AI Glossary

Validation and test sets

Also called held-out data, test set

What does “Validation and test sets” mean?

Validation and test sets: Portions of data kept out of training so a model can be checked on examples it has never seen. The validation set guides choices during development, such as when to stop training; the test set is saved for a final, honest measure of performance.

Validation and test sets visual guide: Portions of data kept out of training so a model can be checked on examples it has never seen. The validation set guides choices during development, such as when to stop training; the test set is saved for a final, honest measure of performance.
Visual guide Open full size (new tab)

What is an example of “Validation and test sets”?

A team might train on 80% of its labeled emails, tune on 10% and report final accuracy on the last 10%.

Keep learning

See also

OverfittingTermDatasetTermBenchmark contaminationTermEvalsTerm

Same topic

EpochTermRegularizationTermUnderfittingTermTransfer learningTermSynthetic dataTerm