Module graph_weight_optimizer 

Graph Weight Optimization (Simplified DW-GRPO)

This module implements a simplified version of Dynamic Weighted Group Relative Policy Optimization (DW-GRPO) for optimizing relationship weights in the knowledge graph.

Key features:

• Heuristic-based optimization (not full reinforcement learning)
• Gradient-free hill climbing for weight adjustment
• Multi-objective optimization (relevance, faithfulness, conciseness)
• Stagnation detection and dynamic weight adjustment
• Performance tracking across iterations

Structs

GraphWeightOptimizer

Graph weight optimizer using simplified DW-GRPO approach

ObjectiveWeights

Weights for combining multiple objectives

OptimizationStep

A single optimization iteration with metrics

OptimizerConfig

Configuration for the optimizer

TestQuery

Test query with expected answer for evaluation