IssueTracker

I think the title is self-exploratory. I have started using AutoML for Vision tasks, and despite things I would wish to be there and improve which I am confident they will, I would rather to see model versioning as a built-in feature when training models. At the moment, one can see tiles of models with individual detailed evaluation. However, with increasing number of trained models with changes of model architecture, accuracy, precision, recalls. input data etc., it is getting harder to have overview of model versions and choose the best. There are offline solutions open-sourced like MLflow. That would be an awesome feature to have such functionality integrated in AutoML for insights into a collection of trained models.

Problem you have encountered:

Lack of model versioning

What you expected to happen:

Built-in model versioning like MLflow (https://github.com/mlflow/mlflow)