Operationalize and Monitor Models SNOWPRO-ADVANCED-DATA-SCIENTIST Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A healthcare company has a model that predicts patient no-show rates. The model is used to schedule appointments. The data science team noticed that the model's performance has degraded over the past month. They suspect data drift, but they are not sure if it is due to a change in patient demographics or a change in the appointment scheduling process. What is the best approach to diagnose the cause?

   • AIncrease the frequency of retraining to monthly instead of quarterly.
   • BAnalyze feature distributions and compare them to the training data, and also investigate any process changes.
   • CRoll back to a previous model version that performed well.
   • DRetrain the model with the most recent data to see if performance improves.
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   What is the primary benefit of using model versioning in a production ML system?

   • AIt automatically improves the model's accuracy
   • BIt eliminates the need for model monitoring
   • CIt allows tracking changes and rolling back to a known-good version if needed
   • DIt reduces the inference latency of the model
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A media company deploys a content recommendation model. The model's response time has been increasing over the past week, and the operations team is concerned about user experience. They want to monitor latency as part of their model health metrics. Which approach is most appropriate for tracking latency?

   • ATrack the maximum latency of any request and alert if it exceeds 500ms.
   • BTrack the 95th percentile latency and alert if it exceeds 500ms.
   • CTrack the average latency of all requests and alert if it exceeds 500ms.
   • DTrack the number of requests that time out and alert if it exceeds a threshold.
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   What is a common approach to retraining a model when new labeled data becomes available periodically?

   • ARetraining only when the model's latency exceeds a threshold
   • BScheduled retraining on a fixed cadence
   • CRetraining only when the model crashes
   • DRetraining only when a new model algorithm is published
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A ride-sharing company has a model that predicts ride demand. The model is retrained weekly. Due to a major event in the city, the demand pattern changed dramatically for a few days. The model's performance degraded during that period, but after the event, the demand pattern returned to normal. What is the best retraining strategy?

   • ADo not retrain immediately; wait for the next scheduled retraining and exclude the event data from the training set.
   • BRetrain the model immediately with the data from the event to capture the new pattern.
   • CRetrain the model with a weighted sample that gives more importance to the event data.
   • DRoll back to a previous model version that was trained before the event.
   Reveal Solution Discussion Select an answer first