Introduced on day 15
Hyperparameter tuning
Searching over settings that are not learned from the training loss (depth, penalty, learning rate, stopping round) by scoring candidates on validation data.
In detail
Each candidate is a try. The best validation score among k tries is a maximum of noisy scores and is biased upward, increasingly so with more tries and smaller validation blocks. Report the number of tries, give a comparison baseline a comparable budget, and score the chosen configuration on rows that took no part in the search.