dongqifu / vcr-graphormer

VCR-Graphormer: A Mini-batch Graph Transformer via Virtual Connections, ICLR 2024

VCR-Graphormer: A Mini-batch Graph Transformer via Virtual Connections, ICLR 2024

https://arxiv.org/pdf/2403.16030

How to run

1. Place datasets into "dataset" folder

   • Reddit, Aminer, and Amazon2M are here, they can also be traced from the original source here.
   • Squirrel and Actor are here, they can also be traced from the DGL.
   • Other datasets are built in the library and can be directly accessed through our code.

2. Commands are in "demo.txt"

3. Dependencies are in "environment.yml". Major libraries are below.

   • dgl == 0.9.1
   • torch == 2.0.1
   • torch-geometric == 2.3.1
   • torch-scatter == 2.1.1
   • torch-sparse == 0.6.17

Codebase Acknowledgment

1. NAGphormer: A Tokenized Graph Transformer for Node Classification in Large Graphs
2. Influence-Based Mini-Batching for Graph Neural Networks
3. GRAND+: Scalable Graph Random Neural Networks
4. Large Scale Learning on Non-Homophilous Graphs: New Benchmarks and Strong Simple Methods

vcr-graphormer's Issues

When will the code be open source?

Some detailed questions

Hello, your work is interesting and inspires me a lot！
But recently when running your code in this project, I am a bit confused.
When I directly execute commands you provided in demo.txt, there is a decrease in performance compared to the results in your paper, especially in the PubMed, CoraFull, and Computer datasets.
May I ask if there are some other training tricks, or can you provide commands with better hyperparameters in demo.txt for achieving performance in your paper? Thanks.