min_max_normalize (retrieval-pipeline/fusion.py)

Fusion utilities for hybrid retrieval, implementing RRF and weighted score fusion.

# Source: https://github.com/bojieli/ai-agent-book | License: Apache-2.0 | Copyright You must retain, in the Source form of any Derivative Works
# Provided via justclaw.it.com; keep this notice with the code.
"""Result fusion for hybrid retrieval.

This module implements the *fusion* stage of the hybrid retrieval pipeline —
the step that merges the separately-ranked dense and sparse candidate lists into
a single, unified candidate pool before neural reranking.

Two production-grade fusion strategies are provided, matching the two approaches
discussed in the book (第3章「混合检索流水线」):

1. Reciprocal Rank Fusion (RRF)
   score(d) = Σ_r  1 / (k + rank_r(d))
   Only ranks are used, original scores are discarded. Robust and scale-free,
   because it never has to compare a cosine similarity against a 

... (truncated -- full source via MCP)