Deep Graph Infomax

Deep Graph Infomax (DGI) is an unsupervised algorithm for finding representations of graphs that can be used in downstream tasks like node classification.

This is a TensorFlow implementation of DGI, based on the Graph Convolutional Network implementation by Thomas Kipf.

Installation

python setup.py install

Requirements

• tensorflow (>0.12)
• networkx

Run

First train a DGI model:

python train.py --model dgi

Once the model is trained, the graph embeddings are saved as a pickle file in the runs folder. Take note of its path (e.g. runs/2018-11-04-164053/embeddings.p and use it to train a logistic regression model on the node classification task:

python train.py --model logreg --embeddings_path runs/2018-11-04-164053/embeddings.p

Data

In order to use your own data, you have to provide

• an N by N adjacency matrix (N is the number of nodes),
• an N by D feature matrix (D is the number of features per node), and
• an N by E binary label matrix (E is the number of classes).

Have a look at the load_data() function in utils.py for an example.

In this example, we load citation network data (Cora, Citeseer or Pubmed). The original datasets can be found here: http://linqs.cs.umd.edu/projects/projects/lbc/. 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

(or by editing train.py)

Models

You can choose between the following models:

• dgi: Deep Graph Infomax (Petar Velicković et al., Deep Graph Infomax, 2018)
• 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)
• dense: Basic multi-layer perceptron that supports sparse inputs