Grid search
Also called Grid Search
Try every combination of a fixed set of settings, exhaustively, and pick the best.
Think of it like
Testing every pairing on a menu — each starter with each main — to find the one you like most.
Example
A team lists 3 learning rates and 3 batch sizes and runs all 9 combinations to see which wins.
How it actually works
Grid search enumerates the full Cartesian product of candidate values. It is simple and reproducible but scales terribly — adding one more hyperparameter multiplies the runs — and it wastes effort on dimensions that barely matter. Random search often finds comparable settings for less compute, which is why grid search is fading for anything beyond a couple of knobs.
For product teams
Thorough but expensive; fine for a tiny search, wasteful for a big one.
For engineers
Exhaustive evaluation over the Cartesian product of hyperparameter values; cost grows exponentially with dimensions.
Related
- Hyperparameter tuning — The broader activity it is one method for.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome