Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/benzyx/DomRL

DomRL is a simulation environment for the card game Dominion, created by Donald X Vaccarino, meant to simplify the development and testing of various AI strategies, specifically Reinforcement Learning algorithms.
https://github.com/benzyx/DomRL

Last synced: 8 days ago
JSON representation

DomRL is a simulation environment for the card game Dominion, created by Donald X Vaccarino, meant to simplify the development and testing of various AI strategies, specifically Reinforcement Learning algorithms.

Awesome Lists containing this project

README

        

# DomRL

DomRL is a simulation environment for the card game Dominion, created by Donald X Vaccarino, meant to simplify the development and testing of various AI strategies, specifically Reinforcement Learning algorithms.

The goal is to make an engine that stores a serializable state of the game and exposes a comprehensive state object to agents, which contain all the information that human players would have. This improved state management will be helpful for building more complex AI agents, beyond simple priority logic supported by most agents.

The engine currently supports all cards in Base Dominion.

## Usage

If you want to play from the command line, simply run:

```
python main.py
```

Currently, this runs two `StdinAgent` instances against each other, so you can choose all the moves from command line.

If you want to run a game from a Jupyter notebook, between two separate agents, the following should suffice:

```
import engine.game.Game

Game(YourAgent(...), YourAgent(...)).run()
```

## Writing a Bot

Implement an agent, which is a class derived from `engine.agent.Agent` that implements

```
policy(self, state_view, decision) -> List
```

where the output is the list of indices corresponding to the moves taken in this decision object.

## Upcoming Features

The following features are necessary before this project can reach an acceptable state for proper RL agent development.

- Testing for correctness.
- Baseline bots (such as big money)
- Undo feature (mostly for human convenience)
- Vector encoding interface for decisions.
- Improved logging (currently do not log draws), and unambiguous events (topdecking from hand vs topdecking from supply pile).