This repository holds an implementation of Automorphic Equivalence-aware Graph Neural Network NeurIPS 2021.
This implementation depends on the following enviornment and packages:
python == 2.7
pytorch == 1.4
numpy == 1.17.2
scipy == 1.6.3
sklearn ==0.24.1
h5py == 2.9.0
GPUtil ==1.4.0
setproctitle == 1.1.10
To train and evaluate the model(s) in the paper, run this command:
python grape_model.py --data='cite|cora' --gpu='0' --lr=0.003 --wd=0.00003 --dropout=0.5 --hid=32
Our model achieves the following classification accuracy:
| Hamilton | Lehigh | Rochester | JHU | Amherst | Cora | Citeseer | Amazon | |
|---|---|---|---|---|---|---|---|---|
| GRAPE | 28.1% | 27.3% | 25.0% | 34.6% | 32.6% | 87.1% | 74.6% | 58.6% |