Class Introduction

Function

Re-ranks BM25 and dense vector hybrid retrieval results. It inherits the abstract class Reranker.

Prototype

from mx_rag.reranker.local import MixRetrieveReranker
MixRetrieveReranker(k, baseline, amplitude, slope, midpoint)

Parameters

Parameter	Data Type	Required/Optional	Description
k	Integer	Optional	The most relevant k results after re-ranking. The value range is [1, 10000]. The default value is 100.
baseline	Float	Optional	Basic value of the weight calculation formula. The value range is [0.0, 1.0]. The default value is 0.4.
amplitude	Float	Optional	Amplitude of the weight calculation formula. The value range is [0.0, 1.0]. The default value is 0.3.
slope	Float	Optional	Transition steepness. The value is greater than 0. The default value is 1.
midpoint	Float	Optional	Query length when the function value reaches the middle point between the basic value and the maximum value. The value is greater than 0. The default value is 6.

Note: The weight calculation formula is as follows. Ensure that the calculation result is in the range of [0, 1].

Return Value

MixRetrieveReranker object.

Example

from langchain_core.documents import Document
from mx_rag.reranker.local import MixRetrieveReranker

doc_1 = Document(
    page_content="document1",
    metadata={
        "score": 1.0,
        "retrieval_type": "dense",
    }
)
doc_2 = Document(
    page_content="document2",
    metadata={
        "score": 2.0,
        "retrieval_type": "sparse",
    }
)
docs = [doc_1, doc_2]
query = "This is a question."
reranker = MixRetrieveReranker(k=100, baseline=0.4, amplitude=0.3, slope=1, midpoint=6)
res = reranker.rerank(query, docs)

Parent topic: MixRetrieveReranker