Hi,
I am trying to use the random_rewire function to generate a new network, using the 'correlated' model. The network is pretty small (208 vertices, 594 edges - no loops or parallel edges). If I just run it with default parameters, it rejects about 300-400 of the moves and 10000 iterations takes about 1 minute. To improve the shuffling a bit, I tried to add persist=True. Now the problem is, after 2 hours, it has not even completed a single iteration. I am suspecting that it is stuck in an infinite loop somewhere, but I cannot figure out why. Would there be any special property of a graph which causes this?
For reference, these are the two commands I have tried executing:
K = 10000 # Finishes in about 1 minutes for i in range(K): graph_tool.generation.random_rewire(testg, model='correlated')
# After 2 hours and counting, still has not finished one iteration for i in range(K): graph_tool.generation.random_rewire(testg, model='correlated', persist=True) print "%i done" % i
Also, the question I am trying to answer is, given the original graph and properties, what is the size of the largest connected component observed when the network is rewired at random, preserving node degree.
Thanks in advance, David
On 05/14/2014 06:45 PM, david@harsk.dk wrote:
Hi,
I am trying to use the random_rewire function to generate a new network, using the 'correlated' model. The network is pretty small (208 vertices, 594 edges - no loops or parallel edges). If I just run it with default parameters, it rejects about 300-400 of the moves and 10000 iterations takes about 1 minute. To improve the shuffling a bit, I tried to add persist=True. Now the problem is, after 2 hours, it has not even completed a single iteration. I am suspecting that it is stuck in an infinite loop somewhere, but I cannot figure out why. Would there be any special property of a graph which causes this?
Yes, if you forbid parallel edges or self-loops, not all proposed edge rewires are possible (since they may generate parallel edges or self-loops). If you have persist=False, and if a move is not allowed, it is not made, and the algorithm moves to next edge. If you put persist=True it keeps trying until it succeeds, but you may get stuck in situations where all edge movements generate parallel edges or self-loops, which is what you are experiencing.
The persist option is there because it is useful in some very specific circumstances. In the vast majority of cases, you should just use persist=False, and increase the value of "niter" until you achieve equilibration.
Best, Tiago
-
david＠harsk.dk -
Tiago de Paula Peixoto