MuZero

Implementation of the MuZero algorithm

Introduction

This repo contains:

• a simple but working implementation of the MuZero algorithm
• an agent trained using the MuZero algorithm to play an openAI gym game (CartPole-v1)

Project details

The code is an implementation of the official MuZero pseudocode.

This is an implementation of an agent that uses MuZero in order to play the openAI gym game of CartPole.

Getting Started

Execute the code in the notebook to train the agent!

Dependencies

To set up your python environment to run the code in this repository, follow the instructions below.

1. Create (and activate) a new environment with Python 3.6.

   • Linux or Mac:

   conda create --name MuZero python=3.6
   source activate MuZero

   • Windows:

   conda create --name MuZero python=3.6 
   activate MuZero

2. Clone the repository, and then, install the required packages (see requirements).

git clone https://github.com/ciamic/MuZero.git

4. Create an IPython kernel for the MuZero environment.

python -m ipykernel install --user --name MuZero --display-name "MuZero"

5. Before running code in a notebook, change the kernel to match the MuZero environment by using the drop-down contextual Kernel menu.

Requirements

• Python 3
• numpy
• matplotlib
• gym
• Tensorflow