Contract a dataframe of an edge list by summing the contracted edge weights from/to two nodes

I have a dataframe df that contains data on edge weights between two pairs of nodes:

df <- data.frame(c("A","A","B","B","C","C"),
c("B","C","A","C","A","B"),
c(2,3,6,4,9,1))
colnames(df) <- c("node_from", "node_to", "weight")

print(df)
# Output:
  node_from node_to  weight
1     A     B       2
2     A     C       3
3     B     A       6
4     B     C       4
5     C     A       9
6     C     B       1

I would like to contract this dataframe by merging nodes A and B and summing all edge weights to and from these nodes with any other node, in this case C only. The result should be an edge list where the edges between A and B have disappeared and AB is now one node:

# some code to merge nodes A and B

print(df_contracted)
# Output:
  node_from node_to weight
1    AB     C      7
3     C    AB      10

Is there a way to do this efficiently for larger dataframes?

I could convert the dataframe to an actual graph using graph_from_data_frame from the igraph package and then the contract function, but given that I have to do this operation multiple times I'd rather not have to convert it then reconvert it back every time.

base R approach

With base R we can use aggregate + subset like below

aggregate(
  weight ~ .,
  subset(
    transform(
      df,
      node_from = gsub("A|B", "AB", node_from),
      node_to = gsub("A|B", "AB", node_to)
    ),
    node_from != node_to
  ),
  sum
)

which gives

  node_from node_to weight
1         C      AB     10
2        AB       C      7

igraph approach

Here is an option using contract from igraph

df %>%
  graph_from_data_frame() %>%
  contract(c(1, 1, 2), function(v) paste0(v, collapse = "")) %>%
  simplify() %>%
  get.data.frame()

which gives

  from to weight
1   AB  C      7
2    C AB     10

Here's a dplyr solution:

library(dplyr)
to.merge <- c('A', 'B')
merged.name <- paste(to.merge, collapse='')

df %>%
  mutate(across(c(node_from, node_to), 
                ~ if_else(.x %in% to.merge, merged.name, .x))) %>%
  group_by(node_from, node_to) %>%
  summarise(weight = sum(weight), .groups = "drop") %>%
  filter(node_from != node_to)
# # A tibble: 2 × 3
#   node_from node_to weight
#   <chr>     <chr>    <dbl>
# 1 AB        C            7
# 2 C         AB          10

It changes all from and to node names that are "A" or "B" to "AB", groups rows with the same combination of from_node and to_node, sums weights within these groups, and finally removes the AB<->AB self-loop.

You may subset the AB and BA rows away, next summarize by a grepl on 'C', and rbind.

subset(df, rowSums(sapply(df[1:2], grepl, pat='A|B')) != 2) |>
  {\(.) by(., grepl('C', .$node_from), \(x) {
    data.frame(t(sapply(x[1:2], \(z) paste(unique(z), collapse=''))), weight=sum(x$weight))
  })}() |> unname() |> do.call(what='rbind')
#   node_from node_to weight
# 1        AB       C      7
# 2         C      AB     10

Data:

df <- structure(list(node_from = c("A", "A", "B", "B", "C", "C"), node_to = c("B", 
"C", "A", "C", "A", "B"), weight = c(2, 3, 6, 4, 9, 1)), class = "data.frame", row.names = c(NA, 
-6L))