Ah I see, thank you Szabolcs!
I am using several clustering methods available in igraph at once to compare outputs, so this is completely inconsistent between methods!!!
cluster_edge_betweenness - uses NULL to omit edge weights cluster_fast_greedy - uses NULL to omit edge weights (though it is not clear on this point?) cluster_label_prop - uses NA, but in the ‘usage’ states NULL cluster_leading_eigen - uses NA, but in the ‘usage’ states NULL cluster_louvain - uses NA, but in the ‘usage’ states NULL cluster_optimal - uses NA, but in the ‘usage’ states NULL cluster_walktrap - does not provide any guidance, I assumed NULL was used cluster_spinglass - uses NA, but usage states NULL
I am still quite new to R so perhaps the ‘usage’ is meant to be read differently, but I thought it meant NULL could be used as a meaningful input
Please could this be fixed so there is a uniform and clear approach in the next update to igraph, as it is very confusing at present
Thanks,
Edmund
On 25 May 2017 at 19:43, Edmund Hunt <address@hidden> wrote:Hi Gabor,
Thanks for your reply.
Here are 4 different commands and their result, I guess I am just a bit
confused how they relate to each other.
The first two are using the cluster_leading_eigen alone, the second two use
that command to find the communities and then the modularity function to get
the modularity value out of it
Would I be right in understanding that cluster_leading_eigen only uses the
weights argument after the communities have been found - but then why does
it return the same value below for the first two commands - and why is it
different to the third command
Thanks
cluster_leading_eigen(net, weights = E(net)$weight)
IGRAPH clustering leading eigenvector, groups: 2, mod: 0.055
+ groups:
$`1`
[1] "YV" "B" "P"
$`2`
[1] "DG" "V"
cluster_leading_eigen(net, weights = NULL)
IGRAPH clustering leading eigenvector, groups: 2, mod: 0.055
+ groups:
$`1`
[1] "YV" "B" "P"
$`2`
[1] "DG" "V"
According to the documentation, you need to supply weights=NA, and notweights=NULL, to ignore any existing weight values in the graph.
modularity(net,membership(cluster_leading_eigen(net, weights =
E(net)$weight)),weights=NULL)
[1] 0.03061224
modularity(net,membership(cluster_leading_eigen(net, weights =
E(net)$weight)),weights=E(net)$weight)
[1] 0.0546875
On 25 May 2017, at 06:51, Gábor Csárdi <address@hidden> wrote:
IIRC the original algorithm can be extended easily to take weights
into account.
If you think the igraph is not doing that (and the docs say that it
would), can you please provide a small example that gives you the same
results with or without (large enough) weights? Thanks.
Gabor
On Wed, May 24, 2017 at 10:11 AM, Edmund Hunt <address@hidden> wrote:
Hello,
I have a question/comment about the leading.eigenvector.community function
in igraph
It has an argument for weights, but this seems to make no difference to the
calculated clusters/resulting modularity
Indeed I don’t think Newman’s algorithm takes edge weights into account?
Is it the case that the weights are only used after the community detection
has taken place, to calculate a modularity value? Is it appropriate to use
the weights to calculate modularity, can anyone advise me what is the
‘right’ thing to do with a weighted, undirected network - is it definitely
to use the weights in the modularity calculation, or is there a free choice
Perhaps these issues could be made clearer in the function help
Thanks
_______________________________________________
igraph-help mailing list
address@hidden
https://lists.nongnu.org/mailman/listinfo/igraph-help
_______________________________________________
igraph-help mailing list
address@hidden
https://lists.nongnu.org/mailman/listinfo/igraph-help
_______________________________________________
igraph-help mailing list
address@hidden
https://lists.nongnu.org/mailman/listinfo/igraph-help
_______________________________________________igraph-help mailing listaddress@hiddenhttps://lists.nongnu.org/mailman/listinfo/igraph-help
|