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 include 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 include IMDB-BINARY
, IMDB-MULTI
, PROTEINS
, COLLAB
, DD
, NCI1
If you find this work helpful to your research, please consider citing our paper:
@inproceedings{cheng2023wiener,
title={Wiener graph deconvolutional network improves graph self-supervised learning},
author={Cheng, Jiashun and Li, Man and Li, Jia and Tsung, Fugee},
booktitle={Proceedings of the AAAI conference on artificial intelligence},
volume={37},
number={6},
pages={7131--7139},
year={2023}
}