Azure Machine Learning - Generally availability updates for March 2023

New features now available in GA include the ability to visualize timeseries models accurately over time, and deploy a compute cluster or compute instance without any public IP addresses. Additionally, you can now deploy your models to batch endpoints and run them on top of your Kubernetes clusters, and control and restrict data access so that sensitive data can only be accessed when working on compute instance.

Visualize forecasting horizon metrics for AutoML - You can now plot the relationship between the model’s predicted value and its actual value mapped over time. You can also choose which validation fold (up to 5) and which time series (up to 20) to visualize the results. 

Secure your compute cluster and Compute Instance with No Public IP - You can now provision your compute cluster and instance without any public IP addresses. This will help you access simpler network configuration without inbound access requirements, prevent data exfiltration, and lower costs by removing the network resources requirement. 

Kubernetes support for Batch Endpoints - You can now retain control of inferencing infrastructure using a Kubernetes cluster with batch endpoints. This functionality is operational for all features of batch endpoints regardless of the compute type. 

Create compute instance with Managed Identity - You can now provision compute instances with managed identity enabled, which can then be used to permit access to data storage. This will allow admins to restrict user access to sensitive data so that it can only be accessed when working on compute instance.