Practical overview
When it fits
Use Muzero when you want a documented TensorFlow codebase for reinforcement-learning experiments, especially training or comparing agents in the named Gym environments and investigating learned MuZero models.
Who it is for
- Researchers and students experimenting with model-based reinforcement learning.
- Developers who want to configure or extend MuZero and AlphaZero models.
- Practitioners comparing the algorithms in supported Gym environments.
Documented uses
- Configure and train a MuZero or AlphaZero agent.
- Pit the two algorithms against each other.
- Study learned MuZero models by varying observation frequency during environment interaction.
- Explore AlphaZero-style methods in single-player domains.
Requirements & limits
The repository is public and lists an MIT license.
- The documented stack requires Python 3.7 or later, TensorFlow, and tqdm; standalone Keras is also listed as a minimal requirement.
- The README lists Python 3.7.9, TensorFlow 2.1.0, and Keras 2.3.1 as tested versions on Windows and Linux.
- MuZero board-game experiments were not extensively tested; the README says computation time and tuning were challenges.
- The implementation omits parallelization used in the original MuZero paper.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Train an agent with a JSON configuration
Run the documented training workflow for an agent using a selected game.
Before you start
- Python 3.7 or later
- TensorFlow, standalone Keras, and tqdm
- A JSON configuration file specifying agent parameters
- A neural-network architecture specified in the configuration; existing architectures are referenced in Agents/__init__.py
Steps
Create a JSON configuration file using the examples in Configurations/ModelConfigs and specify the agent parameters.
Specify a neural-network architecture in the configuration, using an existing architecture referenced in Agents/__init__.py if appropriate.
Run Main.py with the training flags, substituting the configuration path and GPU integer: the command below.
python Main.py train -c my/config/file.json --game gym_Cartpole-v1 --gpu [INT]
Expected outcome
The command starts agent training for the selected game using the configuration.
Guide limitations
- The repository notes that MuZero board-game environments were not tested and that computation time became an issue.
- The README reports CartPole-v1, MountainCar-v0, and Pendulum-v0 as well-tested environments; canonical MuZero can be unstable depending on hyperparameters.
- The README lists tested versions as Python 3.7.9, TensorFlow 2.1.0, and Keras 2.3.1 on Windows and Linux.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- kaesve/muzero · Official homepage title · retained in its source language
Related options
AlphaZero-General — The README identifies it as the framework this implementation follows and lists it as another AlphaZero implementation.
MuZero-General — The README lists it as a parallelized PyTorch MuZero implementation.
MuZero in TensorFlow by johan-gras — The README lists it as another sequential TensorFlow implementation.