Skip to content

rmongeca/gnn

Repository files navigation

Graph Neural Networks

This repository contains the code support for the Master thesis for the Master in Innovation & Research in Informatics - Data science of BarcelonaTech University, Faculty of Informatics.

The purpose of the repository is two-fold. On the one hand, it provides an implementation of the gnn Python module to create Graph Neural Networks (GNN) using Tensorflow 2, based on the Message Passing Neural Network framework on [1].

On the other hand it provides two examples of graph neural networks using said gnn module as well as the equivalent implementation using the iGNNition library, which provides a fast way for prototyping GNN without the need for expert implementations.

The example models are the following:

  • Quantum Chemistry QM9 dataset molecules' properties prediction, see [1].

  • Graph Neural Networks for Scalable Radio Resource Management, see [2].

Environemnt setup

In order to set-up the repository environment, Python 3.8 and Conda are needed. Refer to you OS specific instructions.

Next, we need to restore the Conda environment, gnn, to isolate the Python dependencies of the project:

conda env create -f environment.yml

Whenever we open a new terminal, we will need to activate this environment with the following command:

conda activate gnn

Next, we need to install the iGNNition library. The repository contains a Wheel file with the 1.0.2 version of the framework, to provide reproducibility. To install, activate the environment and run:

pip install ignnition/ignnition-1.0.2-py3-none-any.whl

Alternatively, one could directly install the latest version on PyPi using:

pip install ignnition

Note that the latest changes in the iGNNition framework may break some of the implementations of this repository, which are tested for version 1.0.2.

References

  1. Gilmer, J., Schoenholz, S.S., Riley, P.F., Vinyals, O. & Dahl, G.E.. (2017). Neural Message Passing for Quantum Chemistry. Proceedings of the 34th International Conference on Machine Learning, in Proceedings of Machine Learning Research 70:1263-1272 Available here.

  2. Yifei Shen, Yuanming Shi, Jun Zhang, Khaled B. Letaief: Graph Neural Networks for Scalable Radio Resource Management: Architecture Design and Theoretical Analysis. CoRR abs/2007.07632 (2020). Available here.