This is the code for our ICLR 2023 paper NAGphormer: A Tokenized Graph Transformer for Node Classification in Large Graphs.
Python == 3.8
Pytorch == 1.11
dgl == 0.9
CUDA == 10.2
You can run each command in "commands.txt".
You could change the hyper-parameters of NAGphormer if necessary.
Due to the space limitation, we only provide several small datasets in the "dataset" folder.
For small-scale datasets, you can download them from https://docs.dgl.ai/tutorials/blitz/index.html.
For large-scale datasets, you can download them from https://github.com/wzfhaha/GRAND-plus.
If you find this code useful, please consider citing the original work by authors:
@inproceedings{chennagphormer,
title={NAGphormer: A Tokenized Graph Transformer for Node Classification in Large Graphs},
author={Chen, Jinsong and Gao, Kaiyuan and Li, Gaichao and He, Kun},
booktitle={Proceedings of the International Conference on Learning Representations},
year={2023}
}