RE: How to use NormalizedMutualInformation

[Ilias Trichopoulos <ilias.trichopoulos AT cern.ch>]

Hello Guillaume,

Thanks a lot for your enlightening help! I managed to make it work!
In GraphStream's media directory, in CSSS2012 
(http://graphstream-project.org/media/other/CSSS2012/) I found some small 
datasets in the GraphStreamWorkspace.zip like the karate.gml, dolphins.gml 
etc. Now I am looking for the ground truth of these examples in the 
literature, so I can test my algorithm and see what NMI value I get, but 
didn't find anything so far. Do you have something in mind that could help?

Thanks a lot!

Ilias.

-----Original Message-----
From: 
guillaume-jean AT herbiet.net
 
[mailto:guillaume-jean AT herbiet.net]
 
Sent: 05 February 2014 00:15
To: 
graphstream-users AT litislab.fr;
 Ilias Trichopoulos
Subject: Re: How to use NormalizedMutualInformation

Hi Ilias,

Sorry for the late reply. I'm pleased to see my modest contribution to 
GraphStream and community detection used in such a great institution as the 
CERN!

NormalizedMutualInformation extends the CommunityRelativeMeasure class, that 
is it computes and updates a relative measure based on the comparison between 
the current community assignment and a reference assignment on a given graph.

You therefore need to have two community attributes declared in your nodes 
(e.g. referenceCommunity(string) and community(string)).
referenceCommunity(string) is likely static and corresponds to the community 
assignment found in the literature, community(string) represents the dynamic 
community assignment computed by your algorithm.

Then you can have the NMI measure respective to the referenceCommunity 
assignment and the current community assignment as follows:

NormalizedMutualInformation NMI = new
NormalizedMutualInformation("community","referenceCommunity");
NMI.init(graph);
NMI.getMeasure();

Hope this helps !

-Guillaume