Yes, it is reliable way of training. Learner App does nothing extra, it uses same internal function and just a good looking UI on top of it. It may happen that the performance metric differ, but surely the difference won't be big. these differing behaviour is because of random intialisation and training. We may not expect the results to come exactly same everytime. the little bit of randomness in the model make the values to differ slightly.
I hope this clears your confusion.
Cheers.
