Add MACE-POLAR-1 pretrained models - #164
Conversation
| if torch.abs(volume) > 0: | ||
| rcell = 2 * torch.pi * torch.linalg.inv(cellTensor.mT) | ||
| else: | ||
| rcell = torch.zeros((3, 3), dtype=dtype, device=ptr.device) |
There was a problem hiding this comment.
This is more complicated than it needs to be. Because of the requirements we have for box vectors, the box volume is just the product of the diagonal elements and is guaranteed to be positive.
There was a problem hiding this comment.
This was copied from their implementation which was copied from inside the MACE library. They probably can't make the same guarantees we can about the box vectors. I'll update it.
| has_graph_longrange = False | ||
| if has_graph_longrange: |
There was a problem hiding this comment.
You could get rid of the has_graph_longrange variable and just put the calculations inside the try block.
|
I also want to ask them about whether the |
2 participants
Adds MACE-POLAR-1 pretrained MACE models and fixes #143.
Like MACELES, requires an extra dependency to be installed from a repository (
git+https://github.com/WillBaldwin0/graph_electrostatics.git@v0.4.0) as documented; this is a versioned release though.