[Andy_G11]

Great, well-written article.

Interesting how mathematicians simplify the the number system to test theories. In Fermat's Last Theorem, the author describes Andrew Wile using the same tool of a reduced set of numbers cycling as if on a clock face in order to test his hypotheses.

When I have a tricky bit of coding to get on with, I often just resort to prototyping with very simple test cases - e.g. just simple arrays like [[dog, 1], [cat, 2], [cat, 1], [bird, 3]] (OK, not that simple, but you get the idea). Yet I often find that others are unwilling to go back to basics as if it is too time consuming - they would rather plug away with whatever unwieldy data set and tangle of relationships that comprise the ultimate problem set of input data.

Breaking something down to very simple fundamentals can be key.

[canjobear]

I supervise a lot of grad students writing numerical/machine learning code and my experience is if they’re not prorotyping their code on simple cases in your way, then it is almost certainly riddled with bugs they are unaware of.