AI-200 Practice Questions & Answers

Loading your practice set…

In designing a machine learning training pipeline using Azure Machine Learning Workspaces, you need to ensure that your data is stored in a secure and accessible location. Which Azure service should you use to store your training dataset, and how should it be configured to ensure secure access and efficient data loading into the compute cluster?

Azure Files

Azure Blob Storage with SAS tokens (Correct)

Azure Table Storage

Azure Queue Storage

Explanation: The correct answer is to use Azure Blob Storage with SAS (Shared Access Signature) tokens for secure and controlled access. Azure Blob Storage provides a scalable and durable storage solution for large datasets. Using SAS tokens allows you to grant temporary access to specific users or services while maintaining security. Other options like Azure Files or Azure Table Storage may not offer the same level of scalability and performance for large datasets.

When setting up a compute cluster for training a machine learning model in Azure Machine Learning Workspaces, which type of compute cluster should you choose if you require high-performance computing resources and want to leverage GPU acceleration for faster training times?

CPU-only cluster

Azure Machine Learning Compute Cluster with GPU instances (Correct)

Virtual Machine without GPU support

Azure Batch

Explanation: The correct answer is to choose an Azure Machine Learning Compute Cluster with GPU instances. This type of compute cluster is specifically designed for high-performance computing tasks and can take advantage of GPU acceleration to speed up training times. Other options like CPU-only clusters or virtual machines without GPU support would not provide the necessary performance for GPU-accelerated training.

In an Azure Machine Learning project, you need to optimize the hyperparameters of a machine learning model using HyperDrive. Which of the following best describes how you would set up the HyperDrive configuration to achieve this goal?

Define a fixed set of hyperparameters and manually run multiple training jobs.

Define a search space for hyperparameters and configure HyperDrive to perform automated hyperparameter tuning. (Correct)

Use a single set of hyperparameters and rely on the model's default settings.

Create a custom environment container and use it to run training jobs with different hyperparameters.

Explanation: The correct answer is B because HyperDrive in Azure Machine Learning allows you to define a search space for hyperparameters and automatically runs multiple training jobs with different hyperparameter values to find the optimal set. Option A is incorrect because it refers to a manual process which does not leverage automation. Option C is incorrect as it suggests using a fixed set of hyperparameters, which defeats the purpose of hyperparameter tuning. Option D is incorrect because it mentions a custom environment container, which is not directly related to setting up HyperDrive for hyperparameter tuning.

You are working on a machine learning project and want to track the performance metrics of your experiments using MLflow. How would you integrate MLflow into your Azure Machine Learning pipeline to ensure that all experiment runs are properly logged and tracked?

Use a different logging system that is not compatible with Azure Machine Learning.

Install the MLflow Python package and use its API to log metrics, parameters, and artifacts from your training scripts. (Correct)

Disable logging since MLflow is not necessary for tracking experiments in Azure Machine Learning.

Create a custom environment container and use it to run training jobs with MLflow tracking enabled.

Explanation: The correct answer is B because MLflow can be integrated into an Azure Machine Learning pipeline by using the MLflow SDK to log metrics, parameters, and artifacts from your training scripts. Option A is incorrect because it suggests using a separate logging system without integrating with MLflow. Option C is incorrect as it implies that MLflow cannot be used for tracking, which is false. Option D is incorrect because it mentions a custom environment container, which is not directly related to integrating MLflow for tracking.

You have deployed a machine learning model using Azure Machine Learning and need to manage traffic between two versions of the model to gradually roll out updates without downtime. Which method should you use to achieve this blue-green deployment?

Use Azure Kubernetes Service (AKS) to manage traffic.

Use Azure Load Balancer to manage traffic between the two versions of the model. (Correct)

Deploy the models in separate virtual machines and manually switch the load.

Use Azure Functions to route traffic based on the version of the model.

Explanation: Blue-green deployment is a strategy where two identical production environments (blue and green) exist at the same time. Traffic can be switched between them without any downtime. In Azure ML, you can create two online endpoints, each pointing to a different version of the model, and then switch traffic between them using Azure Load Balancer or Azure Application Gateway. Option B is correct because it describes the use of Azure Load Balancer to manage traffic between the two versions of the model during a blue-green deployment. Options A, C, and D are incorrect as they do not describe the proper method for managing traffic in a blue-green deployment scenario.

Your organization wants to deploy a machine learning model that will process data in batches rather than continuously. Which type of endpoint should you choose in Azure Machine Learning, and what additional configuration might be necessary?

Choose an online endpoint and configure it to process data continuously.

Choose a batch endpoint and configure it to process data in batches. (Correct)

Select a web app and deploy the model there, configuring it to handle batch requests.

Deploy the model as a function app and set up triggers for batch processing.

Explanation: Batch endpoints in Azure Machine Learning are designed to handle data processing in batches, which means the model is invoked only when new data is available. You would select a batch endpoint for this scenario. Additional configuration might include setting up the scoring script to handle batch requests and configuring the endpoint to trigger on new data. Option B is correct because it correctly identifies the batch endpoint as the appropriate choice for processing data in batches. Options A, C, and D are incorrect as they do not align with the requirements for batch processing.