如何在R中生成具有不同图层级中心化程度的igraph图？

Great question! The built-in sample_* functions in R's igraph package are fantastic for generating graphs from standard models, but they don't let you directly specify a target centralization value. Luckily, there are a couple of practical workarounds to create graphs with your desired centralization levels (0.50, 0.30, 0.00, etc.). Let's walk through the most reliable approaches.

1. Iterative Adjustment (Most Flexible Method) #

This approach starts with an extreme graph (like a star graph with centralization = 1.0) and iteratively modifies edges to nudge the centralization toward your target. It works for any centralization metric (degree, betweenness, closeness) and gives you precise control.

Step-by-Step Code #

First, let's write a reusable function that generates a graph with a target degree centralization (you can swap in other centralization metrics easily):

library(igraph)

generate_target_centralization_graph <- function(n_nodes, target_cent, max_iter = 1500, tolerance = 0.01) {
  # Start with a star graph (max possible degree centralization = 1.0)
  g <- make_star(n_nodes, mode = "undirected")
  current_cent <- centralization_degree(g)$centralization
  
  # If target is near 0, generate a random graph with the same number of edges
  if (target_cent < tolerance) {
    return(sample_gnm(n_nodes, m = ecount(g), directed = FALSE))
  }
  
  iter <- 0
  # Iteratively adjust edges until we hit the target (or max iterations)
  while (abs(current_cent - target_cent) > tolerance && iter < max_iter) {
    # Pick a random edge connected to the star's center node
    center_edges <- E(g)[V(g)[1] %--% V(g)[-1]]
    if (length(center_edges) == 0) break  # No more center edges to modify
    
    # Remove that center edge
    g <- delete_edges(g, sample(center_edges, 1))
    
    # Add a new edge between two random non-center nodes
    non_center <- V(g)[-1]
    new_edge_pair <- sample(non_center, 2, replace = FALSE)
    g <- add_edges(g, c(new_edge_pair$name, new_edge_pair$name[2]))
    
    # Recalculate current centralization
    current_cent <- centralization_degree(g)$centralization
    iter <- iter + 1
  }
  
  # Warn if we couldn't hit the target within max iterations
  if (abs(current_cent - target_cent) > tolerance) {
    warning(paste("Warning: Could not reach target centralization. Current value:", round(current_cent, 3)))
  }
  
  return(g)
}

Test the Function #

Now you can generate your graphs with specific centralization values:

# Set seed for reproducibility
set.seed(123)

# Generate graphs with target centralizations
graph_1 <- generate_target_centralization_graph(n_nodes = 20, target_cent = 0.50)
graph_2 <- generate_target_centralization_graph(n_nodes = 20, target_cent = 0.30)
graph_3 <- generate_target_centralization_graph(n_nodes = 20, target_cent = 0.00)

# Verify the results
cat("Graph 1 centralization:", round(centralization_degree(graph_1)$centralization, 2), "\n")
cat("Graph 2 centralization:", round(centralization_degree(graph_2)$centralization, 2), "\n")
cat("Graph 3 centralization:", round(centralization_degree(graph_3)$centralization, 2), "\n")

2. Rewiring for Approximate Values #

If you don't need pinpoint precision, you can use igraph's rewire() function to gradually transform a high-centralization graph into a low-centralization one. This is faster but less precise.

# Start with a star graph (centralization = 1.0)
star_graph <- make_star(20, mode = "undirected")

# Rewire 30% of edges to get moderate centralization (~0.7)
graph_moderate <- rewire(star_graph, with = keeping_degseq(niter = ecount(star_graph)*0.3))

# Rewire 80% of edges to get low centralization (~0.2)
graph_low <- rewire(star_graph, with = keeping_degseq(niter = ecount(star_graph)*0.8))

Key Notes #

• Centralization Metric: The examples use degree centralization (centralization_degree). To use betweenness or closeness centralization, just replace that function with centralization_betweenness or centralization_closeness.
• Constraints: Centralization values are bounded by the graph size. For an undirected graph with n nodes, maximum centralization is 1.0 (star graph), and minimum is near 0 (random graph).
• Reproducibility: Always set set.seed() if you need consistent results across runs.

内容的提问来源于stack exchange，提问作者Erwin