引導式測驗 · 硬 等級

時間點正確的功能加入測驗

了解事件時間連接、TTL、後期創建的更正、未來特徵洩漏,以及為什麼僅靠時間點連接無法保證乾淨的資料集。

問題 1 的 8

What problem does a point-in-time correct join prevent?

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. What problem does a point-in-time correct join prevent?
    • Joining feature rows whose event timestamps are later than the example’s decision timestamp
    • Slow query performance on large tables
    • Duplicate rows appearing in a feature table
    • Features being stored in the wrong file format
  2. Why would training a churn model on a 'total lifetime purchases' feature computed after the churn date be a problem?
    • It's not a problem, since it improves accuracy
    • It would include purchases the customer made after already churning, which wouldn't be known at prediction time
    • Lifetime purchase counts are always inaccurate
    • It duplicates the label column
  3. Besides an entity key, what timestamp lets a historical feature lookup know which event-time value to retrieve?
    • The time the training job starts
    • The date the model was published
    • A feature event timestamp and an entity-row decision timestamp
    • The current wall-clock time only
  4. For each labeled row, what feature value does a correct as-of join select?
    • The feature value from exactly one year before the label
    • Any feature value regardless of timestamp
    • The average of all historical feature values
    • The most recent feature value at or before the label's timestamp
  5. What mistake does the guide describe as joining on a feature table's current snapshot?
    • It silently includes feature updates that happened after the labeled event
    • It correctly reflects historical state
    • It removes all timestamp information from the join
    • It only works for numeric features
  6. How do a feature event timestamp and a created or ingestion timestamp differ?
    • They are always identical in practice
    • Event time records when the underlying event happened; created time records when the feature row was created or became available
    • Ingestion timestamp only applies to streaming data, never batch data
    • Event timestamp is set by the feature store automatically and cannot be changed
  7. According to the guide, how is an as-of join commonly implemented at scale?
    • By manually inspecting each row in a spreadsheet
    • By deleting all historical feature versions except the latest
    • With a window function partitioned by entity key and ordered by feature timestamp, or dedicated feature store SDK support
    • By running the join only on a random sample of rows
  8. According to the guide, what is one limitation of a point-in-time join?
    • It always removes every possible form of leakage
    • It replaces the need for entity keys
    • It automatically fixes late-arriving data in every store
    • It cannot detect every future-derived feature or incorrect label cutoff