Learning the scaler per-atom quantities

[xfanak]

Dear DP developers,

I see that DP can fit the atomic dipole now. Is there an even simplified function that allows learning of scaler per-atom quantities? For example, I want to fit the atomic activation energies, or displacement after some time. For the latter quantity, can I provide the travelled distance along each axis, namely dx, dy, dz, and just use the 'dipole' tag in fitting_net?

Thanks.

[amcadmus]

per-atom scaler one can train the model with atom_ener labels that are stored in atom_ener.npy files. The atomic scaler values of one frame are written on one line and the raw file contains number of frames lines. One can include the loss of atom_ener in the total loss by setting start_pref_ae and limit_pref_ae to non-zero values. More details please ref to the documentation of deepmd-kit inputs

one can use dipole to train the per-atom displacement.

[xfanak]

Thanks. I will try.