Contents Menu Expand Light mode Dark mode Auto light/dark, in light mode Auto light/dark, in dark mode Skip to content
Starfold manual
Starfold manual

Start

  • What Starfold does
  • Install Starfold
  • Your first run

Concepts

  • Concepts
    • UMAP: a map of your table
    • HDBSCAN: dense clumps and noise
    • Optuna: searching the settings
    • How Starfold chains the tools
    • Copies and gaps
    • Refits: reproducibility, not truth

Tutorials

  • Tutorials
    • Tutorial: groups in the scikit-learn digits dataset, part 1
    • Tutorial: groups in the scikit-learn digits dataset, part 2
    • Tutorial: two groups in the APOGEE DR19 example table, part 1
    • Tutorial: two groups in the APOGEE DR19 example table, part 2

Metrics

  • Metrics
    • Prediction agreement
    • Coverage
    • Modularity
    • Score
    • Gap
    • Default comparison
    • Group recurrence
    • Support and own support
    • Unresolved mass
    • Count interval
    • Co-association, contrast and the consensus check
    • Calibration p-value
    • HDBSCAN strengths
    • Group properties and their bands

How-to guides

  • How-to guides
    • Prepare your table for recover
    • Choose between standard and robust scaling
    • Narrow the search space
    • Pass measurement errors to recover
    • Set the budgets and run on a laptop
    • Run Starfold on a GPU
    • Run fits in parallel with n_jobs
    • Reproduce a run
    • Run the calibration test
    • Save a result and load it back
    • Compare the CPU and cuML partitions
    • Check a result against known labels
    • Run a seed sweep on a SLURM cluster
    • Plot your result
    • Report your result in a paper

Reference

  • Reference
    • API reference
    • Configuration fields
    • Result fields
    • Flags
    • Saved result files
  • Troubleshooting
  • What Starfold does not claim
  • Learn more
  • Citing Starfold
  • Glossary
Back to top
Copyright © 2026, Andreas W. Neitzel
Made with Sphinx and @pradyunsg's Furo