Proposal: more readable edge mutations based on node attributes

[Fuschi]

Hi,

first of all, thanks for tidygraph, which I find very useful for network analysis within the tidyverse ecosystem.

I would like to propose a small helper for edge mutation in tidygraph, aimed at making common node-based edge operations more readable and closer to a dplyr-style expression.

Motivation

When mutating edges based on node attributes, the current idiom typically relies on .N() together with explicit indexing via from / to. For example:

flo_tidy %>%
  activate("edges") %>%
  mutate(
    to_medici = (.N()$name[from] == "Medici" | .N()$name[to]   == "Medici"))

This approach is flexible and expressive, but can feel somewhat redundant, since node attributes need to be accessed explicitly through .N() and indexing.

Proposal

I experimented with a minimal helper that precomputes the two edge endpoints and exposes a small set of functions inside mutate(), such as:

• from(var) / to(var) to access node attributes at edge endpoints
• either(expr), both(expr), one(expr), neither(expr) for symmetric nodes logic

Using this approach, the example above becomes:

flo_tidy %>%
  proposed_edge_mutate(
    to_medici = either(name == "Medici")
  )

Example using a node-level variable

flo_tidy %>%
  activate("nodes") %>%
  mutate(centrality = centrality_power()) %>%
  activate("edges") %>%
  mutate(
    mean_centrality = (.N()$centrality[from] + .N()$centrality[to]) / 2)

Equivalent proposed syntax:

flo_tidy %>%
  proposed_edge_mutate(
    mean_centrality = (from(centrality) + to(centrality)) / 2)

Results are identical (verified via all.equal()).

Scope

I would be interested in your thoughts on whether an interface like this could fit within tidygraph, or if it would be considered out of scope.

Best,
Alessandro

---

Minimal proof-of-concept implementation

proposed_edge_mutate <- function(graph, ...) {
  if (!inherits(graph, "tbl_graph")) {
    rlang::abort("`graph` must be a tidygraph `tbl_graph`.")
  }

  local_env <- environment()

  nodes <- graph %>%
    tidygraph::activate("nodes") %>%
    tibble::as_tibble() %>%
    dplyr::mutate(.node_id = dplyr::row_number())

  edges <- graph %>%
    tidygraph::activate("edges") %>%
    tibble::as_tibble()

  if (!all(c("from", "to") %in% names(edges))) {
    rlang::abort("Edge table must contain `from` and `to` columns.")
  }

  tbl_from <- edges %>%
    dplyr::select(from) %>%
    dplyr::left_join(nodes, by = dplyr::join_by(from == .node_id))

  tbl_to <- edges %>%
    dplyr::select(to) %>%
    dplyr::left_join(nodes, by = dplyr::join_by(to == .node_id))

  from <- function(var) {
    var_sym <- rlang::ensym(var)
    dplyr::pull(tbl_from, !!var_sym)
  }

  to <- function(var) {
    var_sym <- rlang::ensym(var)
    dplyr::pull(tbl_to, !!var_sym)
  }

  either <- function(expr) {
    expr_quo <- rlang::enquo(expr)
    rlang::eval_tidy(expr_quo, data = tbl_from) |
      rlang::eval_tidy(expr_quo, data = tbl_to)
  }

  both <- function(expr) {
    expr_quo <- rlang::enquo(expr)
    rlang::eval_tidy(expr_quo, data = tbl_from) &
      rlang::eval_tidy(expr_quo, data = tbl_to)
  }

  neither <- function(expr) {
    expr_quo <- rlang::enquo(expr)
    !(rlang::eval_tidy(expr_quo, data = tbl_from) |
        rlang::eval_tidy(expr_quo, data = tbl_to))
  }

  one <- function(expr) {
    expr_quo <- rlang::enquo(expr)
    rlang::eval_tidy(expr_quo, data = tbl_from) !=
      rlang::eval_tidy(expr_quo, data = tbl_to)
  }

  quos <- rlang::enquos(...)
  quos <- lapply(quos, function(q) {
    rlang::new_quosure(
      expr = rlang::quo_get_expr(q),
      env  = local_env
    )
  })

  graph %>%
    tidygraph::activate("edges") %>%
    dplyr::mutate(!!!quos)
}