Operationalize and Monitor Models SNOWPRO-ADVANCED-DATA-SCIENTIST Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   Why is detecting data drift important for a deployed machine learning model?

   • ABecause it measures the time taken for the model to make predictions
   • BBecause it directly indicates that the model's predictions are incorrect
   • CBecause it can signal that the model's performance may degrade even if ground truth labels are unavailable
   • DBecause it identifies which model version is currently deployed
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A healthcare startup deploys a model that predicts patient readmission risk. The model is used by clinicians to make discharge decisions. The operations team needs to set up an alerting system to notify them when the model's performance degrades. Which alerting strategy is most appropriate for this high-stakes environment?

   • AAlert on any single metric that falls below a threshold, and page the on-call engineer immediately.
   • BAlert only when the model's predictions are significantly different from clinician decisions, and log the details for later analysis.
   • CMonitor multiple metrics (accuracy, drift, latency) and alert when a combination of metrics indicates a problem, with a clear escalation path.
   • DSet up a dashboard that shows model metrics, and manually check it daily.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   An online retailer has a model that predicts product returns. The model is deployed in production. The operations team wants to monitor the model's performance using a dashboard. Which combination of metrics should they include to get a comprehensive view of model health?

   • AAccuracy, data drift, latency, and prediction volume.
   • BData drift and prediction volume only.
   • CPrecision and recall only.
   • DAccuracy and latency only.
   Reveal Solution Discussion Select an answer first
5. 24expert · hard

   A logistics company has a model that predicts delivery delays. The model is retrained monthly. The data science team has noticed that the model's performance is stable, but the business environment is changing rapidly. They want to implement a more responsive retraining strategy. What is the best approach?

   • ARetrain the model only when new data is available, regardless of performance.
   • BImplement a trigger-based retraining strategy that retrains when data drift or performance degradation is detected.
   • CKeep the monthly schedule and monitor performance closely.
   • DRetrain the model weekly instead of monthly.
   Reveal Solution Discussion Select an answer first
6. 25expert · hard

   A manufacturing company has a predictive maintenance model that uses sensor data from machines. The model is retrained every month. Recently, the data science team noticed that the distribution of sensor readings has shifted due to a change in the manufacturing process. The model's performance has degraded, but the team is unsure whether to retrain or roll back. What is the best approach?

   • ARetrain the model immediately with the new sensor data, as the drift indicates the model is outdated.
   • BKeep the current model and monitor for another month to see if the drift stabilizes.
   • CInvestigate the drift to understand the cause, then decide whether to retrain with the new data or adjust the model.
   • DRoll back to the previous model version, as it was trained on more stable data.
   Reveal Solution Discussion Select an answer first