Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

At 950 epochs, the best results on the zinc dataset are still more than 1.4, for a total of just 1,000 epochs #1

Open
QSNZNJING opened this issue Oct 16, 2023 · 1 comment

Comments

@QSNZNJING
Copy link

The experiment seems to have lost some code files,such as pygdataclass.py ...And what is returned in the SpecformerZINC model is a tuple with a different y size. But the above are not the most important, I do not know what is the problem, the code in zinc data set about 950 rounds, the effect is not improved at all, could you please provide a complete code?

@bdy9527
Copy link
Collaborator

bdy9527 commented Feb 23, 2024

Hi, there. Sorry that I mis-merged the loss functions of ZINC, PCBA, and HIV.

You can replace the following code

y_idx = y == y
loss = loss_fn(logits.to(torch.float32)[y_idx], y.to(torch.float32)[y_idx])

with

loss = loss_fn(logits.squeeze(),y)

for the ZINC dataset. And you will see the correct results.

Here is part of my training log:

image

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants