This repo aims to provide a working AlphaZero implementation that's simple enough to be able to understand what's going on at a quick glance, without sacrificing too much.

The connect2.py example gives a rough idea of what's needed to train an agent. Being this basic, the game does not require everything a complete AlphaZero training loop requires but the main things should be there.

The main reference for this implementation can be found at this link, which is the preprint of the Nature paper published by DeepMind in 2017.

Requirements:

• torch
• matplotlib (for connect2.py example)
• tqdm (for progress bar)