Map of GitHub

Original Article ↗ Hacker News Discussion ↗

An interactive “map of GitHub” turns millions of repositories and their shared stargazers into a world of islands and countries, where proximity reflects how often the same users star different projects. Commenters explore what the layout reveals about language ecosystems (from Rustland and Lispaña to Pythonia and PHP kingdoms), as well as odd placements that highlight gaps between aspirational interest and real usage. Many praise the visualization’s creativity and performance while noting methodological quirks of using Jaccard similarity on stars—especially how celebrity projects and over-active starrers can distort perceived relationships.

Overall Reception

  • Many commenters find the visualization “phenomenal,” “artful,” and surprisingly usable and fast, even on mobile.
  • The playful country names (e.g., “Lispaña,” “Sussex,” “Homelabia,” “Quitlessia,” “The GitHub Archipelago”) are widely enjoyed and become a running joke.
  • Some treat it as a game: trying to locate specific projects without search or “sailing” from one project to another via paths.

Data Source, Similarity, and Layout

  • Repos are positioned based on overlapping stargazers. Dots are close if they share many stargazers.
  • Edges between repos are derived from a similarity metric, primarily Jaccard similarity over star sets, with a threshold to decide which edges exist (exact threshold not specified).
  • Lines only appear when zoomed into a region.
  • Popular “celebrity” projects tend to cluster together due to generic popularity rather than semantic similarity; commenters note this as a known limitation.
  • Suggestions include using TF–IDF over the user–star matrix to downweight “overstarring” users, or code embeddings, though resource costs are questioned.
  • The author experimented with multiple similarity metrics and chose Jaccard subjectively as “best” for this use.
  • Clustering uses community-detection–style algorithms (Louvain/Leiden plus custom methods). Hierarchical clustering ideas (e.g., HDBSCAN) ran into memory issues at this scale.

Interpretation Quirks and Ecosystem Insights

  • Several projects appear in “unexpected” lands (e.g., Linux near frontend/awesome lists, HTMX in Djangonia, Django in Pythonia, MicroPython/CircuitPython placement, Magisk forks in different regions).
  • Explanations offered:
    • Users star surrounding ecosystem projects more than core ones (e.g., Linux kernel, Django).
    • “Aspirational” star patterns (e.g., people star Julia alongside Python ML/AI projects without fully moving ecosystems).
    • Overlap of interest communities (e.g., crypto with AI).
  • Some observe smaller-than-expected regions for Rust, Node, or Azure, and very large ones for JavaScript, YAML/DevOps, Python/AI, Vim/Emacs.
  • One hypothesis: ecosystems with lower friction to publishing packages (e.g., JavaScript) yield larger “islands.”
  • PHP’s prominent “kingdom” is noted as evidence it remains widely used and actively developed.

Critiques of the Map Metaphor and Stars

  • Some question the country/map metaphor and fuzzy region names; they propose hierarchical cluster diagrams with clearer labels.
  • Others appreciate that it is “just a view, not a thesis,” and like the personality over stricter analytical clarity.
  • Skeptics note that stars can be noisy or gamed (bots, vanity projects), so importance and quality are not faithfully represented.