Principle:Marker Inc Korea AutoRAG Strategy Based Module Selection

Knowledge Sources	AutoRAG Docs
Domains	Optimization, Evaluation
Last Updated	2026-02-08 06:00 GMT

Overview

A multi-criteria optimization technique that selects the best RAG module based on evaluation metrics, speed, and configurable selection strategies.

Description

Strategy-Based Module Selection is the decision engine at the heart of AutoRAG's optimization. After evaluating all module combinations within a node, the strategy selects the best module based on multiple metric columns. Three selection strategies are available: mean (simple average across metrics), rank (reciprocal rank fusion across metrics), and normalize_mean (min-max normalized average). An optional speed threshold can filter out modules that are too slow. The selected module becomes the node's output for the downstream pipeline.

Usage

This principle is configured per-node in the YAML config via the "strategy" key. Choose "mean" for simple optimization, "rank" when metrics have different scales, or "normalize_mean" for balanced multi-metric optimization.

Theoretical Basis

Mean strategy: $s c o r e_{i} = \frac{1}{| M |} \sum_{m \in M} m e a n (r e s u l t s_{i} [m])$ Select the module with the highest average metric score.

Rank strategy (Reciprocal Rank): For each metric, rank modules from best to worst: $R R_s c o r e_{i} = \sum_{m \in M} \frac{1}{r a n k_{i} (m)}$ Select the module with the highest reciprocal rank sum.

Normalize-mean strategy: Min-max normalize each metric column, then take the mean: $n o r m (x) = \frac{x - m i n}{m a x - m i n}$ $s c o r e_{i} = \frac{1}{| M |} \sum_{m \in M} n o r m (m e a n (r e s u l t s_{i} [m]))$

Related Pages

Implemented By

Implementation:Marker_Inc_Korea_AutoRAG_Select_Best

Uses Heuristic

Heuristic:Marker_Inc_Korea_AutoRAG_Module_Selection_Strategies

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment