traning set and test set

[lammps-vasp]

Is deepmd kit DNN back propagation? Will the value of numb_test in input.josn file affect the model? Will increasing the value of numb_test reduce the test error in lurcve.out? When I run deepmd with only energy, the test error will not converge, and the training error will only decrease by an order of magnitude.
Looking forward to your reply!

[marian-code]

No, numb_test does not affect the quality of the fit. numb_test is essentially batch_size for test set. The total number of test structures is determined solely by the size of last(test) set.

I do not have the answers to your other questions. But the poor quality of the fit certainly has nothing to do with numb_test setting.

[DingChangjie]

Did you include virial information (virial.npy) in your training set ? I also encountered similar problems. After included viral, things worked well and the error could reach a reasonable level.