What is the purpose of a validation set in machine learning?
- Store raw data
- Tune hyperparameters and make model selection decisions during development ✓
- Evaluate final model performance
- Train the model
Correct answer: Tune hyperparameters and make model selection decisions during development
Option B is correct because the validation set is used during the training and development phase to tune hyperparameters, compare model architectures, and guide model selection decisions, all without contaminating the final test evaluation. Option A is wrong because raw data storage is a data engineering concern, not a role of the validation split. Option C describes the test set, which is held out until after all development decisions are finalized and is used only once to report unbiased final performance. Option D describes the training set, which is what the model actually learns from via gradient updates or other fitting procedures.
Topic: · machine learning, validation set, hyperparameter tuning, model selection