chao1224/GraphMVP

Why the classification task and the regression task have different model checkpoint?

Closed this issue · 2 comments

StefanIsSmart commented

And what is the relationship with GraphMVP, GraphMVP-G, GraphMVP-C?

StefanIsSmart commented

By the way, maybe the ogb version has been updated.

size mismatch for atom_encoder.atom_embedding_list.1.weight: copying a param with shape torch.Size([4, 300]) from checkpoint, the shape in current model is torch.Size([5, 300]).

The dimension of the pre-trained model is not correct

chao1224 commented

Hi,

  • These are two variants: C for contrastive and G for generative. You can check Eq 8 in the paper.
  • Yes, the checkpoints are using the older version of OGB. You can also check #23