Not to raise an error when the training is not successful

[Yin-Jia Zhang]

In amp/amp/ init.py, an error is raised when the energy_rmse and force_rmse convergence criteria is not reached by the following line.

if result is False: raise TrainingConvergenceError('Amp did not converge upon ' 'training. See log file for' ' more information.')

Could we let it not raise an error in such case? The output .amp file and the log file have already indicated that the training is unsuccessful. An error raise will interrupt the subsequent operations after the training.

[andrew_peterson]

This was intentional. In this case you can re-start training right in your python script if it's not successful. E.g.,

try:
    calc.train(images)
except TrainingConvergenceError:
    # Some code here to restart training...

What problems does it create with you? Can you solve it by using a try...except clause?

[Yin-Jia Zhang]

Ah, that would work. I am thinking to let qmml to continue even if the convergence criteria is not reached. Thanks a lot!

[andrew_peterson]

Ok, I just added a note to the docs showing how to do this. 15dc5f4