Clue

This repo implements an environment for the board game Clue that can be used to train a reinforcement learning agent to play the game.

It uses the OpenAI gym interface and petting zoo to represent clue as a multi-agent reinforcement learning problem.

The key files are:

clue/env/clue_environment_<version>.py - The environment.
clue/map.py - The map of the board and utility functions for working with it.
clue/map49.csv - A csv representation of the board from the 1949 rules
clue/state.py - The logic of the game and the state representation.
clue/train_agents_<version>.py - different attemps to train agents to play the game.
clue/interactive.py - A script to play the game interactively against a trained agent.

Current status

The environment captures the rules of the game ok, but the trained agents don't play very well.

I've tried a few different approaches to training agents, but none of them are much better than random.

I suspect that there is a problem with my training setup or model architecture.

A csv file represents the board layout.

Key:

"" - part of a room
0 - unaccessible part of the grid
1 - a regular square on the grid
d[nsew]:(.*) a door into a room
- the second char indicates which side of the cell the door way is
- the text after the colon is the name of the room
s[a-z]{1,2} - the starting point of a player
[A-Z].* - The name of a room

Install the git hooks:

poetry run pre-commit install

You can check out what a policy would experience by running the interactive script:

poetry run python clue/interactive.py

It will create a log of the session in a file called interactive.<timestamp>.log