Game Documentation

Description

The environment for conducting reinforcement learning experiments is called a game. While it does not explicitly have to be a game, it should contain a certain set of functionalities mentioned later. The game module is considered a place for training and testing RL models.

Functionality

Contains built-in environment logic (e.g. rules of the game)
Should provide functions and methods for interacting with the environment

Game module structure

Prerequisites

The game module should contain an abstract class called Game and an enum class State.

Example of Game and State in Python:

from enum import Enum
from abc import ABC, abstractmethod

class State(Enum):
    ONGOING = 0
    WON = 1
    LOST = 2

class Game(ABC):
    @abstractmethod
    def get_moves(self) -> list:
        pass

    @abstractmethod
    def make_move(self, move: tuple) -> bool:
        pass

    @abstractmethod
    def get_state(self) -> State:
        pass

    @abstractmethod
    def get_board(self) -> list:
        pass

    @abstractmethod
    def start_game(self) -> None:
        pass

State enum class

State enum class provides a highly readable way of describing the state of the game. Using such a class is highly recommended as it makes it easy to tell whether model needs to undertake another game and to check its outcome.

Game functions and methods

game_moves

Input: None

Output: list of possible moves

make_move

Input: move described as a tuple

Output: boolean value indicating whether the move was successful

get_state

Input: None

Output: one of State values

get_board

Input: None

Output: list representing the board

start_game

Input: None

Output: None