ryankeiper / gym-chess

An OpenAI gym environment for chess

gym-chess

This documentation is slightly out of date and will be updated soon. For now what you need to know is that calling env.action_space.n returns a list of legal moves

Pre-requisites:

Below are a list of packages that are required to get the game to work, along with links for installation instructions.

python-chess

For usage and installation instructions, see: https://pypi.python.org/pypi/python-chess

OpenAI Gym

For installation instructions, see: https://www.youtube.com/watch?v=Io1wOuHEyW0&feature=youtu.be

gym_chess Documentation

This environment is still actively being developed. Always check the docs for the most up to date information.

gym_chess Installation

While in the root directory you can run the following code:

pip install -e .

Running chess as a gym environment

import gym
import gym_chess

env = gym.make('chess-v0')
env.render()

Return Values

state

The environment state consists of 2 parts: [[an 8x8 array of the game board with pieces represented as integers],[A list of all legal moves]] Pieces are assigned numerical values as such:

1: Pawn
2: Knight
3: Bishop
4: Rook
5: Queen
6: King

Values are always positive for the side who's turn it is.

reward

Float value defined in the REWARD_LOOKUP dict.

Reward values are currently set as such:

Check: 0.05
Checkmate: 100
Stalemate: 0
Pawn promotion: 0.1
**Taking a piece:**
Pawn: 0.1
Knight: 0.3
Bishop: 0.3
Rook: 0.5
Queen: 0.9

is_terminated

Boolean value representing if the game has ended in checkmate, stalemate, insufficient material, seventyfive-move rule, fivefold repetition, or a variant end condition.

info

Dictionary containing any debugging information

Params

action

The move to be played

Must be represented in UCI format (i.e. 'a2a4' or 'b7b8q')

Methods

step(action)

Plays a move

Returns state, reward, is_terminated, info

reset()

Resets the board to the starting position

Returns state

render()

Prints the board to the console with pieces represented as letters

alt_step(action)

Similar to step(). However, calling step() will remove all moves made by alt_step(), and revert to the last move used in step().

This can be used to allow the agent to look ahead several moves.

Returns state, is_terminated, info

alt_pop()

Pops a single move performed by alt_step()

This can be used to build trees of possible future board positions.

Returns state

alt_reset()

Pops all moves performed by alt_step()

Returns state