Overlapping Community Detection with Graph Neural Networks

May 28, 2020 ยท View on GitHub

Pytorch implementation of the Deep Variational Generative Overlapping Community Detection

Usage

The main algorithm and other utilities(metrics, evaluation .etc) are implemented in the embedding_cd package that can be installed as

python setup.py install

The main analysis and visualization code are in Jupyter notebooks

run pip install -r requirements.txt first.

Requirements

numpy>=1.16.4
pytorch>=1.2.0
scipy>=1.3.1
dgl>=0.42
...

Some modifications in DGL version:

  1. append F.relu after logits
  2. BatchNorm after GCNLayey
  3. nocd.utils.l2_reg_loss(model, scale=args.weight_decay) rather than setting weight_decay in Adam
  4. dropout -> Relu -> BN