Implementation of the paper "Variational Inference for Graph Convolutional Networks in the Absence of Graph Data and Adversarial Settings", accepted for publication at NeurIPS 2020. The paper will be publicly available from the NeurIPs website but a preprint can be found in arXiv.
This code has been tested in Python 3.6.8. Once you have a copy of the source, you can navigate to the root directory and create a new virtual environment (recommended):
$ virtualenv ~/VGCN-env --python=python3.6
Then you can activate your environment:
$ source ~/VGCN-env/bin/activate
and install requirements with:
$ pip install -r requirements.txt
There are two scripts that will run examples of the experiments described in the paper.
To run an example of the no-graph case by building a prior based on a KNNG:
From your working directory, run
$ ./experiments/example_nograph.sh
This runs an example of the no-graph case on the CiteSeer dataset and saves the results in the local directory ./results/test_knn.
To run an example of the adversarial case on a graph corrupted by adversarial noise:
From your working directory, run
$ ./experiments/example_adversarial.sh
This runs an example of the adversarial setting on the CiteSeer dataset for a specific adversarially-corrupted graph and saves the results in the local directory ./results/test_adversarial
Pantelis Elinas, Edwin V. Bonilla and Louis Tiao.
This project is licensed under the MIT License.