taynaud
commented
4 years ago Hello, When I run your snippet, I get very close modularities on both graph. Louvain method is not strictly deterministic (for instance, iteration order on nodes is important but no order is the correct order). Thus you may get different communities on several run, even on the same graph. I try to analyze this in http://hal.archives-ouvertes.fr/inria-00492058/en/
0.38 and 0.41 seems a rather extreme difference, but I have note manager to reproduce this.
Hi, I am confusing that I got different modularity in the same graph with different node labels. First, I got one modularity 0.382 in my network. Then I rename the node labels based on some sorted and I got 0.412. Anyone can tell me the reason? For simplicity, I test the karate club network and find the same problem. Here is an example for my code.
~~~~~~~~~~~~~~~import community import networkx as nx
G1 = nx.karate_club_graph()
rename nodes id
G2=nx.Graph() index={} id=0 for (i,j) in G1.edges(): if i not in index.keys(): index[i]=id id=id+1 if j not in index.keys(): index[j]=id id=id+1 G2.add_edge(index[i], index[j])
calculate the modularity
partition1 = community.best_partition(G1) Q1 = community.modularity(partition1,G1)
partition2=community.best_partition(G2) Q2 = community.modularity(partition2,G2)
print len(G1),len(G2) print G1.number_of_edges(), G2.number_of_edges() print Q1,Q2