modularity() results differ when called on communities vs with graph + membership

[ashander]

This is similar to an old question on SO but not due to weights.

The two different ways given in ?modularity give different results yet based on the answer to the SO question I don't think this should be the case.

The two different ways to compute modularity given in the help are:

    wtc <- cluster_walktrap(g)
    modularity(wtc)
    # and
    modularity(g, membership(wtc))

For this example these give different results

g2 <- make_graph(c(1,2,2,3,3,1), directed = FALSE)

set.seed(444)
wtc2 <- cluster_walktrap(g2)
modularity(wtc2)
## 0
wtc2
##IGRAPH clustering walktrap, groups: 3, mod: 0
##+ groups:
## $`1`
##  [1] 1
## 
##  $`2`
## [1] 2
##  
##  $`3`
##  [1] 3
modularity(g2, membership(wtc2))
## -0.3333

This is with igraph_1.2.5.

Note

This shows up on much more complicated graphs as well -- this is the simplest case where I could reproduce

[szhorvat]

This looks like an issue with the Walktrap code in that the returned community structure is not the one with the highest modularity.

Also, Walktrap returns a sequence of modularity values, one for each merge step of the dendrogram. The one for entirely separate vertices is 0, not -0.3333.

Moving to the C core repo, since this is not an R-specific bug.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 14 days if no further activity occurs. Thank you for your contributions.