Problem framing

How this question breaks down

1. 1Goal to objectiveCore19 minA model can only push up one number it sees in its training data. Framing means finding the decision a prediction will change, then picking the label, the population and the window so that pushing that number up also pushes the business goal up.
2. 2Task typesCore19 minA probability, a number, an order, a shortlist, new text, an oddness score and a set of groups are different answers, and whichever one the product needs fixes the labels, the loss, the metric and the serving shape before any model is chosen.
3. 3When not MLCore20 minA model earns its place only when a learned rule beats the best simple rule by more than it costs to build, feed and watch; many requests are better served by a rule, a lookup, a solver, a human or an experiment.

Work through the parts in order and you’ve covered the full design. Each part is one topic.