qslim/soft-mask-gnn

PythonMIT

Soft-mask: Adaptive Substructure Extractions for Graph Neural Networks

This is the code of the paper "Soft-mask: Adaptive Substructure Extractions for Graph Neural Networks".

Prerequisites

The code is built upon PyTorch Geometric.

The following packages need to be installed:

python 3.7.3
pytorch 1.2
pytorch-geometric 1.3.0
Additional modules: numpy, easydict

Running the tests

Set hyper-parameters in configs.
Run tests.
- To run 10-fold cross-validation on MUTAG, run the classification_main.py script:
```
python classification_main.py --config=configs/MUTAG.json
```
  To run other classification benchmarks, replace MUTAG.json with corresponding configuration files.
- To run the QM9 test, run the regression_main.py script:
```
python regression_main.py --config=configs/QM9.json
```

Reference

@inproceedings{yang2021softmask,
	author = {Yang, Mingqi and Shen, Yanming and Qi, Heng and Yin, Baocai},
	title = {Soft-Mask: Adaptive Substructure Extractions for Graph Neural Networks},
	year = {2021},
	url = {https://doi.org/10.1145/3442381.3449929},
	booktitle = {Proceedings of the Web Conference 2021},
	pages = {2058–2068},
	numpages = {11},
	series = {WWW'21}
}

License