Size-Invariant Graph Representations for Graph Classification Extrapolations

This repository contains the official code of the paper Size-Invariant Graph Representations for Graph Classification Extrapolations (ICML 2021 Long Talk).

Manual dependencies (CUDA)

• PyTorch 1.7.1
• torch-cluster 1.5.8
• torch-geometric 1.6.3
• torch-scatter 2.0.5
• torch-sparse 0.6.8
• torch-spline-conv 1.2.0
• torch-geometric 1.6.3
• ray[tune] 1.1.0

Install the additional dependencies as follows:

$ pip install -r requirements.txt

Download Data

Please, run the following commands to download and set up the data folder.

$ wget https://www.dropbox.com/s/38eg3twe4dd1hbt/data.zip
$ unzip data.zip

The command above will place the data already sampled in the folder data/. Please specify its absolute path in base_config.yaml.

Hypertune

The provided configurations allow you to run the hyperparameter tuning of $\Gamma_\text{GIN}$ on NCI1.

To tune for other datasets and/or models:

• In hyper_config.yaml specify the hyperparameters values. For details on the range of the hyperparameter refer to the Appendix.
• In base_config.yaml set dataset_name to NCI1, NCI109, PROTEINS or brain-net (i.e. schizophrenia).
• In base_config.yaml set the model to KaryGNN (i.e. $\Gamma_\text{GNN}$), KaryRPGNN (i.e. $\Gamma_\text{RPGNN}$), GraphletCounting (i.e. $\Gamma_\text{1-hot}$), GNN or RPGNN. You can specify the GNN in gnn_type as pna, gcn or gin and the XU-READOUT in graph_pooling as mean, max or sum.

Run

$ python hypertuning.py

Run a single configuration

The provided configurations allow you to run $\Gamma_\text{GNN}$ on NCI1 with the best hyperparameters.

To run for other datasets and/or models specify the parameters in base_config.yaml.

Run

$ python lightning_modules.py

Credits

If you use this code, please cite

@inproceedings{bevilacqua2021size,
  title={Size-invariant graph representations for graph classification extrapolations},
  author={Bevilacqua, Beatrice and Zhou, Yangze and Ribeiro, Bruno},
  booktitle={International Conference on Machine Learning},
  pages={837--851},
  year={2021},
  organization={PMLR}
}