Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/werner-duvaud/muzero-general
MuZero
https://github.com/werner-duvaud/muzero-general
alphago alphazero deep-learning deep-reinforcement-learning gym machine-learning mcts model-based-rl monte-carlo-tree-search muzero muzero-general neural-network python3 pytorch reinforcement-learning residual-network rl self-learning tensorboard
Last synced: 1 day ago
JSON representation
MuZero
- Host: GitHub
- URL: https://github.com/werner-duvaud/muzero-general
- Owner: werner-duvaud
- License: mit
- Created: 2019-12-27T02:33:51.000Z (almost 5 years ago)
- Default Branch: master
- Last Pushed: 2024-09-03T21:32:07.000Z (3 months ago)
- Last Synced: 2024-12-04T13:05:36.534Z (8 days ago)
- Topics: alphago, alphazero, deep-learning, deep-reinforcement-learning, gym, machine-learning, mcts, model-based-rl, monte-carlo-tree-search, muzero, muzero-general, neural-network, python3, pytorch, reinforcement-learning, residual-network, rl, self-learning, tensorboard
- Language: Python
- Homepage: https://github.com/werner-duvaud/muzero-general/wiki/MuZero-Documentation
- Size: 7.09 MB
- Stars: 2,523
- Watchers: 74
- Forks: 613
- Open Issues: 58
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome-ray - muzero-general - A commented and documented implementation of MuZero based on the Google DeepMind paper (Schrittwieser et al., Nov 2019) and the associated pseudocode. (Models and Projects / Reinforcmenet Learning)
README
-929292)
[](https://github.com/psf/black)
[](https://discord.gg/GB2vwsF)
# MuZero General
A commented and [documented](https://github.com/werner-duvaud/muzero-general/wiki/MuZero-Documentation) implementation of MuZero based on the Google DeepMind [paper](https://arxiv.org/abs/1911.08265) (Schrittwieser et al., Nov 2019) and the associated [pseudocode](https://arxiv.org/src/1911.08265v2/anc/pseudocode.py).
It is designed to be easily adaptable for every games or reinforcement learning environments (like [gym](https://github.com/openai/gym)). You only need to add a [game file](https://github.com/werner-duvaud/muzero-general/tree/master/games) with the hyperparameters and the game class. Please refer to the [documentation](https://github.com/werner-duvaud/muzero-general/wiki/MuZero-Documentation) and the [example](https://github.com/werner-duvaud/muzero-general/blob/master/games/cartpole.py).
This implementation is primarily for educational purpose.\
[Explanatory video of MuZero](https://youtu.be/We20YSAJZSE)
MuZero is a state of the art RL algorithm for board games (Chess, Go, ...) and Atari games.
It is the successor to [AlphaZero](https://arxiv.org/abs/1712.01815) but without any knowledge of the environment underlying dynamics. MuZero learns a model of the environment and uses an internal representation that contains only the useful information for predicting the reward, value, policy and transitions. MuZero is also close to [Value prediction networks](https://arxiv.org/abs/1707.03497). See [How it works](https://github.com/werner-duvaud/muzero-general/wiki/How-MuZero-works).
## Features
* [x] Residual Network and Fully connected network in [PyTorch](https://github.com/pytorch/pytorch)
* [x] Multi-Threaded/Asynchronous/[Cluster](https://docs.ray.io/en/latest/cluster-index.html) with [Ray](https://github.com/ray-project/ray)
* [X] Multi GPU support for the training and the selfplay
* [x] TensorBoard real-time monitoring
* [x] Model weights automatically saved at checkpoints
* [x] Single and two player mode
* [x] Commented and [documented](https://github.com/werner-duvaud/muzero-general/wiki/MuZero-Documentation)
* [x] Easily adaptable for new games
* [x] [Examples](https://github.com/werner-duvaud/muzero-general/blob/master/games/cartpole.py) of board games, Gym and Atari games (See [list of implemented games](https://github.com/werner-duvaud/muzero-general#games-already-implemented))
* [x] [Pretrained weights](https://github.com/werner-duvaud/muzero-general/tree/master/results) available
* [ ] Windows support (Experimental / Workaround: Use the [notebook](https://github.com/werner-duvaud/muzero-general/blob/master/notebook.ipynb) in [Google Colab](https://colab.research.google.com))
### Further improvements
Here is a list of features which could be interesting to add but which are not in MuZero's paper. We are open to contributions and other ideas.
* [x] [Hyperparameter search](https://github.com/werner-duvaud/muzero-general/wiki/Hyperparameter-Optimization)
* [x] [Continuous action space](https://github.com/werner-duvaud/muzero-general/tree/continuous)
* [x] [Tool to understand the learned model](https://github.com/werner-duvaud/muzero-general/blob/master/diagnose_model.py)
* [ ] Batch MCTS
* [ ] Support of more than two player games
## Demo
All performances are tracked and displayed in real time in [TensorBoard](https://www.tensorflow.org/tensorboard) :
Testing Lunar Lander :
## Games already implemented
* Cartpole (Tested with the fully connected network)
* Lunar Lander (Tested in deterministic mode with the fully connected network)
* Gridworld (Tested with the fully connected network)
* Tic-tac-toe (Tested with the fully connected network and the residual network)
* Connect4 (Slightly tested with the residual network)
* Gomoku
* Twenty-One / Blackjack (Tested with the residual network)
* Atari Breakout
Tests are done on Ubuntu with 16 GB RAM / Intel i7 / GTX 1050Ti Max-Q. We make sure to obtain a progression and a level which ensures that it has learned. But we do not systematically reach a human level. For certain environments, we notice a regression after a certain time. The proposed configurations are certainly not optimal and we do not focus for now on the optimization of hyperparameters. Any help is welcome.
## Code structure
Network summary:
## Getting started
### Installation
```bash
git clone https://github.com/werner-duvaud/muzero-general.git
cd muzero-general
pip install -r requirements.lock
```
### Run
```bash
python muzero.py
```
To visualize the training results, run in a new terminal:
```bash
tensorboard --logdir ./results
```
### Config
You can adapt the configurations of each game by editing the `MuZeroConfig` class of the respective file in the [games folder](https://github.com/werner-duvaud/muzero-general/tree/master/games).
## Related work
* [EfficientZero](https://arxiv.org/abs/2111.00210) (Weirui Ye, Shaohuai Liu, Thanard Kurutach, Pieter Abbeel, Yang Gao)
* [Sampled MuZero](https://arxiv.org/abs/2104.06303) (Thomas Hubert, Julian Schrittwieser, Ioannis Antonoglou, Mohammadamin Barekatain, Simon Schmitt, David Silver)
## Authors
* Werner Duvaud
* Aurèle Hainaut
* Paul Lenoir
* [Contributors](https://github.com/werner-duvaud/muzero-general/graphs/contributors)
Please use this bibtex if you want to cite this repository (master branch) in your publications:
```bash
@misc{muzero-general,
author = {Werner Duvaud, Aurèle Hainaut},
title = {MuZero General: Open Reimplementation of MuZero},
year = {2019},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/werner-duvaud/muzero-general}},
}
```
## Getting involved
* [GitHub Issues](https://github.com/werner-duvaud/muzero-general/issues): For reporting bugs.
* [Pull Requests](https://github.com/werner-duvaud/muzero-general/pulls): For submitting code contributions.
* [Discord server](https://discord.gg/GB2vwsF): For discussions about development or any general questions.