges from node information in R?

1 Answers

Visualization