Automated ML (Machine Learning) AMA (Ask Microsoft Anything)

Hi Nick, AutoML Python SDK uses completely new Python packages, and the way you do it is by installing in a Conda environment the Azure ML Python SDK v2 packages, explained here:

https://github.com/Azure/azureml-examples/tree/main/sdk

Basically, it's simply this pip command on top of your activated conda environment:

pip install --pre azure-ai-ml

Gotchas / new things:

- By default, models in v2 are saved as MLFlow models (MLFlow model) but you can also deploy it as a regular .pkl file model as in v1. In reality, the models created by AutoML in v1 and v2 are "the same" because AutoML backend is the same. What's new in v2 is the Python SDK and CLI.

- CLI is completely new for AutoML, since AutoML was not part of Azure ML CLI v1, previously.

- Instead of using Tabular Datasets (v1) you use MLTable (v2) which you currently define in a .YAML file pointing to your data (i.e. a .CSV file). In the next Public Preview you will also have Python functions to create an MLTable from your data files using "data plane" functions for MLTable.