alittera/sgc

Implementation of Simplified Graph Convolutional Networks in TensorFlow

Simplified Graph Convolutional Networks

This is a TensorFlow implementation of Simplified Graph Convolutional Networks for the task of (semi-supervised) classification of nodes in a graph carried out as a project for the examination of Neural Networks, at Sapienza university of Rome.

The project is based on:

Thomas N. Kipf, Max Welling, Semi-Supervised Classification with Graph Convolutional Networks

Wu , Zhang ,de Souza Jrm, Simplifying Graph Convolutional Networks

Installation

python setup.py install

Requirements

• tensorflow (>0.12)
• networkx

Run the demo

cd gcn
python train.py -model sgcn -dataset pubmed

Data

In this example, we load citation network data (Cora, Citeseer or Pubmed). The original datasets can be found here: http://www.cs.umd.edu/~sen/lbc-proj/LBC.html. In our version (see data folder) we use dataset splits provided by https://github.com/kimiyoung/planetoid (Zhilin Yang, William W. Cohen, Ruslan Salakhutdinov, Revisiting Semi-Supervised Learning with Graph Embeddings, ICML 2016).

You can specify a dataset as follows:

python train.py --dataset citeseer
python train.py --dataset cora
python train.py --dataset pubmed

(or by editing train.py)

Models

You can choose between the following models:

• gcn: Graph convolutional network (Thomas N. Kipf, Max Welling, Semi-Supervised Classification with Graph Convolutional Networks, 2016)
• gcn_cheby: Chebyshev polynomial version of graph convolutional network as described in (Michaël Defferrard, Xavier Bresson, Pierre Vandergheynst, Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering, NIPS 2016)
• sgcn: Simplifying Graph Convolutional Networks
• dense: Basic multi-layer perceptron that supports sparse inputs