converting adjacency matrix into numpy is slow

[idc9]

Converting the adjacency matrix into a numpy array using the following method is pretty slow. I tried a few other ways of doing this to no avail.

G.summary()

IGRAPH DN-- 27885 234312 --

%%time
A = np.array(G.get_adjacency().data)

CPU times: user 3min 23s, sys: 1min 18s, total: 4min 42s
Wall time: 5min 31s

.get_adjacency() took about 3.5 min and np.array(--) took about 2 minutes.

Q: is there a faster way of getting the adjacency matrix as a numpy array?

Q: why is this method so slow (just curious)?

I'm using python 2.7, OS X 10.11.6, igraph 0.7.1, and numpy 1.11.1

[ntamas]

why is this method so slow (just curious)?

G.get_adjacency() gives you a dense matrix with 27885 x 27885 elements, and all of it is then copied into a NumPy array - that's why it's slow. np.array() would also be a dense matrix for what it's worth, so it's gonna be slow anyway as soon as you start doing some operations on the dense matrix. You need a sparse matrix instead.

is there a faster way of getting the adjacency matrix as a numpy array?

If you really want a NumPy array and not a SciPy sparse matrix:

n = G.vcount()
a = np.zeros((n, n))
rows, cols = zip(*G.get_edgelist())
a[rows, cols] = 1

If you are on Python 3.x, you probably need to wrap the zip() call in list() first.