Changelog

All notable changes to VoronoiMap are documented here.

[1.0.0] — 2026-02-14

Fixed

• Statistical bias in get_sum() — Zero-area degenerate regions no longer corrupt the estimation average. The old code decremented N but left zero entries in the summation array at wrong indices, causing systematic downward bias. Now uses a clean valid-estimates-only list. (Fixes #14)

Changed

• O(1) KDTree lookup in get_NN() — Replaced O(n) identity scan over _data_cache with _kdtree_by_id dictionary keyed by id(data). This runs on every NN query, so the speedup is meaningful for large datasets with many queries.
• Cleaned up new_dir() — Removed dead variable assignments (c1x, c1y, c2x, c2y) that were set to magic numbers but never read.

Added

• Comprehensive MkDocs documentation site with:
• Installation guide with dependency groups
• Quick start tutorial
• Full CLI reference
• Python API documentation for all public functions
• Algorithm overview with mathematical notation
• Implementation details and architecture docs
• Regression tests for the get_sum() bias fix
• Test for _kdtree_by_id cache population

---

[0.9.0] — 2026-02-13

Added

• CI/CD workflow with lint, test, and coverage stages
• GitHub Pages deployment for interactive demo
• Copilot agent setup (.github/copilot-setup-steps.yml)
• CodeQL security scanning
• Dependabot configuration
• Docker workflow
• Branch protection rules
• Professional README with badges, API docs, and project structure

Fixed

• Path traversal vulnerability in load_data() (prevents ../../etc/passwd)
• NaN/Inf coordinate rejection
• Near-parallel line intersection producing wildly large coordinates
• Binary search convergence guard (100 iteration limit)
• Voronoi vertex count limit (MAX_VERTICES = 50)

Changed

• KDTree-accelerated nearest neighbor (O(log n) with SciPy)
• Auto-detect search bounds from data extents
• Cross-product collinearity test (replaces fragile slope comparison)
• Iterative retry loop in get_sum() (replaces recursive calls)