Compare general outputs with the scikit-learn machine_learning_map?

How does the output of this compare to http://scikit-learn.org/stable/tutorial/machine_learning_map/ ... which I took as general best practice, but I have no idea how current it is with state of the art.

I don't mean in terms of quality, I meant in terms of the general advice that the map gives, and how often TPOT happens to converge on the same one that the map would have suggested. Which could either reinforce the map (great!) or point out interesting areas where it diverges.