Alejandro Candioti
Alejandro Candioti
> Thanks! I appreciate that. I really was convinced it was the order of the nodes. Happy to help! Yeah, you were right though in that the order of nodes...
I took a look at subgraph logic. There is nothing wrong with ego graph code. Your code is hitting a subgraph optimization that makes the node iterator go over the...
> If you have sortable nodes, you can use something like this to get a "sorted graph". > Note that while node order is the order in which nodes are...
> > Even if we change subgraph to use dict, if the caller passes an iterable of nodes with a different order than the underlying graph, then subgraph won't preserve...
> I don't know enough about behind the scenes of the library in full, but one simple thing I see is allowing the user to have an option to sort...
That's great to hear!
If you want to optimize for dense cases you can do the same trick as in https://github.com/networkx/networkx/pull/7971. What do you think? ``` post = list(nx.dfs_postorder_nodes(G.reverse(copy=False), source=source)) n = len(post) seen...
Seems like it's much better now! ``` n=10, p=1.0 fast_kscc time: 0.0076 seconds networkx kosaraju time: 0.0008 seconds tarjan scc time: 0.0006 seconds ale scc time: 0.0002 seconds networkx scc...
Is Erdos Renyi really representative here? How many SCCs do these random graphs usually have? Do you have any benchmark suite covering the worst case?
Thanks! Can you clear the unrelated files you committed?