How to reproduce 'consistent' partitions?

[judekurn]

How can I reproduce the same partitions everytime I run best_partition?

I use best_partition to produce community clusters for my research and wrote an entire chapter on the result. But when I restart the notebook, I got a different number of partitions with different memberships. That means, I cannot 'reproduce' the result I have written and my entire chapter is no longer reproducible. How can I recreate the partitions which I found previously? Btw, this problem happens on a graph with 131 nodes, but works fine (i.e. reproducible) for a graph with 51 nodes. Thanks.

[dalwar23]

@judekurn I am also using louvain method for community detection and when I saw this issue I tried to reproduce communities after restarting kernel and It worked fine for me. Every time same communities with same nodes. Please have a look at my louvain ipython notebook You will also find a copy of karate_club data as plain .txt format in the same directory. I hope this helps :+1:

[dalwar23]

@judekurn On another note. If you are using networkx package for creating graphs from edgelist file with nx.read_edgelist please consider putting the parameter nodetype as int that way it will give you consistent graph thus a consistent outcome on community detection.

I removed nodetype form nx.read_edgelist and I was finally see what you were facing. The communities were having different nodes.

[judekurn]

Thank you for your replies, they are very helpful. I am using nx.from_pandas_edgelist method and the nodes are in str format. As you suggested, I retraced my steps back to networkx and found the culprit: Giant = max(nx.connected_component_subgraphs(G), key=len). I use this because I want to apply the Louvain method on the 'giant' connected graph only. When I remove this code, everything works the way it should be. Again, thanks Dalwar for your help.