I spent a decent chunk of my morning trying to figure out how to construct a sparse adjacency matrix for use with graph.adjacency(). I’d have thought that this would be rather straight forward, but I tripped over a few subtle issues with the Matrix package. My biggest problem (which in retrospect seems rather trivial) was that elements in my adjacency matrix were occupied by the pipe symbol.

> adjacency[1:10,1:10]
10 x 10 sparse Matrix of class 'ngCMatrix'
                         
 [1,] . . . . . | . . . .
 [2,] . . . . . . . | . .
 [3,] . . . . . . . . . .
 [4,] . . . . . . . . . .
 [5,] . . . . | . . . . .
 [6,] . . . . . . . . . .
 [7,] . . . . . . . . . .
 [8,] . . . . . . . . . .
 [9,] . . . . . . . . . .
[10,] . | . . . . . . . .

Of course, the error message I was encountering didn’t point me to this fact. No, that would have been far too simple! The solution is highlighted in the sample code below: you need to specify the symbol used for the occupied sites in the sparse matrix.

> library(Matrix)
> 
> set.seed(1)
> 
> edges = data.frame(i = 1:20, j = sample(1:20, 20, replace = TRUE))
> 
> adjacency = sparseMatrix(i = as.integer(edges$i),
+                          j = as.integer(edges$j),
+                          x = 1,
+                          dims = rep(20, 2),
+                          use.last.ij = TRUE
+ )

The resulting adjacency matrix then looks like this:

> adjacency[1:10,1:10]
10 x 10 sparse Matrix of class 'dgCMatrix'
                         
 [1,] . . . . . 1 . . . .
 [2,] . . . . . . . 1 . .
 [3,] . . . . . . . . . .
 [4,] . . . . . . . . . .
 [5,] . . . . 1 . . . . .
 [6,] . . . . . . . . . .
 [7,] . . . . . . . . . .
 [8,] . . . . . . . . . .
 [9,] . . . . . . . . . .
[10,] . 1 . . . . . . . .

And can be passed into graph.adjacency() without any further issues.

> library(igraph)
> graph = graph.adjacency(adjacency, mode = "undirected")

The post Graph from Sparse Adjacency Matrix appeared first on Exegetic Analytics.