This repository is an ongoing implementation of Combinatorial optimization with physics-inspired graph neural networks done with PyTorch Lightning and PyTorch Geometric.

The current version implements only the Maxcut and Maximum Independent Set (MIS) problems on random d-regular graphs, as explained in the paper. Please beware that the results are still inconclusive and this is an ongoing implementation, your feedback is more than welcomed.

The file structure is quite simple and straightfoward. main.py is the file used to run the experiments, you can find the various arguments controlling the runs inside. data.py generates a given number of random d-regular graphs, models.py contains the GNN model with various possible architectures, and util.py contains the code to compute the Hamiltonians for each problem. As a simple example, you can run the following commands to check that the code runs:

python main.py --maxcut --epochs 3 # maxcut
python main.py --epochs 3 # maxcut