/compare

name: compare
description: Compare multiple ML experiment runs side-by-side to identify the best configuration.

Compare multiple ML experiment runs side-by-side to identify the best configuration.

Steps

1. Load experiment records from the tracking store. 2. Select experiments to compare: 3. Build a comparison table: 4. Analyze parameter sensitivity: 5. Generate visualizations: 6. Identify the winning configuration: 7. Recommend next experiments to try.

Format

Comparison: <N> experiments
Best Run: <experiment name>
Key Findings:
  - <parameter X> has <impact> on <metric Y>

Rules

• Only compare experiments with the same dataset version.
• Use consistent metrics across all compared runs.
• Statistical significance matters; do not draw conclusions from single runs.