Results not reproducible with rleiden.community

[SABiagini]

Hi,

I've been using the rleiden.community() function for clustering detection. However, I've noticed that when I run the same analysis multiple times, results tend to change. I also tried to set a seed, but even with the same seed results change.

Do you have any suggestion on how to prevent this behavior and make results reproducible?

Thank you.

[evanbiederstedt]

Hi @SABiagini

I need much more information to be helpful here.

Could you show me the code? What are you running and are the results changing?

Thanks, Evan

[SABiagini]

Hi Evan,

my code is pretty simple:

data<-fread(5cM.coef)
edges <- as.data.frame(data)
colnames(edges) <- c("from", "to", "weight")
g <- graph_from_data_frame(edges, directed = FALSE)
leiden<-rleiden.community(g, max.depth = 4, min.community.size = 100)

write.table(leiden$membership,"Leiden_Communities.txt",sep="\t",row.names = T, col.names = F, quote = FALSE)

The input is a three column file (ID1 ID2 weight)

In the list I save at the end of the code, the community assignment to each individual change any time I rerun the code.

Thank you.

Best,

Simone

[evanbiederstedt]

Hi Simone @SABiagini

I'll need to reproduce this myself. I don't have access to 5cM.coef. Please put the outputs of what you see as well, i.e. attach this table in this GitHub issue.

So yes, please give me a reproducible example I can run.

Thanks, Evan

[SABiagini]

Hi Evan,

The file is 10Mb and it cannot be copy/pasted here.

Also, I'm sorry but I can't publicly share that file. Is it possible to reach you in a different way?

Thank you.

Best,

Simone

[evanbiederstedt]

Email me if you could.

But please try to use some publicly available R data as well.

https://github.com/kharchenkolab/leidenAlg/blob/main/tests/testthat/test_functions.R

It would be best if you could replicate this behavior with our unit tests using the data I saved in the package here. If you could show me the behavior here, that would be helpful and save time.

Thanks, Evan

[evanbiederstedt]

Could you try installing the package from this branch? https://github.com/kharchenkolab/leidenAlg/tree/no_cpp

Please check if the problem remains. It looks like the results become stable.

Thanks, Evan

[evanbiederstedt]

@SABiagini

It looks like this existed with previous versions of leidenAlg and igraph (i.e. igraph 1.2.6 from 2020 and leidenAlg 1.0.0)

I'm using this code:

library(data.table)
foobar = '/home/evanb/leidenAlg-1.0.0/5cM.coef' 

library(leidenAlg)

data<-data.table::fread(foobar)

edges <- as.data.frame(data)
colnames(edges) <- c("from", "to", "weight")
g <- igraph::graph_from_data_frame(edges, directed = FALSE)
leiden<-rleiden.community(g, max.depth = 4, min.community.size = 100)

write.table(leiden$membership,"Leiden_CommunitiesC.txt",sep="\t",row.names = T, col.names = F, quote = FALSE)

Are you able to see reproducible clusters using igraph::cluster_leiden()? If so, we could try using this.

@VPetukhov mentioned once in another thread there was some odd behavior, but I think this was fixed with explicitly setting a parameter + fiddling with the resolution parameter.

Best, Evan