Angie · learn

Random search

Sampling combinations at random instead of testing all of them. Cheap coverage of a space too large to walk, at the cost of possibly missing the best corner.

Updated 2026-09-03

A random search samples parameter combinations at random rather than trying every one. With the same budget of tests it usually finds a better result than a grid search, which is counterintuitive and well established.

The reason is that most parameters do not matter much. A grid spends its budget evenly across every dimension, including the ones that make no difference, so with a thousand tests over three parameters it only ever tries ten distinct values of the one parameter that actually matters. Random sampling tries a thousand different values of it.

It is the right default when you have more than two or three parameters and no strong prior about which ones matter. It inherits every overfitting risk of a grid search, and the same defence applies: judge the region, not the single best point.

See also

  • Grid search — Testing every combination of the settings you chose. Exhaustive and certain, and the count explodes as soon as you add a third or fourth knob.
  • Evolutionary search — Seeding a population of settings, keeping the strongest, then mutating and recombining them over generations to home in on what works.

← See this one in the deck

Educational content, not investment advice. Angie explains how a valuation is built so you can judge it yourself. What you do with that is your call.

IsiofiaIsiofia