Revisiting force training keyword

Issue #60 resolved
Alireza Khorshidi created an issue

There are different coefficients related to force training: "train_forces", "force_coefficient", "force_rmse", and "force_maxresid". We need to think about which one can be removed, which one is the major keyword, and different scenarios for the user.

Comments (4)

  1. andrew_peterson repo owner

    Yes -- I think the user still has to specify in multiple places how to turn off force training. This needs to be made simpler.

  2. andrew_peterson repo owner

    Btw, it might be three places. The force_coefficient in LossFunction, the force_rmse in LossFunction's convergence dictionary, and the train_forces keyword in calc.train!

  3. andrew_peterson repo owner

    This should be taken care of in commit 81f4508. However, I almost certainly broke the tensorflow module in this commit, and I don't have a working installation at the moment, so I expect a new issue to arise while we get that figured out!

  4. Log in to comment