Experiments

The Experiments tab (Cmd+3) is where your training history lives. Every completed training run is saved as an experiment that you can compare, tag, and revisit.

Experiment Table

The main view is a sortable table showing all experiments. Each row displays:

• Formula — the LaTeX expression used
• Dataset — which dataset was trained on
• Metrics — key performance numbers (accuracy, MSE, etc.)
• Project — which workspace the experiment belongs to
• Tags — custom labels for organization
• Timestamp — when the training completed

Expanding Details

Click any row to expand it and see the full experiment detail:

• All metrics (not just the summary ones)
• Hyperparameters used (learning rate, epochs, optimizer, etc.)
• Dataset info (rows, columns, target variable)
• Transform summary — amber chips showing what data transforms were applied

Comparing Experiments

Select two or more experiments using the checkboxes, then the comparison view activates:

• Loss curve overlay — Plotly chart showing all selected experiments' loss curves on one graph
• Metric bar charts — side-by-side comparison of key metrics

Tags

Add tags to experiments for organization. Tags are editable inline — click the tag area in the expanded detail row. A typeahead suggests existing tags from your workspace.

Use the tag filter bar at the top of the experiments table to filter by tag.

Cross-Project Experiments

Experiments are stored globally, not per-project. The project column and blue left border indicate which project each experiment came from. This lets you:

• Compare results across different projects
• Find the best approach regardless of which workspace you tried it in

Open in Canvas

Click Open in Canvas on any experiment to load its formula and dataset back onto the canvas. This lets you iterate on a previous experiment without re-entering the formula.

If the experiment is from a different project, the button shows Import instead — it copies the formula to your current workspace.