Louvain algorithm and its multi-level nature

[anastasia...@gmail.com]

Dear all,

I have a question regarding the resolution problem in modularity clustering. In the original paper by Blondel, Guillaume and Lambiot (2008), the authors claim that "the so-called resolution limit problem of modularity also seems to be circumvented thanks to the intrinsic multi-level nature of our algorithm" (p. 5). As I understand it, the communities created in each pass are meant to be accessible thus avoiding the loss of smaller communities within larger ones.

However, in visone the output is not hierarchical, and I assume that it demonstrates the outcomes of the last pass. This would mean that for this particular implementation resolution is an issue - is that correct?

(I implemented the algorithm with default node order without group nodes, with uniform initial clusters and uniform edge weight.)

All the best,

Anastasia

[Müller Julian]

Dear Anastasia,

 

Yes, visone only outputs the partition in the Louvain clustering hierarchy that has the largest associated modularity. Thus, the result is subject to the resolution limit, as this a property of modularity maximization in general.

 

Generally, we could produce all levels of the hierarchy, and I think visone 2.10 to 2.17 actually did output them as additional node attributes.

However, these versions are five years old by now and it seems some bug in the Louvain clustering algorithm was fixed in 2.18, so they might no longer run well or yield incorrect results.

 

Best,

Julian