globus-labs / faster-molecular-hessians

Reduce the amount of computation needed for Hessians using machine learning
MIT License
0 stars 1 forks source link

Add SOAP forcefields #19

Closed WardLT closed 8 months ago

WardLT commented 9 months ago

Use SOAP, as implemented in DScribe, as features for a GPR regression model.

Still need to

WardLT commented 8 months ago

There's still more to do with SOAP, but let's merge what we have now. The forces predicted by SOAP are very poor compared to those from MBTR, and I suspect this has to do with how the model's trained or the choice of hyperparameters rather than a mistake in the software (see unit test comparing numerical to analytic forces).