DP-100 Practice Questions

For forecasting tasks, AutoML has powerful built-in time-series featurization capabilities. By correctly identifying the time column in the forecasting settings and leaving featurization on 'auto' (the default), AutoML will automatically generate a rich set of features from the date/time column, such as year, month, day, day of week, week of year, and more.

The semantic ranker (also known as semantic search) is a premium feature in Azure AI Search that uses deep learning models to understand the contextual meaning and intent behind a query. It re-ranks the initial set of results from keyword or vector search based on semantic relevance, significantly improving the quality of results for RAG applications.

An OutOfMemoryError in a batch scoring job typically occurs when the amount of data being processed in a single batch exceeds the available RAM on the worker node. The mini_batch_size parameter controls how many files or records are processed at once. By reducing this value, less data is loaded into memory simultaneously, which should resolve the memory issue without needing to change the compute hardware.

The mlflow.search_runs function is used to query past runs. It allows you to filter by experiment ID and order the results by a specific metric (e.g., 'metrics.rmse' in ascending order) to find the best performing run.

Once the best run is identified, the mlflow.register_model function is used to take the model artifact from that run (specified by its run URI) and register it in the MLflow Model Registry with a given name.

Fine-tuning is the ideal approach when the goal is to teach a model a new skill, style, or format that is consistent across many interactions. By training the model on a large, high-quality dataset of expert conversations, you directly adjust the model's weights to adopt the specific brand voice and response patterns demonstrated in the logs. RAG is for providing knowledge, and prompt engineering is less effective for instilling a consistent, nuanced style.

Azure ML data assets support versioning. You can create new versions as the underlying data changes. By referencing the asset in your script using the @latest tag (e.g., azureml:my_data@latest), you ensure that the job always resolves to the most recently created version of that asset, achieving the goal without code modifications.

Prompt flow is specifically designed for this type of complex LLM orchestration. Its visual graph allows developers to clearly define and visualize the flow of logic, including conditional branches. It has native support for Python tools to encapsulate API calls, LLM nodes for prompt execution, and a secure 'Connections' system to manage secrets like API keys. The built-in tracing and debugging capabilities would directly address the team's difficulties in understanding and troubleshooting the conversational flow.

A recursive character text splitter is ideal for documents with varied structure. It attempts to split along semantic boundaries (paragraphs, sections) first before falling back to smaller units. This preserves the context within chunks, which is crucial for accurate retrieval from both short policies and long guides. A moderate chunk size with overlap ensures that sentences or ideas are not awkwardly split between chunks.

Azure Machine Learning components can be configured with retry settings directly in their YAML definition. By adding a retry_settings block with properties like count, delay, and backoff, you instruct the pipeline orchestrator to automatically re-run the component if it fails. This is the correct approach for handling transient errors without altering the component's source code.

Creating separate workspaces provides the strongest isolation boundary. Each team can manage its own compute, data, and experiments independently, satisfying the requirement that they cannot access each other's resources.

An Azure Machine Learning registry is designed for sharing assets like models, environments, and components across multiple workspaces. This allows a central MLOps team to manage and distribute curated assets to all the individual team workspaces.