search

PetarV- / GAT

Graph Attention Networks (https://arxiv.org/abs/1710.10903)
https://petar-v.com/GAT/
MIT License
3.18k stars 642 forks source link

Pubmed Dataset Accuracy #21

Closed mufeili closed 5 years ago

mufeili commented 5 years ago

Hi,

I've tried running your code on the Pubmed dataset using the configuration you mentioned in the paper. With 100 runs, the accuracy is about 77.61 +- 0.78 while the paper reports 79.0 +- 0.3. I'm wondering if you have done some additional modifications as in the issue about Citeseer accuracy and it will be great if you can provide these details. Thanks a lot!

PetarV- commented 5 years ago

Hi Mufei,

You are correct indeed -- please see #12 . I should also remark (as you can see in its discussion) that for such a small training dataset, an entirely simpler attention mechanism may be more desirable.

Thanks, Petar

mufeili commented 5 years ago

Thanks Petar!