How Iterative Studio Works

Iterative Studio works with the data, metrics, hyperparameters and model metadata that you add to your ML project Git repositories. It works very closely with your Git ecosystem.

This video illustrates how Iterative Studio works closely with your Git ecosystem.

How your project data is saved

• Using DVC and Git, you will push all your ML experiments to your GitHub, GitLab or Bitbucket repositories as Git commits.
• Using Iterative Studio, or using the command line interface (CLI) of GTO, and possibly MLEM, you will push all your ML model details to the Git repositories as Git commits and Git tags.

How Iterative Studio extracts your project data

• When you connect to these Git repositories from Iterative Studio, the project's dvc.yaml is used to identify all the data, metrics and hyperparameters in your experiments.
• If you are not using DVC, you can add the metrics and hyperparameters to your Git repositories manually.
• Details of your ML models, including versions and stage assignments, are extracted from the Git commits and tags.

How Iterative Studio presents your project data

• Iterative Studio creates an interactive, tabular representation of all the identified values.
• All the projects that you have created are presented in a central projects dashboard.
• All identified models are included in an interactive models dashboard and individual model details pages.

How Iterative Studio saves updates to your ML projects

• When you run new experiments or add models to your model registry in Iterative Studio, it creates Git commits and pull requests with the changes.
• You can set up your CI/CD actions (e.g. GitHub Actions) to run model training upon the creation of Git commits, tags or pull requests. You can use CML in your CI/CD actions for continuous machine learning.
• When you register new versions of your ML models or assign stages to them, Iterative Studio creates annotated Git tags representing the actions.