Classical models built from the math you already know
Suppose a model aced a thousand training examples. How sure can anyone be it keeps that up? Taste one spoonful from a huge pot of soup and the read can be unlucky: a stray clump of salt, a cold patch near the ladle. That single spoonful is a shaky guide to the whole pot's true flavor. Taste several spoonfuls from all around the pot, though, and the average taste settles onto the truth, and it becomes very unlikely that reading is far off. A training set works the same way. It is a finite spoonful drawn from the vast, uncountable pot of every input a model might ever face.
🔒 This is a Pro lesson — the interactive figure, worked examples, quiz and practice open with Pro access.
▶ How Much Data Is Enough? A Real Generalization Bound