The networks trained were really small, with only one hidden layer, and a width of 16.