引導式測驗 · 中等 等級
用於 ML 資料測驗的 Lakehouses 和 Delta Lake
了解 Delta 事務日誌、架構控制和時間旅行如何支援 ML 工作流程,包括其保留和再現性限制。
問題 1 的 8
What does the Delta transaction log record?
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。
- What does the Delta transaction log record?
- Every model parameter automatically
- A permanent backup of all source data
- Committed table changes and metadata used to construct table snapshots
- Only dashboard views
- How can a Delta table version help an ML experiment?
- It records the full model environment automatically
- It guarantees that all external sources remain available
- It ensures the labels are correct
- It can identify a historical table snapshot for the dataset input
- What can make an older Delta time-travel version unavailable?
- Adding a model metric
- VACUUM or retention cleanup removing required data files or log history
- Changing a notebook title
- Recording a random seed
- What storage assumptions underpin Delta’s ACID guarantees?
- Storage behavior such as atomic visibility, mutual exclusion, and consistent listing
- That all files are stored in RAM
- That every engine uses SQL only
- That no concurrent reads occur
- Does Delta table versioning automatically capture preprocessing code and random seeds?
- Yes, every version stores the full Python environment
- Yes, but only for Spark
- No; teams must record those experiment details separately
- No, because table versions contain no data
- How are schema enforcement and schema evolution different?
- They are identical behaviors
- Enforcement rejects incompatible writes; evolution permits configured schema changes
- Enforcement changes every schema automatically
- Evolution is a kind of time travel
- What should a team verify before relying on a Delta feature across engines?
- Only the cloud provider’s logo
- That all engines have identical support
- Nothing, because Delta is a file extension
- The engine and protocol support the needed feature
- What should be saved for more complete ML reproducibility?
- Dataset version, code, preprocessing, parameters, environment, and relevant seeds
- Only the table name
- Only the model’s accuracy
- A screenshot of a dashboard