Changelog

All notable changes to simmetry will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.2.0] - 2026-06-05

hamming_str — normalized Hamming similarity for equal-length strings. Registered as a string metric.
bm25 — BM25 relevance score in [0, 1]. Asymmetric: measures how well a document satisfies a query. Registered as a string metric.
pairwise() now dispatches on input type: pass a list of strings → pairwise_strings, a list of tuples → pairwise_points, or an ndarray → pairwise_numpy. No manual import needed.
available("string") now lists hamming_str and bm25.
Module docstrings on api.py, registry.py, and index.py.
pytest-cov added to [dev] extras.
ANN backend tests (auto-skip if hnswlib / faiss not installed).
Custom metric registration test.
NaN / inf propagation tests (88 passing, 2 skipped).

haversine_sim(a, b, scale_km=...) — geographic similarity in [0, 1] registered as the haversine_sim metric.
hamming — normalized Hamming similarity for equal-length sequences. Registered as a vector metric.
cosine_distance — 1 - cosine, registered as a vector metric for sklearn-compatible workflows.
tversky — Tversky index for sets: |A∩B| / (|A∩B| + α|A\B| + β|B\A|). Defaults (α=β=1) equal Jaccard; α=β=0.5 equals Dice.
pairwise() now supports hamming and cosine_distance.
SimIndex.query() now converts ANN backend distances to similarities consistently.
Field presence validation in similarity() for composite dict records.
[ann] extra now installs both hnswlib and faiss-cpu.
CI test matrix covering Python 3.10–3.13 and NumPy 1.x / 2.x.
GitHub Pages documentation site.

Breaking: haversine_km is no longer a registered metric. Use haversine_sim as the metric; import haversine_km directly for raw km values.
Breaking: Auto-metric for 2-number tuples now returns haversine_sim. _is_point_like restricted to tuples (not lists) within valid lat/lon ranges.
pairwise_points and topk_points default metric changed from haversine_km to haversine_sim.
Package description, Homepage URL, and classifier updated in pyproject.toml.
Version bumped to 1.1.0.

haversine_km now returns geographic distance in kilometers instead of a normalized score.
topk_points(..., metric="haversine_km") ranks by nearest distance first.

Optional Numba acceleration for euclidean_sim / manhattan_sim via simmetry[fast].

Initial release: auto similarity, string/vector/point/set metrics, ANN support, composite records.