Operations and Processes
Free CompTIA DataX practice — 6 questions on Operations and Processes, with explanations. No sign-up. Full 12-question mixed test →
A team monitors a production credit-risk model using the Population Stability Index (PSI), computed weekly between the current feature distribution and the training baseline. This week the PSI for the applicant-income feature is 0.31. Based on standard drift-monitoring thresholds tested on the exam, which action is BEST justified?
Industry-standard PSI thresholds classify <0.1 as no significant shift, 0.1-0.25 as moderate shift worth watching, and >0.25 as significant shift requiring investigation and likely retraining.
A team is building a real-time fraud-scoring service. They need identical feature computation logic used at both training time and inference time, low-latency lookups (single-digit milliseconds) for pre-computed features, and versioned feature definitions shared across multiple models. Which component BEST satisfies these requirements?
A feature store maintains consistent, versioned feature definitions and provides both a low-latency online store for real-time serving and an offline store for training, directly preventing train-serve skew.
A team deploys a new fraud-detection model version using a canary release strategy, initially routing 5% of production traffic to it while comparing latency and accuracy metrics against the existing model before increasing the rollout percentage. Which statement correctly describes this configuration?
Canary deployment reduces blast radius by exposing only a small, real subset of production traffic to the new version, allowing teams to detect regressions before increasing exposure.
What BEST distinguishes a champion/challenger evaluation framework from a standard fixed-duration A/B test in an MLOps context?
Champion/challenger is an ongoing evaluation pattern where challengers are continuously scored against the reigning production model with defined promotion thresholds, whereas an A/B test is typically time-boxed with a defined stopping point and statistical decision.
A CI/CD pipeline automatically promotes new model versions to production after passing validation tests. Shortly after a new version deploys, monitoring alerts trigger for a significant spike in prediction latency and error rate. Which MLOps practice BEST enables rapid recovery from this incident?
A model registry retains prior validated, versioned artifacts, enabling near-instant rollback to a known-good state, which is the fastest and most reliable recovery path when a new deployment causes regressions.
A data science team needs to orchestrate a pipeline involving conditional branching, explicit task dependencies across ingestion, feature engineering, and training stages, automatic retries on failure, and scheduled execution. Which orchestration approach is BEST suited for this requirement?
DAG-based orchestrators explicitly model task dependencies, support conditional branching, retries, and scheduling, making them the standard tool for managing complex multi-stage data science pipelines.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →