-
Notifications
You must be signed in to change notification settings - Fork 212
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
Cluster force #675
base: refactor_data
Are you sure you want to change the base?
Cluster force #675
Conversation
Is there a script to process an xyz to find clusters and annotate the file? |
Will add annotation script from Ioan. Another improvement to current implementation:
|
Make sure the script you add generates unique IDs, and then when reading the file you can check it. Why does the ID need to be unique across the data set rather than just within the config? You could generate a unique id by concatenating the config ID and the cluster ID and this would then be unique irrespective of the cluster ID uniqueness |
Yes exactly, thats the to do ("double batched"). I'll edit it tomorrow. |
Looks good. Could you add a small test for the loss + one in the run_train test? |
The clustering scripts are taken from Ioan Magdau's repo: https://github.com/imagdau/aseMolec/blob/main/aseMolec/anaAtoms.py |
@LarsSchaaf I think it would be nice to add somme config_cluster_weight, inside the loss, as you already added them to the AtomicData.
And then do a bunch of scatter etc to create a mask per cluster and mask out clusters that should not be weighted. |
I don't know if it's currently possible or not, but it would be nice if not all atoms needed to be part of a cluster, just some. |
Fitting to cluster forces for better intermolecular interactions.
Description
Atoms are assumed to be separated into clusters, eg molecules. Each cluster has a unique id (which is passed to cluster_key). The cluster force is all forces belonging to a cluster summed together. The additional loss term fits the predicted to ground truth cluster force.
Possible extensions: