Implementation of the paper "Community Detection with Graph Neural Networks", by J. Bruna and L. Li https://arxiv.org/abs/1705.08415.
The code is based on Lua Torch. See here for installation and basic tutorials.
The experiments on the Stochastic Block Model can be run with
th -i sbm.lua [options]
An example with fixed probabilities (a=5, b=1), using adam as optimizer and using 30 gnn layers:
th -i sbm.lua -mixture 0 -p 5 -q 1 -path /scratch/sbmp5q1/ -optmethod adam -gpunum 2 -layers 30
The experiments on real-world community detection are based on the Snap graphs with ground-truth community; see Stanford Network Analysis Project for more details. An example of running the model on such data is
th -i snap.lua -gpunum 1 -graph 'youtube'