引導式測驗 · 中等 等級
Azure Machine Learning 測驗
了解 Azure ML 工作區、運算、作業、管道和端點,同時檢查身分、成本、配額和評估實務。
問題 1 的 8
Which description best captures an Azure ML workspace?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- Which description best captures an Azure ML workspace?
- A service that automatically runs every training job without configured compute
- A project scope that groups jobs, data references, models, and endpoint resources
- A storage account that holds only registered model binaries
- A compute cluster that executes one job at a time
- Which workload can an autoscaling Azure ML compute cluster support?
- Serving synchronous HTTP requests from an online deployment
- Running submitted jobs on allocated nodes that can scale by configuration
- Providing an interactive notebook environment on a persistent instance
- Registering model metadata without allocating compute
- Which endpoint is intended for synchronous real-time scoring?
- A batch endpoint
- A compute instance
- A registered data asset
- An online endpoint
- What does a pipeline connect?
- A set of reusable workflow steps that pass inputs and outputs
- A collection of workspace user accounts
- A list of deployed endpoint URLs
- A model registry with no executable steps
- Why use a managed identity for a job?
- To grant all workspace users owner access
- To remove the need to assign storage permissions
- To encrypt every training label without configuration
- To give a job scoped service identity without embedding a personal credential
- What affects Azure ML compute cost?
- Only offline accuracy and model size
- Machine type, region, duration, storage, and endpoint uptime
- Only the number of steps in a pipeline
- Only the workspace display name
- What should be done before adding complex pipelines to a new project?
- Deploy every model version as soon as training ends
- Tune repeatedly on the held-out test set
- Grant broad owner access to all jobs before testing
- Run a small job and validate its data, environment, and outputs
- Which statement about a model registry is accurate?
- It tracks model versions but does not replace release review
- It certifies that a model is safe for every deployment
- It automatically detects and repairs data drift
- It removes the need to version training data