AlphaZeroGeneral is a project developed as part of the "Laboratórios de IACD" course. The primary objective of this project is to create a general AlphaZero implementation that can generalize and learn to play any game. In particular, we have tested the implementation with the games of Go and Attaxx.
- Generalized AlphaZero architecture.
- Learning and playing games knowing only the rules.
- Support for different games given correct architecture.
- Training and evaluation pipelines.
The project includes several scripts and modules for training and playing games. All of the necessary code to run the model in contained in main_notebook.ipynb, which allows you to select the game, load any previous model or create a new one, train or test it or connect the agent to a server. Arguments, aliases and save locations can be manually changed in the notebook.
Currently, AlphaZeroGeneral has been tested with the following games:
- Go
- Attaxx
You can extend the project to support additional games by implementing the necessary interfaces and rules for the new game.