問題文
A one-hot encoding is fitted on the training data, which contains 12 payment methods. In production, a 13th method appears. Which option correctly identifies the two things that should be part of the design? (Select two.)
選択肢
- A defined behavior for categories that were not seen during fitting, so the unseen value does not break the inference.
- Dropping the rows with the unseen category, because a row that the encoding cannot represent should not be scored at all under any circumstances.
- Switching to label encoding, because integer codes fit any number of categories.
- A monitor or check that reports when unseen categories start appearing, so the team learns that the world has changed.
- Refitting the encoding at inference time on the incoming rows, so the 13th method receives its own column automatically as soon as it appears in production traffic.