Signs of good learning

Hello

I wondering every loss parameter (like hard_loss_metric, full_loss_metric, full_loss_num, and mean_dist) should be how much that indicate a good training and good hyperparameter setting?

Thanks for your attention