larzsearch

Full-text search with real relevance (BM25). Pure Python, zero dependencies.

Index documents, then search them ranked by BM25 — the relevance model behind Lucene and Elasticsearch — not just "contains the word". Rarer query terms count for more, term frequency has diminishing returns, and long documents don't win by length. Optional fuzzy matching tolerates typos.

from larzsearch import SearchIndex

idx = SearchIndex()
idx.add("a", "the quick brown fox jumps")
idx.add("b", "the lazy brown dog sleeps")

idx.search("brown fox")            # [{'id': 'a', 'score': .., 'doc': ..}, ...]
idx.search("quik fox", fuzzy=True) # matches despite the typo

Why

• Actual ranking, not filtering. BM25 with IDF weighting and length normalization gives you results ordered by relevance — the thing naive substring search can't do.
• Typo-tolerant. fuzzy=True matches query terms within one edit, so "pythou" finds "python".
• Documents your way. Index plain strings or dicts (all values are indexed); the original doc comes back with each result.
• Zero dependencies. No Elasticsearch, no Whoosh, no C extensions — great for in-app search, docs sites, CLIs, and pairing with larzdb.

Install

pip install larzsearch

Usage

idx = SearchIndex(k1=1.5, b=0.75)         # tune BM25 if you like
idx.add("doc1", "text ...")
idx.add("doc2", {"title": "...", "body": "..."})
idx.add_many([{"id": 1, "text": "..."}, ...])
idx.remove("doc1"); idx.get("doc2"); len(idx)

results = idx.search("query terms", limit=10, fuzzy=False)
# [{"id": ..., "score": float, "doc": original}, ...]  (best first)

Tests

python -m unittest discover -s tests -v   # 15 tests incl. ranking + fuzzy

The Larz stack

One of 30+ pure-Python, zero-dependency libraries at github.com/larz-scripter.

License

MIT © larz-scripter