Mature fields ideally are driven by theory. Newer fields are driven by experiment, until the theory catches up. I've learned that it is a waste of time (and br

Mature fields ideally are driven by theory. Newer fields are driven by experiment, until the theory catches up. I've learned that it is a waste of time (and brainpower) to blindly tweak code without understanding the algorithm, hoping it will magically work. Yet, I've observed many newcomers to machine learning essentially doing this. It might be somewhat useful at this moment, but if it is easy to do then it will likely become automated.