The Alpha Zero, a well-known computer program with the ability to achieve superhuman level of play in go, chess and shogi within 24 hours where only game rules are pro- vided, is based on the Monte-Carlo tree search (MCTS) al- gorithm using a single Deep Neural Network. Our goal for the entire project is trying to improve the Alpha Zero by introducing an ensemble of assorted neural networks.

To start training a model for go:

python Gomain.py

There are two major targets consist- ing of establishing rules for the Game Go and constructing a single Deep Residual Network (ResNet) rather than a single Deep Neural Network to train the Game Go.

• Fork from https://github.com/suragnair/alpha-zero-general
• Used two repositories(https://github.com/ambbber/alpha-zero-go) at first, then edwardchor loaded and merged all code into his repository.