Skip to content
forked from pmuens/alphago

Reference implementation of DeepMinds AlphaGo based on "Deep Learning and the Game of Go"

Notifications You must be signed in to change notification settings

Ting817/alphago

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

77 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AlphaGo

This repository contains a reference implementation of the AlphaGo AI by DeepMind.

How to play

Go

Bot vs. Bot

Run python bot_v_bot.py to let 2 Bots play against each other.

Human vs. Bot

Run python mcts_go.py to play against a bot.

Tic-Tac-Toe

Human vs. Bot

Run python play_ttt.py to play against an unbeatable bot.

Reinforcement Learning

  1. Run python init_ac_agent.py --board-size 9 --output-file ./agents/ac_v1.h5

  2. Run python self_play_ac.py --board-size 9 --learning-agent ./agents/ac_v1.h5 --num-games 5000 --experience-out ./experiences/exp_0001.h5 to let a bot play against itself and store experiences gathered during self play.

  3. Run python train_ac.py --learning-agent ./agents/ac_v1.h5 --agent-out ./agents/ac_v2.h5 ./--lr 0.01 --bs 1024 experiences/exp_0001.h5 to use experience data for agent improvements via Deep Reinforcement Learning.

  4. Run python eval_ac_bot.py --agent1 ./agents/ac_v2.h5 --agent2 ./agents/ac_v1.h5 --num-games 100 to check whether the new bot is stronger.

If the new agent is stronger start with it at 2.

Otherwise go to 2. again to generate more training data. Use multiple experience data files in 3.

Rinse and repeat.

Resources

About

Reference implementation of DeepMinds AlphaGo based on "Deep Learning and the Game of Go"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 74.8%
  • JavaScript 15.7%
  • HTML 9.4%
  • Nix 0.1%