-
Notifications
You must be signed in to change notification settings - Fork 69
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
Global and Local structure #16
Comments
Hi, Two encoders tend to encode real bonds and virtual bonds within a threshold respectively, which in practice helps the performance. |
Hi Minkai, I am trying to reproduce the results of your GeoDiff paper. However, to help understand the code better, I am looking at the losses; I see 3 losses calculated using an objective close to the RMSD, What is the difference between the local and global loss. I see they are both based on some:
Following the local edge, max I find it is based on whether an edge type is local or not. local_edge_mask = is_local_edge(edge_type) # (E, ) def is_local_edge(edge_type): Now this does not make much sense. I don’t know what edge types are > 0 ?? Could you explain something about the edge_type and the local_edge_mask that either makes them local or global e.g. what are the different edge types? And what numbers are they given as per the code. IN THE train.py SCRIPT I see we load the batch to cuda using the torch_geometric.data Dataset, Your help is highly appreciated. |
Hi @mulangonando, Thanks for your interest! Pos and atom_type are atom position and type respectively. Local edges are the bonds existing in molecular graphs, while global edges refer to those added when two atoms are close enough. Global edges are labeled as type 0. I recommend that you can have a closer look at the module at Lines 231 to 254 in c6f26dc
|
Hi Minkai Xu,
What's the motivation for that you designed two separate architectures to learn local and global structures?
In loss, the loss is divided into local loss and global loss
(node_eq_global - target_pos_global)**2 + (node_eq_local - target_pos_local)**2
The text was updated successfully, but these errors were encountered: