Implement ML model lifecycle and operations: free practice questions

Using a weighted or macro-averaged metric such as AUC_weighted or average_precision_score_weighted ensures that Automated ML's model selection is driven by performance on the minority class, rather than being dominated by the majority class as with raw accuracy. This is the recommended primary metric change for imbalanced classification. Setting featurization mode to 'auto' enables Automated ML's automatic featurization pipeline, which includes class imbalance detection and mitigation techniques such as oversampling (SMOTE) and class-weight adjustments. There is no enable_class_balancing parameter in the Azure Machine Learning Automated ML SDK v2 API; this is a distractor reflecting a plausible but non-existent setting. Setting n_cross_validations to 1 effectively disables cross-validation, which would produce unreliable performance estimates especially on an imbalanced dataset, and is counterproductive. LogisticRegression supports class_weight='balanced' natively and is one of the models that handles imbalance reasonably well; blocking it would reduce the model search space without benefit.

Fairness assessment measures performance disparities across demographic or categorical subgroups (here, language subgroups), satisfying requirement 1. Interpretability (feature importance) surfaces global and local feature attribution scores, explaining which input features drive individual predictions, satisfying requirement 2. Error analysis identifies cohorts and input segments with the highest error rates, helping engineers find systematic failure modes, satisfying requirement 3. Counterfactual analysis generates what-if perturbations for individual instances to explain prediction boundaries but does not address group fairness or systematic error patterns. Data explorer provides statistical summaries and distributions of the dataset but is not an evaluation component for the three stated requirements. Causal analysis estimates causal treatment effects of features on outcomes, which is relevant for intervention planning but does not address fairness, feature attribution for predictions, or error cohort identification.

Azure Machine Learning studio's model version lineage tab natively surfaces the complete provenance graph: the training job that produced the model, the versioned data assets consumed, the registered environment, and the pipeline that orchestrated training—all accessible from the model registry entry. Azure Monitor Logs stores operational events but does not provide a structured lineage graph linking model versions to datasets and environments. The MLflow Model Registry UI shows run-to-model associations but does not surface Azure ML-specific lineage details such as dataset versions or pipeline run hierarchies as richly as the studio UI. Azure Purview can catalog ML assets discovered by scanning Azure ML workspaces but is a governance tool, not the primary surface for real-time lineage within an Azure ML workspace.

Azure Machine Learning managed online endpoints natively support multiple deployments under a single endpoint with configurable traffic weights that must sum to 100. Setting weights of 90 and 10 routes that proportion of live requests to each deployment without any external component. Option A requires an external load balancer and defeats the purpose of the native feature. Option C is incorrect: there is no separate 'A/B testing mode' toggle in managed online endpoint advanced settings; traffic weights achieve this directly. Option D is wrong because batch endpoints process offline data in bulk and do not perform live traffic splitting.

Azure Machine Learning pipeline step caching (output reuse) is controlled by the is_deterministic flag on each step. When set to True (the default), AzureML computes a cache key based on the step's code, environment, and input data. If a matching cache entry exists from a prior run, AzureML skips re-execution and reuses the cached output, saving compute time. The team should verify this flag has not been inadvertently set to False and that the inputs are genuinely unchanged. Changing the datastore storage tier affects read/write throughput but has no effect on whether AzureML decides to re-execute a step; caching decisions are independent of storage performance. add_dependency() controls execution order (ensuring steps run sequentially rather than in parallel) but does not enable output caching; it is used to express data flow dependencies, not to skip re-execution. Manually copying outputs and hardcoding dataset versions is a fragile, manual workaround that defeats the purpose of a managed pipeline and does not leverage AzureML's built-in caching mechanism.