In Azure ML Studio, how to change Environment to register model as MLmodel?

Question

Using Azure Machine Learning Studio, I ran a job that generated a regression model. I want to use the function 'Responsible AI' to do error analysis.

The problem: I cannot register the model type as 'MLflow', because 'Output must be a folder containing an mlflow 'MLmodel' file.'

I figured that this is probably because my Environment (AzureML-Designer version 54) does not have the mlflow loaded. (Is this really the cause of my problem?)

So, I went to look how to change the Environment (to sklearn-1.1 version 26 for example). I have read the documentation and checked the entire Azure ML Studio Graphical User Interface for the setting. However, I cannot figure out how to fix the issue. Your help is very welcome!

If your answer includes the usage of SDK or CLI, please also explain why it cannot be done in the GUI.

Answer 1

To obtain an mlflow model, the training script used in the pipeline component should resemble this script. However, in Azure ML Designer's Classic prebuilt training component, the script doesn't log an mlflow model.

It's not that it only supports different environments, but the script used for training should use mlflow to track and log the models.

If you have a custom training script, use the Python SDK with a command job to train and log the model. Check this notebook for more information.

According to this documentation, mlflow is only supported in Azure CLI and Python SDK.

The same SDK is used in the training script while building pipeline components.

If you want prebuilt components that log mlflow models, consider Azure ML Designer v2.

You can find the training component in the components tab, which you can select for training.