Training and/or Validation set accuracies (not Test set accuracy) are common selection criteria in classic ML, and there's nothing wrong with that per se

→ View original post on X — @rasbt