📏 Features: the world as numbers
Why: Turn real things into numbers — features and labels, the raw material of all ML.
Unlocks: Training and testing — how we know a model actually learned.
A machine can't look at an apple. It can only look at numbers — so the first job in any ML project is choosing which measurements describe the thing. Those measurements are called features; the answer we want is the label.
Fruit on a scale: weight in grams, and a "redness" score 0-10. Two features per fruit. The label: apple or watermelon.
With features this clean, you can almost SEE the rule: everything near 150g is an apple, everything in the kilograms is a watermelon. Write that insight as code — a one-feature model, learned by your own eyes:
You're building a model to predict apartment prices. Which of these is a FEATURE (not the label)?
Why did weight alone nearly solve the fruit problem?