Loading…
Glossary term
Data kept separate until the end to measure how well the final model performs.
The test set is used only at the very end to give an honest estimate of how the finished model will do on new data. It must never influence training or tuning.
Peeking at it repeatedly, or letting it leak into training data, makes reported results look better than reality.
An untouched test set is the standard for trustworthy performance claims, and contamination of it undermines benchmark scores.