Serving Models PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 6)

Questions 26–30

1. 26foundation · easy

   What is the purpose of the Vertex AI Model Registry?

   • ATo monitor model performance
   • BTo serve predictions
   • CTo train new models
   • DTo store and organize trained models with metadata
   Reveal Solution Discussion Select an answer first
2. 27expert · hard

   A company has a model that requires a complex feature transformation using a custom library. The transformation must be applied consistently across online and batch inference. They want to avoid duplicating the transformation code. What is the best approach?

   • ATrain the model to accept raw features and perform the transformation internally.
   • BCreate a shared preprocessing library and include it in both the online serving container and the batch prediction container.
   • CUse Vertex AI Feature Store to perform the transformation.
   • DImplement the transformation in the client application and pass the transformed features to the model.
   Reveal Solution Discussion Select an answer first
4. 28foundation · easy

   What is a common use case for model versioning in the Model Registry?

   • ARolling back to a previous version if a new version performs poorly
   • BTraining a new model from scratch
   • CDeleting old models to save storage
   • DMonitoring model drift
   Reveal Solution Discussion Select an answer first
5. 29application · medium

   A fraud detection model outputs a raw probability score. The business requires a binary 'fraud' or 'not fraud' decision, and they want to be able to adjust the threshold without redeploying the model. What is the best approach?

   • ARetrain the model with a different loss function to output binary labels directly.
   • BUse Vertex AI Prediction's built-in thresholding feature.
   • CApply a threshold in the serving container's postprocessing logic and make it configurable via an environment variable.
   • DHave the client application apply the threshold after receiving the probability.
   Reveal Solution Discussion Select an answer first
6. 30foundation · easy

   Which action can you perform directly from Model Garden?

   • AStore a trained model artifact
   • BMonitor model drift
   • CCreate a new custom model from scratch
   • DDeploy a pre-trained model to an endpoint
   Reveal Solution Discussion Select an answer first