フリー問題

問題文

Which check belongs in the automated pipeline that runs on every change to the training code?

選択肢

1. A full training run on the complete dataset, because only the full run can reveal a regression that appears at production scale and running anything smaller would give false confidence about a change that behaves differently once the whole input is processed by the pipeline.
2. A comparison of the fitted coefficients against the previous run's coefficients, and any change in the code shows up in those numbers before it shows up in the metric.
3. A fast run on a small fixed sample that asserts the pipeline completes and that the evaluation metric stays above a recorded floor.
4. A manual review of the produced model by a data scientist, and a person reading the metrics and the coefficients will catch problems that an assertion in code would miss.