igraph 中的模块化：2021 年 10 月前后

Question

In igraph when I cluster the karate.network ( kar ) with, say, cluster_fast_greedy , to give karfg , using modularity(karfg) gives 0.4345215. The deprecated option of including membership
modularity(kar, membership(karfg)) gives 0.399096. Is there an explanation for this discrepancy? (It is not fastgreedy specific, other community detection methods also give a discrepancy).

library(igraph)
library(igraphdata)
data(karate)
kar <- karate
karfg<-cluster_fast_greedy(karate)

modularity(kar, membership(karfg))
0.3990796
modularity(karfg)
0.4345215

Answer 1

This happens because the modularity function does not use edge weights by default, while most community detection functions do. The graph you are working with has a weight edge attribute, which will be used by cluster_fast_greedy() unless you explicitly prevent that.

This is how you can use it in modularity as well:

modularity(karate, membership(karfg), weights=E(karate)$weight)

0.4345215

---

Note that not all of igraph's community detection functions make use of weights, and those that do may not use it in the same way. Thus, when you call modularity explicitly, and separately from the community detection function, be careful to specify weights that match with how the community detection was done.