This is the official implementation of the following paper:
Wiener Graph Deconvolutional Network Improves Self-Supervised Learning
Jiashun Cheng, Man Li, Jia Li, Fugee Tsung; AAAI 2023
- python >= 3.7
- torch >= 1.11.0
- torch_geometric >= 2.0.3
- ogb >= 1.3.4
- argparse >= 1.1.0
- numpy >= 1.12.2
- scikit_learn >= 1.0.2
- scipy >= 1.4.1
To reproduce the reported results, please run the script with --use_cfg.
Node classification
# With best configurations
python main_node.py --dataset PubMed --use_cfg
# Or you can customize the configurations (such as propagation kernel, decoder aggregation and etc.)
python main_node.py --dataset PubMed --kernel heat --dec_aggr sum
Supported datasets includes Cora, CiteSeer, PubMed, CS, Physics, Computers, Photo, ogbn-arxiv
Graph classification
# With best configurations
python main_graph.py --dataset IMDB_BINARY --use_cfg --seed 2 12 22 32 42
# Or you can customize the configurations (such as propagation kernel, decoder aggregation, pooler and etc.)
python main_node.py --dataset IMDB_BINARY --kernel heat --dec_aggr sum --pooler max --seed 2 12 22 32 42
Supported datasets includes IMDB-BINARY, IMDB-MULTI, PROTEINS, COLLAB, DD, NCI1
If you find this work is helpful to your research, please consider citing our paper:
@article{cheng2022latent,
title={Latent Augmentation For Better Graph Self-Supervised Learning},
author={Cheng, Jiashun and Li, Man and Li, Jia and Tsung, Fugee},
journal={arXiv preprint arXiv:2206.12933},
year={2022}
}