|
From: | David Hunkins |
Subject: | Re: [igraph] 'decompose.graph' versus 'clusters' |
Date: | Sat, 19 Jul 2008 10:38:51 -0700 |
Thanks Gabor. Don't you take a break on the weekends?! I'm going to try what you suggest, since that seems the best path to find the most interesting 'connectors' in the large clusters. By eliminating clusters up to size 5, I can get rid of 73% of the clusters, which should improve on the O(c(E+V)) calculation. I am loath to kill the clusters up to size 10 since that represents 'interesting' green growth of the social network (and only kills another 10%). Also, thanks for the new version of betweenness--I'll let you know the results. Because 'clusters' appears to be performing so well, could you look at the following and tell me what you think? I had done some tests on 'clusters' versus 'decompose.graph', and I found the following table of results, which suggest to me that somehow 'clusters' is very much faster than 'decompose.graph': clusters decompose.graph 5k .005sec 1.155sec 10k .008sec 3.824sec 20k .008sec 12.257sec 5M 2.905sec <did not finish after 24 hours> 'Clusters' is amazingly fast. I hadn't realized that constructing the subgraph from a large graph would be so computationally intensive compared to just identifying the members of a cluster. But I wonder, if the subgraph were constructed at the same time as the members of the cluster are detected, would it run much faster? I realize I'm in over my head, so feel free to let that one go... Dave David Hunkins im: davehunkins 415 336-8965 On Jul 19, 2008, at 2:18 AM, Gabor Csardi wrote:
|
[Prev in Thread] | Current Thread | [Next in Thread] |