Adding edge list from file too large to fit into memory

Dear Graph-Tool community,

I am trying to construct a graph with a large number of edges, using an np file as an edge_list.
There are 125760 vertices and an edge list of length (7907725920, 2).

In order to use the npy edge_list file, I have needed to load the edge list as a readable memmap because, at a size of 126Gb, it is far too large to load into memory. But, when calling add_edge_list to this memmap, I think it is still being loaded into memory as the RAM will fill and the python session crash. I suppose the alternative is that the graph object becomes too large to hold into memory, but with previous large graphs I did not find this to be the problem. Does anybody have a solution to this issue?

Lastly, after I find a means to add this number of edges, I need to assign weights to the edges, again from a memmap file due to its size, which gives me the same problem. Any advice?

Sample code:

#prime the graph with the number of vertices
g = Graph(directed=False)
g.add_vertex(125760)

#load the edge list as memmap and add it
idx_indi_mmap = np.load('idx_indi.npy', mmap_mode='r’)
idx_indi_mmap.shape #(7907725920, 2)

g.add_edge_list(idx_indi_mmap) #script will crash at this point from filling the RAM

#Then want to add weights by taking the indices from another memmap object
node_matrix = np.load('node_matrix.npy', mmap_mode='r’)
node_matrix.shape #(125760, 125760)
weights = node_matrix[idx_indi_mmap]
ew = g("double")
ew.a = weights
g.ep[‘weight'] = ew

Thank you for your help!
James

attachment.html (2.54 KB)

Dear Graph-Tool community,

I am trying to construct a graph with a large number of edges, using an
np file as an edge_list.
There are 125760 vertices and an edge list of length (7907725920, 2).

In order to use the npy edge_list file, I have needed to load the edge
list as a readable memmap because, at a size of 126Gb, it is far too
large to load into memory. But, when calling add_edge_list to this
memmap, I think it is still being loaded into memory as the RAM will
fill and the python session crash. I suppose the alternative is that the
graph object becomes too large to hold into memory, but with previous
large graphs I did not find this to be the problem. Does anybody have a
solution to this issue?

I'm not sure what kind of solution you are expecting. Do you want
Graph.add_edge_list() *not* to load the edges into memory?

Lastly, after I find a means to add this number of edges, I need to
assign weights to the edges, again from a memmap file due to its size,
which gives me the same problem. Any advice?

*Sample code:*

#prime the graph with the number of vertices
g = Graph(directed=False)
g.add_vertex(125760)

#load the edge list as memmap and add it
idx_indi_mmap = np.load('idx_indi.npy', mmap_mode='r’)
idx_indi_mmap.shape #(7907725920, 2)

g.add_edge_list(idx_indi_mmap) #script will crash at this point from
filling the RAM

#Then want to add weights by taking the indices from another memmap object
node_matrix = np.load('node_matrix.npy', mmap_mode='r’)
node_matrix.shape #(125760, 125760)
weights = node_matrix[idx_indi_mmap]
ew = g("double")
ew.a = weights
g.ep[‘weight'] = ew

You might save some intermediary memory by loading the weights in one go
with Graph.add_edge_list() (see the 'eprops' parameter), which requires
a numpy array with three values, (source, target, weight).

However, if you do not have enough RAM to hold the entire graph +
weights into memory, this is also not going to work.

Best,
Tiago