search

THUDM / GRAND

Source code and dataset of the NeurIPS 2020 paper "Graph Random Neural Network for Semi-Supervised Learning on Graphs"
MIT License
208 stars 40 forks source link

The datasets are not convincing enough #9

Open nixror opened 3 years ago

nixror commented 3 years ago

I wonder if you can add an OGB experiment result? I tried to run GRAND on the ogbn-arXiv dataset, but the result was very bad.

wzfhaha commented 3 years ago

I wonder if you can add an OGB experiment result? I tried to run GRAND on the ogbn-arXiv dataset, but the result was very bad.

Thanks for your interests! We don't use OGB because GRAND is mainly for semi-supervised node classification, where labeled nodes are limited. The ogbn-arXiv has enough labels, in that setting consistency regularization might be not useful and even degrade the performance. You can try to remove consistency loss when running OGB, and to see if the results will become better.