引導式測驗 · 中等 等級

ML 管道測驗的資料契約

了解資料契約的規定內容、可執行測試如何強制執行它,以及為什麼僅聲明不能保證資料的正確性。

相關引導路徑ML 管道的資料契約
問題 1 的 8

Which items can an ODCS data contract describe?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. Which items can an ODCS data contract describe?
    • A guarantee that every model is fair
    • The accuracy of every prediction
    • Schema, semantics, quality expectations, and service levels such as freshness
    • A replacement for monitoring
  2. How can a pipeline detect a changed upstream field against a contract?
    • A producer or consumer system validates actual data against the declared rule
    • The contract file exists somewhere
    • The ML model is retrained automatically
    • The producer team sends no notice
  3. Does “contract” in a data-contract specification necessarily mean a legal contract?
    • 是的,總是
    • Only if it has a YAML file
    • Yes, and every consumer automatically signs it
    • No; the term commonly describes a technical agreement and is not necessarily legal
  4. Which standard does the Data Contract Specification project recommend for new adoption after its deprecation notice?
    • OpenLineage Run Cycle
    • Open Data Contract Standard (ODCS)
    • Parquet schema registry
    • Delta Change Data Feed
  5. What can a passing data-contract check establish?
    • The data are unbiased and semantically perfect
    • Every future batch will pass
    • The tested batch met the specific checks that were implemented
    • The model is safe for deployment
  6. Who should resolve a meaning such as “eligible customer” in a contract?
    • The schema validator alone
    • Producer and consumer domain owners who share the data meaning
    • A hash function
    • The storage system
  7. Which service objective can be represented in a data contract?
    • A maximum age for the newest record using a timestamp field
    • Guaranteed model accuracy on future users
    • Guaranteed fairness without evaluation
    • A promise that values are always true
  8. What should a team define for contract violations?
    • Nothing after validation fails
    • Automatic replacement with guessed values
    • A rule that hides all failures
    • An owner and explicit alerting, quarantine, or remediation behavior