Gym Renju

Gym-renju is an Renju game simulating environment for Open AI Gym. The game is played mainly 15x15 sized board, and 2 players (Black and White) put a stone on the board alternatively.

The player will win the game when he/she make a 5 not breaking stones horizontally, vertically, or diagonally. The detail rule is a little bit complexed, so please refer to the link below if you want to know it.

https://en.wikipedia.org/wiki/Renju

Also, I have to note that some of the detail rules is not implemented. Please refer to [To be implemented] to know it.

Dependencies

Product Dependencies

• gym (https://github.com/openai/gym)
• Chainer (https://chainer.org/)

Test Dependencies

• pytest (https://github.com/pytest-dev/pytest)
• parameterized (https://github.com/wolever/parameterized)
• injector (https://github.com/alecthomas/injector)

Installation

pip install gym-renju

Learning Example

To implement machine learning code, please refer to the sample implementation below. This code is an implementation of AI with DQN on Chainer. https://github.com/aratakokubun/renju-dqn

Quick Example

See src in sample.py

You can test environment with the code below.

# Imports
import gym
import time
# Change import on installed on pip
import sys, os
sys.path.append(os.path.join(os.path.dirname(__file__), ".."))
import gym_renju
env = gym.make('Renju15x15-v0') # default 'beginner' level opponent policy

env.reset()
env.render()

# Play a game
env.reset()
for _ in range(225):
    action = env.action_space.sample() # sample without replacement
    observation, reward, done, info = env.step(action)
    env.render()
    if done:
        print ("Game is Over")
        break

Test

py.test -v

To be implementated

• Cases of 'Not San San'. This is the case that a player create 'San San' but he/she will not be able to create a 'Yon' on either of 'San' because the action to create 'Yon' is not allowed by other rules (ex. 'Yon Yon').
• Add envs and AI policies.