ehoogeboom / e3_diffusion_for_molecules

MIT License
432 stars 113 forks source link

Explicit valence error in training #32

Closed lajictw closed 10 months ago

lajictw commented 10 months ago

Hi! I met a problem in training. The codes works super fine in the 0th epoch. But when it came to the 1st epoch, it pumped out explicit error for atoms. Two screenshots of the run have been attached below. Thanks for your help! image image

ehoogeboom commented 10 months ago

It's been a while, but I think this is just something in eval that's part of the SMILES conversion. SMILES validity is kind of a tricky benchmark anyway, because it solves a lot of things that are incorrect.