I completely agree with you on knowing the math part. I am kind of emphatic about it too when doling out advice. Without that, getting good results just becomes

I completely agree with you on knowing the math part. I am kind of emphatic about it too when doling out advice. Without that, getting good results just becomes a blind exercise in tweaking parameters. With that, you can really troubleshoot your models, and know what model(s) to try next; the path to optimality is somewhat meaningful and reasonable.