Objective 4.1 Deploy a Model AIP-210 Practice Questions (Page 7)

Questions 31–35

1. 31foundation · easy

   What is the goal of scaling a machine learning deployment?

   • ATo simplify the model's code
   • BTo reduce the model's training time
   • CTo increase the model's accuracy
   • DTo handle varying prediction loads efficiently
   Reveal Solution Discussion Select an answer first
2. 32application · medium

   A team has deployed a new version of a churn prediction model to production. Before switching all traffic to the new version, they want to validate that it performs correctly with live data. What is the best approach?

   • AImmediately route all traffic to the new model and monitor for errors
   • BTest the new model only on the original training dataset
   • CCompare the new model's predictions to the old model's predictions in a batch job
   • DRun a shadow deployment where the new model receives a copy of live traffic but its predictions are not used
   Reveal Solution Discussion Select an answer first
4. 33application · medium

   A company has a real-time recommendation model that serves millions of requests per day. The ML team wants to detect when the model's performance degrades due to data drift. What should they monitor?

   • AThe model's training loss on the original training data
   • BOnly the system metrics like CPU and memory usage
   • CThe number of API requests and response times
   • DThe distribution of input features and prediction outputs over time
   Reveal Solution Discussion Select an answer first
5. 34foundation · easy

   Which of the following is a common format for an API endpoint that serves model predictions?

   • AMessage queue protocol
   • BSQL query
   • CFile transfer protocol (FTP)
   • DRESTful API
   Reveal Solution Discussion Select an answer first
6. 35expert · hard

   A company has a model that runs on a GPU cluster. The usage pattern is high during business hours and low at night. They want to reduce costs without sacrificing performance during peak hours. What is the best approach?

   • AUse autoscaling to scale down the cluster during low usage and scale up during peak hours
   • BMove the model to CPU-only instances to reduce cost
   • CLimit the number of predictions during off-peak hours
   • DUse a fixed cluster with enough GPUs to handle peak load
   Reveal Solution Discussion Select an answer first