跳到主要内容
Documentation

RL Driven

Overview

Docs indexEdit page
版本：最新版

RL Driven

Overview​

rl_driven is a selection algorithm for online exploration and personalization.

It aligns to config/algorithm/selection/rl-driven.yaml.

Key Advantages​

  • Supports exploration instead of always exploiting the current best model.
  • Can personalize routing as more interactions accumulate.
  • Keeps online-learning behavior local to one decision.

What Problem Does It Solve?​

If the router should keep learning instead of freezing the current winner, a static selector becomes a bottleneck. rl_driven exposes an exploration-based policy for those cases.

When to Use​

  • the route should keep exploring candidate models online
  • personalization should adapt over time
  • the route can tolerate some exploration cost in exchange for better long-term selection

Configuration​

Use this fragment inside routing.decisions[].algorithm:

algorithm:
  type: rl_driven
  rl_driven:
    exploration_rate: 0.15
    use_thompson_sampling: true
    enable_personalization: true