Split Data Between Endpoints for Realtime Interference MACHINE-LEARNING-ASSOCIATE Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A financial services firm deploys a fraud-detection model to a real-time endpoint. They monitor the endpoint's latency and accuracy weekly. After a new product launch, they observe a sudden increase in latency and a drop in accuracy. What should they do first?

   • ARoll back to the previous model version.
   • BAnalyze the distribution of incoming requests to identify if the data split has changed.
   • CIncrease the endpoint's compute resources to reduce latency.
   • DImmediately retrain the model on the most recent data.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   An e-commerce company deploys a recommendation model to two endpoints: one for new users and one for returning users. They notice the new-user endpoint has high latency and low accuracy. The training data was 90% returning users. What is the most effective way to improve the new-user endpoint's performance?

   • ACollect more data on new users and retrain the model with a balanced dataset.
   • BRoute all new-user traffic to the returning-user endpoint.
   • CIncrease the compute resources for the new-user endpoint.
   • DReduce the new-user endpoint's request timeout to lower latency.
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   A logistics company deploys a delivery-time prediction model to a real-time endpoint. The training data includes a feature 'package_weight' with values between 0.1 and 50 kg. At serving time, the endpoint receives a package_weight of 100 kg due to a system error. The model returns a prediction. What is the most appropriate action to handle this?

   • ASend the raw value of 100 kg to the model.
   • BConfigure the serving pipeline to clip the package_weight to the maximum value seen in training (50 kg).
   • CRetrain the model to include package weights up to 100 kg.
   • DReject the request because the package_weight is out of range.
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A retail company trains a churn-prediction model on 12 months of historical customer data. They deploy the model to a real-time endpoint that receives a single customer's features at checkout. The data science team notices the model's predictions drift because the endpoint receives data in a different format than the training data. Which action best addresses this issue?

   • ARetrain the model weekly on the same 12-month dataset to keep it fresh.
   • BIncrease the endpoint's autoscaling limits to handle more concurrent requests.
   • CApply the same feature-engineering pipeline used during training to the endpoint's input data before inference.
   • DLog all inference requests and manually inspect them for anomalies each week.
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A bank deploys a loan-approval model to a real-time endpoint. The training data was collected over 5 years and includes a feature 'credit_score' with values between 300 and 850. At serving time, the endpoint receives a 'credit_score' of 900 due to a data entry error. The model returns an approval decision. What is the most appropriate action to handle this?

   • ARetrain the model to include credit scores up to 900.
   • BReject the request because the credit_score is out of range.
   • CSend the raw value of 900 to the model.
   • DConfigure the serving pipeline to clip the credit_score to the maximum value seen in training (850).
   Reveal Solution Discussion Select an answer first