Training, validation, and test phases in AI -- explained in a way you'll never forget

If you've heard of validation in the context of machine learning (ML) and AI but you're not quite sure what all the fuss is all about -- validation is only one of the most important applied AI concepts ever, no big deal -- then here's the analogy you've been waiting for. Imagine that Mr. Bean is about to take his first calculus exam… Mr. Bean unearths the single equation he squirreled away and begins studying it for tomorrow's exam. He's got no other examples (datapoints) or resources to help him along and he didn't bother to write down any explicit rules explaining how calculus works, so all he can try doing is search for patterns in his equation: Just like an AI algorithm, his goal is to find a data pattern that he can turn into a recipe ("model") that successfully takes him from the input on the left of the " " to the output on the right-hand side. That is precisely what goes on during the training and tuning steps of an applied AI project (steps 6–7 in my step-by-step guide). Training is all about making a recipe out of patterns in the available examples.