問題文
A machine learning system has four pipelines: feature computation, training, evaluation, and deployment. The team wants a test that would catch a mismatch where the training pipeline expects a column that the feature pipeline no longer writes. What kind of test is that, and what must it cover?
選択肢
- A schema check on the feature table that runs after the feature pipeline completes, independent of the training code; the failure is reported against the table and stops there.
- A unit test of the training function that asserts the presence of every expected column in its input before training.
- An integration test that runs the pipelines in sequence on a small dataset, so the handoff between feature computation and training is actually exercised.
- A smoke test of the deployment pipeline, because a column mismatch surfaces as a failure at inference time, when the request errors.