Why must the test set stay separate from tuning?
Why C is the answer
So that the final score reflects unseen data
Tuning on test data leaks information and inflates the estimate.
In detail
- A test set used in tuning no longer represents unseen data.
- Reusing the test set inflates the estimate.
Use case
An illustrative scenario, not a real incident.
A team tunes on validation data and scores the final model once on the untouched test set.
See it in different ways
flowchart LR A["All data"] --> B["Training: fit the model"] A --> C["Validation: tune the model"] A --> D["Test: final check"]
stateDiagram-v2 [*] --> Train Train --> Validate: tune Validate --> Train: adjust Validate --> Test: choose the final model Test --> [*]
mindmap
root(("Dataset split"))
n1["Training"]
n2["Fit the model"]
n3["Validation"]
n4["Tune"]
n5["Test"]
n6["Final score"]Key point: Keep the test set for the final score.
Why the other options are wrong
- A. So that the model trains for fewer epochs
- It has no effect on epochs.
- B. So that features need no scaling at all
- Scaling is a separate matter.
- D. So that training runs on a smaller instance
- It has no effect on instance size.