|
From: | Gang Su |
Subject: | Re: [igraph] Suggestion for a new community detection function |
Date: | Fri, 09 May 2008 15:12:54 -0400 |
User-agent: | Thunderbird 2.0.0.14 (Windows/20080421) |
Okay, there's actually another paper you may want to read: http://arxiv.org/abs/0704.3759 What the author says makes sense. I have tried on two different datasets:One with 2k nodes, 10k edges, and the other with 10k nodes, 100k edges. Almost all algorithms based on Q give similar number of clusters on two different datasets. That is, the number of clusters given by each algorithm is relatively stable, for large datasets you are likely to get large clusters and you may need to run the algorithm iteratively to get smaller clusters.
Right now i have tried different algorithms, it seems that spinglass and ken wakita's method(not in igraph, taking consolidation ratio into account) works best on my dataset. walk trap and fast greedy works worst, edgebetweenness works somewhere in between but too slow, leading eigenvector never worked :((Doesn't converge even with 1 million iterations on all my datasets.)
Gang Tamas Nepusz wrote:
Tamas, when you say to implement in python that means you are not going to modify the C backend?I'm not planning to implement it in C yet, only in Python - but not in the next two weeks, I assume :( I can send you the source code when (and _if_ ;)) I implemented it. In case there is substantial demand for it and the algorithm gets widespread in the network research community, I will give it a try and translate it to C.
[Prev in Thread] | Current Thread | [Next in Thread] |