DP-100 Question #267: Real Exam Question with Answer & Explanation

Question

Drag and Drop Question You are designing an Azure Machine Learning solution. The model must be trained by using automated machine learning. The compute must be a shared resource with users in the Azure Machine Learning workspace. After you train the model, it must be deployed for batch scoring on a serverless compute. You need to select the appropriate computation options for the solution. Which compute options should you select for training and deployment? To answer, move the appropriate compute options to the correct project activities. You may use each compute option once, more than once, or not at all. You may need to move the split bar between panes or scroll to view content. NOTE: Each correct selection is worth one point. Answer:

Explanation

The question requires identifying the correct Azure Machine Learning compute options for model training on a shared resource and model deployment for serverless batch scoring.

Approach. To answer correctly, the test-taker must drag 'Azure Machine Learning compute cluster' to the 'Train the model' slot and 'Azure Machine Learning serverless compute' to the 'Deploy the model' slot.

• Train the model: The scenario states that the model must be trained using automated machine learning on a 'shared resource with users in the Azure Machine Learning workspace'. An Azure Machine Learning compute cluster is a managed, scalable pool of virtual machines designed for training machine learning models, including automated ML, and can be shared by multiple users within a workspace. This option perfectly matches the training requirements.
• Deploy the model: The scenario specifies that the model must be deployed for 'batch scoring on a serverless compute'. Azure Machine Learning serverless compute refers to the managed, on-demand compute infrastructure that powers Azure ML batch endpoints, providing a serverless experience for batch inference. This option directly fulfills the requirement for 'serverless compute' for batch scoring.

Common mistakes.

• common_mistake. A common mistake, as illustrated in the first image, is to swap 'Azure Machine Learning serverless compute' and 'Azure Machine Learning compute cluster'.
• Placing 'Azure Machine Learning serverless compute' for 'Train the model': Serverless compute in Azure ML is primarily designed for inference (like batch scoring) or interactive development (compute instance), not for large-scale, shared model training, especially for automated machine learning, which typically requires robust, scalable compute clusters.
• Placing 'Azure Machine Learning compute cluster' for 'Deploy the model': Compute clusters are optimized for training workloads and are not the appropriate serverless compute type for batch scoring deployments. Deploying for 'batch scoring on a serverless compute' points to Azure ML's managed batch compute offerings, which are serverless.
• 'Azure Machine Learning endpoints': While an Azure Machine Learning batch endpoint would be used for batch scoring, the question asks for the 'compute option', and 'endpoints' are the deployment construct, not the underlying compute resource type itself. Batch endpoints utilize serverless compute.
• 'Azure Machine Learning Kubernetes': This option is for deploying models to Azure Kubernetes Service (AKS) or Arc-enabled Kubernetes clusters. While Kubernetes can host batch deployments, it is not inherently 'serverless' in the context of Azure ML's managed serverless batch compute. It offers more control but requires managing the Kubernetes cluster itself, which goes against the 'serverless compute' requirement.

Concept tested. Understanding the different Azure Machine Learning compute targets, their specific use cases (training vs. deployment), and the appropriate compute types for shared training resources (compute cluster) and serverless batch scoring (serverless compute).

No community discussion yet for this question.