VikingZero
This repository is an effort to understand the AlphaGo family of algorithms as part of the fullfillment for a MS thesis.
Paper Link: From MDP to AlphaZero
Installation
pip install vikingzero
Or if you are familiar with poetry (https://python-poetry.org/) you can install from source using the following
git clone https://github.com/befeltingu/VikingZero.git
cd VikingZero
poetry install
You can also use the setup.py file or requriements file for installing from source.
Usage
For example of how to train an agent go to notebooks/run_alphago.ipynb and run the cells.
This project uses yaml files for run configurations. notebooks/ has a few example files to get started.
There is a notebooks/step_tutorial.ipynb file that is meant as a tutorial for better understanding of the Monte Carlo Tree search steps. It used pydot with graphviz and requires that you have an installation of graphviz.
Agents
There are a few types of agents that you can experiment with in this project.
- MiniMax
- AlphaBeta Pruning
- AlphaBeta Pruing with depth limited search
- Monte carlo tree search - UCT
- AlphaZero - ( Look to notebooks/connect4.yaml for a connect4 configuration)
Play
Go to https://github.com/befeltingu/VikingDashboard for a dashboard where you can play against some pre trained models
