Here we provide the implementation of GND-Nets (Graph Neural Diffusion Networks) in TensorFlow. The repository is organised as follows:

• data/ contains datasets Cora, Cora-ML, Citeseer, Pubmed, Amazon Computers, and Amazon Photo;
• new_data/ contains datasets Chameleon and Squirrel;
• models/ contains the implementation of the GND-Nets (gndnets_slp.py, gndnets_mlp.py, and gndnets_ds.py);
• utils/ contains:
  • an implementation of three variants of graph neural diffusions (layers.py);
  • preprocessing subroutines (process.py);

Finally, bash run_train execute the experiments.

The script has been tested running under Python 3.7.9, with TensorFlow version as:

• tensorflow==2.6.0

In addition, CUDA 11.4 and cuDNN 11.1 have been used.

MIT