AI-300: Operationalizing Machine Learning and Generative AI SolutionsDemo

By Microsoft · Question Mode

QuestionQ22

Implement machine learning model lifecycle and operations

A team deploys a classification model to production and scores incoming customer data daily.

After several weeks, business stakeholders report unexpected changes in prediction behavior, although the endpoint remains healthy.

Determine whether data drift is occurring and, if it is, identify the appropriate actions.

Which action should be performed for each observed signal? Each action may be used once, more than once, or not at all.

Drag & Drop
Analyze dataset drift metrics.
Review prediction and ground truth trends.
Investigate scaling and infrastructure metrics.
Rebuild the inference container image.
Input feature distributions differ from training data.
Model accuracy drops without code changes.
Endpoint latency increases under load.
Explanation

A shift between production input-feature distributions and the training distribution is data drift and should be assessed with dataset-drift metrics. A model-accuracy decline requires reviewing predictions alongside ground-truth labels to distinguish performance degradation or concept drift. Higher endpoint latency under load is addressed by examining capacity, scaling, and infrastructure telemetry; rebuilding the container image is not indicated by these signals.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!