AI-300 Practice Question 74: Online Data Collector
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 8 · question 4 of 10
The team wants to set up model monitoring for an online deployment, but the deployment does not store any production data. They need the inputs and outputs of scoring requests to be saved to the workspace for use by monitoring signals. What should they configure?
Show answer and explanation
Answer:
Model monitoring compares production data with a reference. To capture that production data from an online endpoint, add a data_collector section to the deployment and enable the model_inputs and model_outputs collections. The scoring script must also create Collector objects for these collections (MLflow models can enable this automatically). The collected data is written to workspace storage as a data asset that monitoring signals can use (option D).
- A is wrong because request telemetry records operational details. It does not capture full model inputs and outputs in a form monitoring signals can read. - B is wrong because the training data is the reference. Production inputs and outputs must be collected from the live deployment. - C is wrong because logger output goes to container logs. It is not a structured, monitorable data set.
Reference: Collect production data from models deployed for real-time inferencing