search

DeepGraphLearning / GearNet

GearNet and Geometric Pretraining Methods for Protein Structure Representation Learning, ICLR'2023 (https://arxiv.org/abs/2203.06125)
MIT License
253 stars 28 forks source link

Non-deterministic embeddings #43

Closed andrew7shen closed 11 months ago

andrew7shen commented 11 months ago

Hi!

I was wondering if there is any reason that the GearNetIEConv encoder would return variable embeddings for the same input file. I encountered this using my own data, but when I set a torch manual_seed, the embeddings became constant for the same input. And is this expected to have any effect on model performance?

Thanks for your help!

Oxer11 commented 11 months ago

Hi, we have set the random seed for torch and numpy in the downstream.py. There is probably some randomness in torch_scatter, but it should not have large influence on the final embeddings. https://github.com/DeepGraphLearning/GearNet/blob/7873e2e594234ab581a1119c6ce2f09800593e0e/script/downstream.py#L56-L66

andrew7shen commented 11 months ago

Gotcha, thanks!