200 examples at most, really?? I have been led to believe that (tens of) thousands is more typical. If you can get excellent results with that few examples, that changes the equation a lot.