Question

You are developing a real-time scoring solution using an Azure Machine Learning managed online endpoint. Before deploying to the cloud, you want to validate that your scoring script and environment work correctly by running the deployment container on your local development machine, catching errors quickly without provisioning cloud compute. Which approach lets you accomplish this?

Accepted Answer

Create the deployment with the `--local` flag using `az ml online-deployment create`, then invoke it with `az ml online-endpoint invoke --local`.

Answer

Deploy to a Kubernetes attached compute target and forward requests through an SSH tunnel to your workstation.

Answer

Set the deployment's instance count to zero so the container runs on the compute instance instead of managed compute.

Answer

Register the model in the workspace registry and use the built-in test tab in Azure Machine Learning studio.