Experiments

Track your training runs, with their parameters, metrics and artifacts, and compare them side by side without leaving the editor. It runs a real MLflow server for you and shows everything in the Experiments panel.

What you can do

• Install MLflow and start a local tracking server with one click, with your tracking data stored alongside your project
• Write ordinary mlflow code in a notebook cell and see the run appear, with no configuration
• Open any run to read its parameters, metric charts and artifacts
• Tick two or more runs to compare them with a parameter diff, overlaid metric charts, a parallel-coordinates plot and a metric-vs-metric scatter
• See your registered models and which version each alias points at, and promote a version to champion
• Scaffold the train.py and evaluate.py a pipeline calls, leaving the model and features to you
• Connect an existing MLflow server instead of the local one

Requirements

• A Python environment. MLflow installs into your active one, or into a new .venv in the workspace if you have none.
• The desktop app, for the metric charts.

Getting started

Open Experiments from the activity bar. If MLflow is not installed yet, click the MLflow item in the status bar and press Install MLflow. With the server running, train from a notebook cell that calls mlflow.autolog(), and the run appears in the panel.

To use a server you already have, press the + button in the panel (Connect a remote MLflow server) and enter its address and sign-in.

Good to know

• The local server runs without authentication, as local MLflow usually does. It survives editor reloads and shuts down with the app.
• A remote server's token is kept as a secret, not in settings. Artifacts on a remote server are read-only here, and deleting a run uses MLflow's soft-delete, so it can be recovered.
• Models are stored as versioned artifacts. The extension does not run them.
• Promoting a version asks you to confirm, records your answer, and goes through the approval rules. The record says a person at this workstation approved, not which person.
• The scaffold keeps a file that already exists rather than overwriting it.
• On Linux with Docker, the container must be able to reach your host machine for the automatic kernel wiring to work.