Case context: A team building a fraud-detection model first trained it on 400 labeled transactions and measured a 30% dev-set error. They then retrained the same model using 1,600 examples and later 6,400 examples, while keeping the dev set unchanged.

Question: According to the general learning-curve relationship in machine learning strategy, what should happen to the dev-set error as the team moves from 400 to 1,600 and then 6,400 training examples?

Sample answer: The dev-set error should trend downward as the training set grows. A larger training set should generally produce a model that performs better on the dev set than the version trained on only 400 examples.

Key points:

• Increasing the training-set size should reduce dev-set error.
• The model trained on more examples should be evaluated as having better dev performance.

Rubric: The student must correctly state that dev-set error is expected to decrease as the number of training examples increases from 400 to 6,400.