問題文
A platform team is standardizing how machine learning projects move from development to production on Databricks. They ask which asset should be promoted between the environments. What does Databricks recommend for most situations, and why?
選択肢
- Promote the training data snapshot, because the model and the code can both be regenerated from it deterministically each time.
- Promote both the code and the model artifact together in every case, because either one alone leaves a gap in the audit trail, one for each check.
- Promote the code, so that the same review and integration testing apply to everything and the production model ends up trained by production code on production data.
- Promote the trained model artifact, because moving a validated binary avoids the risk of the training code behaving differently in production, and the binary is what was validated.