Count-GNN

We provide the code (in pytorch) and datasets for our paper "Learning to Count Isomorphisms with Graph Neural Networks", which is accepted by AAAI23.

Description

The repository is organised as follows:

• datasets/data/: contains data we use. Need to be decompressed and be placed in the same path as Count_GNN/
• Count_GNN/: contains our model.
• converter/: transform the original dataset into the data format that can be inputted into Count_GNN.
• generator/: generate synthetic dataset.
• Appendix.pdf: Appendix of our paper "Learning to Count Isomorphisms with Graph Neural Networks"

Package Dependencies

• tqdm
• numpy
• pandas
• scipy
• tensorboardX
• torch >= 1.3.0
• dgl == 0.4.3post2

Running experiments

• train model: python _train.py --model EDGEMEAN --predict_net FilmSumPredictNet --emb_dim 4 --ppn_hidden_dim 12 --weight_decay_film 0.0001
• test model: python evaluate.py ../dumps/MUTAG

Citation

@inproceedings{yu2023learning,
title={Learning to Count Isomorphisms with Graph Neural Networks},
author={Yu, Xingtong and Liu, Zemin and Fang, Yuan and Zhang, Xinming},
booktitle={Proceedings of the AAAI conference on artificial intelligence},
year={2023}
}