First, I agree with both (1) the encouragement of who might have been
a student "wanting to apply a neural network to *something*," and also
(2) the concern in the workplace re the willingness to advocate/use
machine learning techniques when other approaches are available...
that are "simpler," or even more efficient, or provably optimal, etc.
I qualify "simpler" because I think the problem is almost the
reverse-- I see analysts willing to take the machine learning black
box tool off the shelf, and turn its crank, with *less* required
understanding of the mathematics of how those tools actually work.
But aside from the debate about the machine learning context, I hope
readers will visit the gist link and actually read the code. I am
repeatedly impressed by the combination of clarity and conciseness--
an optimization of a driving simulation, with cool video rendering of
the results, all in less than 500 lines of code.
> I am repeatedly impressed by the combination of clarity and > conciseness
Thanks, Eric! I had actually buried the code link at the bottom of the
article because I didn't feel it wasn't quite up to my usual standards,
and that was because I wasn't planning to discuss the code, just the
concepts. I'm glad to hear it wasn't so unreadable as I feared!