R/gen_adj_mat.R

In yongsu-lee/NOTEARGIS: Learn Directed Acyclic Graphs with Grouped Variables

#' Generate a true adjacency matrix
#'
#' @description Generate a desired graph (with \code{igraph} format) with corresponding adjacency matrix as an binary matrix. If \eqn{(i,j)} is one, it implies there exists an edge \eqn{i -> j}.
#'
#' @param n_nodes positive integer scalar (larger than 1), the number of nodes of the desired graph. For the Bipartite graph, it should be even.
#' @param graph_type character scalar, one of types of graphs in the simulation studies. See details below for more information.
#' @param seed integer scalar, in order to reproduce the same graph.
#'
#' @return \item{A_true}{an adjacency matrix of the \code{graph_true}}
#'
#' @export
#' @import igraph pcalg
#' @details There are seven graph types in the simulation studies: "bi" for Bipartite graph, "sf" for Scale-Free graph, "sw" for Small-World, "tree" for the tree with two children, "rand" for random DAG, and "chain" for a Chain graph. Except "rand", all the other graphs generated based on R package \strong{igraph}. The Random DAG is generated based on R package \strong{pcalg}.
gen_adj_mat=function(n_nodes, graph_type=c("bi","rand","sf","sw","tree"), seed=NA){

  graph_type = match.arg(graph_type)
  if (is.na(seed)) seed = 1

  if (graph_type == "bi" & (n_nodes %% 2 == 1) ) {
    stop ("Number of nodes should be even for Bipartitie graph.")
  }

  if (n_nodes == 2){
    A_true = matrix(c(0,0,1,0), 2, 2)
    graph_true <- igraph::graph_from_adjacency_matrix( A_true )
    A_true = igraph::as_adjacency_matrix(graph_true)

  } else {
    while(TRUE){

      switch(graph_type,

             "bi" = {
               if (!is.na(seed)) set.seed(seed)
               graph_true = igraph::sample_bipartite(
                 n1 = round(0.5*n_nodes), n2 = round(0.5*n_nodes),
                 type = "Gnm", m = n_nodes, directed = TRUE, mode = "in")
               A_true = igraph::as_adjacency_matrix(graph_true) },

             "rand" = {
               if (!is.na(seed)) set.seed(seed)
               graph_temp = pcalg::randDAG(n_nodes, d = 2)
               A_true = (pcalg::wgtMatrix(graph_temp) > 0)*1
               graph_true <- igraph::graph_from_adjacency_matrix( A_true )
               A_true = igraph::as_adjacency_matrix(graph_true) },

             "sf" = {
               if (!is.na(seed)) set.seed(seed)
               graph_true = igraph::sample_pa(n = n_nodes, directed = T)
               A_true = igraph::as_adjacency_matrix(graph_true) },

             "sw" = {
               if (!is.na(seed)) set.seed(seed)
               graph_temp <- igraph::sample_smallworld(
                 dim = 1, size = n_nodes, nei = 2,  p = 0.5)
               A_temp <- igraph::as_adjacency_matrix(graph_temp)
               graph_true = igraph::as.directed(graph_temp, mode="arbitrary")
               A_true = igraph::as_adjacency_matrix(graph_true)},

             "tree" = {
               if (!is.na(seed)) set.seed(seed)
               graph_true = igraph::graph.tree(n = n_nodes, children = 2)
               A_true = igraph::as_adjacency_matrix(graph_true) }

      )

      if (igraph::is_dag(graph_true)) break

      seed <- seed + 1

    }
  }

  A_true = as.matrix(A_true)
  colnames(A_true) = paste0("Z",1:n_nodes)
  rownames(A_true) = paste0("Z",1:n_nodes)
  V(graph_true)$name <- paste0("Z",1:n_nodes)
  attr(A_true, "graph_type") = graph_type
  attr(A_true, "seed") = seed
  attr(A_true, "graph_true") = graph_true
  return(A_true)

}