Review this Python model-evaluation routine.
It always selects max_depth=None with accuracy 1.0. Is that the best model?