-
Notifications
You must be signed in to change notification settings - Fork 124
This issue was moved to a discussion.
You can continue the conversation there. Go to discussion →
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鈥檒l occasionally send you account related emails.
Already on GitHub? Sign in to your account
馃専 [FEATURE] Multi-system dataset support & atomic charges/dipoles prediction #89
Comments
Hi @leoil, thanks for the questions, answers to your two questions below.
You can then simply put that extxyz file as the input to our code and everything should work. Note that if you want energies and forces, you will have to be careful on how to set the loss-function weights (we recommend using a weight of 1 on the energies and a weight of N^2 on the forces, where N is average number of atoms in the system). Let us know if you have any other questions.
|
Thanks for your response.
|
Hi @leoil, that sounds good! Regarding your questions:
The networks should then implicitly learn the correct partial charges. Does that make sense? I highly recommend you do this on our |
Yes, that make sense. Thanks for your detailed guide, I'll give it a try. |
This issue was moved to a discussion.
You can continue the conversation there. Go to discussion →
Is your feature request related to a problem? Please describe.
The trained model shows great performance on dataset with single system like MD17. I'm wondering if it also support the training&fitting on dataset with multiple systems.
I removed
npz_fixed_field_keys
in the config file and run a quick test on thesn2-reaction
dataset used in PhysNet, which includes various structures of different molecules related to that reactions. It uses 0 padding to represent those molecular with fewer atoms. Here is the result.Also in order to model long range interactions, properties like atomic charges and dipoles is needed. I wonder if it's possible to implement this feature based on the existing code to calculate them.
I've tried to implement it myself, but not clear exactly which part of the code should be modified. :(
Describe the solution you'd like
Additional context
atomic_number
in sn2 datasetThe text was updated successfully, but these errors were encountered: