Re: [igraph] arpack implementation in igraph is slow?

[Tamás Nepusz]

there are two reasons for this being slow. One is that ARPACK needs a

lot of iterations to converge, around a thousand. The other is that

your callback function itself takes a long time to compute. I measured

and more than 50% of the computation is spent in the callback.

How about replacing the callback with sparse matrix multiplications from a sparse matrix package (I don't know how it is called in R)? Basically, you could construct the Laplacian manually from the adjacency matrix (which igraph can return in sparse matrix format), and then just use sparse matrix multiplications within the callback. Presumably a decent sparse matrix package should implement these in C, so it is likely to be faster than a for loop.

Best,

Tamas