Train/Validation/Test Split

Easy

machine-learning Meta Google Amazon

A data scientist tunes hyperparameters using the test set and reports test set performance as the final metric. Why is this problematic?

Multiple Choice

It is not problematic; the test set is meant for tuning and evaluation The test set performance becomes an optimistic estimate because the model was indirectly optimized on it It only matters if the test set is smaller than 20% of the data This is acceptable as long as cross-validation was also used

Correct!

Incorrect — try again next time!

Upgrade to Pro to unlock solutions and explanations

Train/Validation/Test Split

Multiple Choice

Related problems

Deep-dive guides