AI-900 Question 136
Select 3You are a data scientist working on deploying a machine learning model using Azure Machine Learning. You want to ensure that the deployed model can handle real-time predictions and that you can monitor its performance after deployment. Which of the following Azure Machine Learning capabilities should you use?
- A
Deploy the model as an Azure Kubernetes Service (AKS) endpoint for real-time inference.
- B
Use Azure Application Insights to monitor the deployed model's performance and telemetry.
- C
Deploy the model as a batch inference pipeline for handling real-time predictions.
- D
Enable Azure Machine Learning Model Registry to store and version your model.
- E
Configure autoscaling for the deployed endpoint to handle varying traffic loads.
Show answer and explanation
Correct answers: A, B, E
Explanation
To meet the requirements of real-time predictions and performance monitoring, the model should be deployed as an AKS endpoint, which supports real-time inference. Azure Application Insights can monitor the deployed model, providing visibility into its performance. Additionally, configuring autoscaling ensures that the endpoint can handle varying workloads efficiently. Batch inference pipelines and the Model Registry, while useful in other aspects of model management, do not fulfill the specific needs of this scenario.
- A. Correct.
Deploying the model as an Azure Kubernetes Service (AKS) endpoint allows for handling real-time predictions, which is a requirement in this scenario.
- B. Correct.
Azure Application Insights can be used to monitor the deployed model's performance, such as latency and telemetry data, making it essential for monitoring after deployment.
- C. Incorrect.
Batch inference pipelines are designed for processing large amounts of data at once and are not suitable for real-time predictions required in this scenario.
- D. Incorrect.
While the Azure Machine Learning Model Registry is crucial for storing and versioning models, it does not directly address the requirements of real-time predictions or monitoring performance after deployment.
- E. Correct.
Configuring autoscaling for the deployed endpoint ensures it can handle varying traffic loads, which is important for maintaining performance during real-time inference.