Question: In a concise analytical response, connect the algorithm's relative performance on the two sets to the positions of their error curves.

Sample answer: A learning algorithm usually performs better on the training set than on the dev set. Better performance corresponds to lower error, so training error is usually lower than dev error. Therefore, the dev-error curve usually lies strictly above the training-error curve.

Key points:

• The algorithm usually performs better on the training set.
• Training error is therefore usually lower than dev error.
• The dev-error curve usually lies strictly above the training-error curve.

Rubric: A strong response states the usual performance difference, translates it into an error difference, and correctly explains the resulting curve positions.