MLA-C01 Practice Question 28: Batch Transform Nightly Scoring
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 3 · question 8 of 10
A propensity model scores about 40 million customer records once a night. The records are read from Amazon S3 and the scores are written back to S3 for a marketing system to pick up in the morning; a four-hour window is plenty. No application ever calls the model interactively. Today a real-time endpoint runs 24 hours a day and a script loops over the records calling InvokeEndpoint one batch at a time. Finance wants the lowest cost for this workload.
Show answer and explanation
Answer:
Batch transform is the option built for getting inferences from large datasets when no persistent endpoint is needed. SageMaker AI starts instances for the job, splits the S3 objects across them, writes an output object per input object, and shuts the instances down when the job finishes, so the team pays only for the minutes the scoring actually runs, and the looping client script disappears (option D).
- A is wrong because a scaled-down endpoint still bills for at least one instance every hour of the day, including the 20 hours nothing is being scored. - B is wrong because serverless inference is sized for spiky, low-volume request traffic. Driving 40 million single requests through it keeps the per-request overhead and is a poor fit for bulk offline scoring. - C is wrong because asynchronous inference targets long-running or large-payload requests that still need a per-request response with a queue in front. It keeps an endpoint and its instances configured, where the workload has no request/response requirement at all.
Reference: Batch transform for inference with Amazon SageMaker AI