Sword Growing RL/Macro

Description

• end-to-end Data parsing / RL / macro tool for Sword Growing (검키우기) game in KakaoTalk Gamebot
• RL environment to build agent
• Macro support (on Macbook yet)

Quick start

0. Preliminaries: uv
1. Sync your python env

   make init

2. Set rl/config.py
   • CHAT_OUTPUT_COORD: coordinate of chatbot output
   • CHAT_INPUT_COORD: coordinate of your input box
3. Run macro

   make macro

4. Functions
   • F1, F2 is hotkey for "강화", "판매" respectively
   • F3 starts loop based on AI inference
   • F4 starts loop based on rule-based strategy
   • F5 quits program

RL support

• Used PPO algorithm using SB3
• Makefile script for convenient training
  • Make train: start training
  • Make test: test 1000 timesteps and plot

Data Parsing

1. Export Kakaotalk chat log
2. put .csv file to ./data/*
3. run process_data.ipynb

AI/Heuristic inference support

• Trained RL agent is used for inference
• Rule-based policy
  • If LEVEL_THRESHOLD achieved, if fail count surpasses FAIL_COUNT_THRESHOLD, sell current sword
  • refer to rl/config.py

Macro (Hotkey) Support

• F1: 강화
• F2: 판매
• F3: AI inference (loop)
• F4: Heuristic inference (loop)
• F5: 매크로 종료

TODOS

• Windows support
• More training
• More data collection (for better environment modeling)