lsj2408 / graphnorm Goto Github PK

View Code? Open in Web Editor NEW

100.0 3.0 22.0 1.58 MB

[ICML 2021] GraphNorm: A Principled Approach to Accelerating Graph Neural Network Training (official implementation)

License: MIT License

Python 91.58% Shell 8.42%

graph-neural-networks normalization

graphnorm's People

Contributors

Stargazers

Watchers

graphnorm's Issues

which one is the best setting for ogb_molhiv dataset

there are two hyper parameter setting for ogb_molhiv, one is at the outer-most README, the other is at GraphNorm/GraphNorm_ws/ogbg_ws/scripts/Example-gcn-test-performance/seed-1-5-gcn_run.sh. I have tried both, but none of them can obtain reported result

Results for ogbg-molpcba

Hi, thanks for open-sourcing the code and releasing extensive experiments! GNN-specific normalization beyond BatchNorm and LayerNorm was definitely a missing component in the toolbox.

Did you experiment with GraphNorm on ogbg-molpcba (the larger and more challenging mol dataset from OGB) as well as the other graph classification tasks? Would love to know your experience or informal results.

WHy

num_features

Hello, thank you very much for your work. I noticed num_features in your code. I want to know what it means and how it is set in use.Thank you for your reply.

Recommend Projects