AlphaZero Gobang Save

Overview

This is a AlphaZero Implementation of Gobang based on Pytorch.

Pytorch 0.3.1 Install

https://ptorch.com/news/145.html

Github

Code can be viewed in my github：https://github.com/xuetf/AlphaZero_Gobang

Design

RL framework

Network Structure

Class Diagram

Illustration can be viewed in my blog: http://xtf615.com/2018/02/24/AlphaZeroDesign/

Final Report(CVPR format)

Final Report: Analysis and Implementation of Deep Reinforcement Learning Based Gobang

Code

• Train.py : Run the train process
• Run.py : Play with Human using the trained model
• Player.py: Base class for different Player
• RolloutPlayer.py: Player with MCTS using random rollout policy
• AlphaZeroPlayer.py: AlphaZero Player with MCTS guided by Residual Network
• HumanPlayer.py: Human Player
• MCTS.py: Base class for different MCTS
• AlphaZeroMCTS.py: MCTS guided by Residual Network
• RolloutMCTS.py: MCTS using random rollout policy
• TreeNode.py: MCTS Tree Node
• PolicyValueNet.py: Redisual Network Implementation based on Pytorch
• Board.py: Board Class for Gobang
• Game.py: Game for Gobang
• VisualTool.py: Tk Tool for visualizing Chess Board
• Config.py: store config. Serve as a snapshot when resuming

Running Code

Training

• Train from scratch:

python3 Train.py

• Train as a background job,then:

nohup python3 -u Train.py > train.log 2>&1 &

• Train from a checkpoint:

python3 Train.py --config data/model_name.pkl

Play game

python3 Run.py

Result

Download or Upload From your OWN remote server

Download the trained model from remote server

scp root@ip:/usr/local/workspace/AlphaZero_Gobang/data/current_policy_resnet_epochs_1500.model /Users/xuetf/Downloads

Upload -P

scp -P 8381 local_file_path [email protected]:/root/