Automating Model Retraining PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A fintech company runs a fraud detection model that must be as fresh as possible to catch new fraud patterns. However, each retraining job is expensive and uses significant compute resources. The team is considering moving from a weekly retraining schedule to a daily one. What is the most important trade-off they should evaluate?

   • AThe need for more frequent model validation versus the risk of deploying an untested model.
   • BThe complexity of automating the daily pipeline versus the simplicity of the weekly schedule.
   • CThe risk of overfitting to recent data versus the benefit of capturing new fraud patterns.
   • DThe increased cost and resource usage versus the potential improvement in model accuracy from fresher data.
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A data science team wants to adopt CI/CD principles for their ML models. They currently train models in notebooks and manually deploy them to production. What is the first step they should take to implement a CI/CD pipeline?

   • ADeploy the model to a staging environment and manually test it before promoting to production.
   • BUse Cloud Build to deploy the model to production automatically without any testing.
   • CSet up a continuous training pipeline that retrains the model every time new data arrives.
   • DVersion control the training code and model artifacts, and create a build pipeline that runs automated tests on the code.
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A company has a CI/CD pipeline that deploys a new model to production. They want to ensure that if the new model performs worse than the current one, they can quickly revert to the previous version. What should they implement?

   • AA manual process where the team can redeploy the old model if needed.
   • BA monitoring system that only alerts on data drift, but does not trigger any action.
   • CA versioning system for the model artifacts, but no automated rollback.
   • DA rollback strategy that keeps the previous model version and can automatically revert if the new model's performance degrades.
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   What is the purpose of pipeline orchestration in an ML pipeline?

   • ATo monitor the model's performance in production
   • BTo visualize the training data
   • CTo store the trained model in a versioned repository
   • DTo coordinate the execution of training, validation, and deployment steps in the correct order
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   Which tool is commonly used in Google Cloud to automate the deployment of a trained model to a serving environment?

   • ACloud Monitoring
   • BCloud Storage
   • CCloud Build
   • DCloud Logging
   Reveal Solution Discussion Select an answer first