sssxccc/alphazero-in-unity

Several board game AI, which are trained based on AlphaZero, used in a Unity game.

AlphaZero-In-Unity

Several board game AI, which are trained based on AlphaZero, used in a Unity game.

I write a simple AlphaZero framework, implement ConnectFour AI, Reversi AI, Gobang AI, and create a unity game to play with those AI.

我写了一个简单的AlphaZero框架，实现了四子棋AI，翻转棋AI，五子棋AI，并创建了一个可以和这些AI下棋的unity游戏。

AlphaZero Reference: Mastering the game of Go without human knowledge, Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm.

博客（blog）：https://blog.csdn.net/SSSxCCC/article/details/86563666

alphazero-in-unity's Issues

How to run this project on Unity

Hi,
I've import this plugin (https://github.com/miyamotok0105/unity-ml-agents/blob/master/docs/Using-TensorFlow-Sharp-in-Unity.md#using-your-own-trained-graphs)
Then run the Unity project.

But I get below issue
DllNotFoundException: libtensorflow assembly: type: member:(null)
TensorFlow.TFGraph..ctor () (at <6ed6db22f8874deba74ffe3e566039be>:0)
Connect4PolicyValueNet..ctor (System.Int32 width, System.Int32 height) (at Assets/Scripts/Connect4/Connect4Model.cs:17)
Connect4Game.Start () (at Assets/Scripts/Connect4/Connect4Game.cs:38)

Can you help me fix it?

Models Save/Restore

Have a question, how did you converted models from tensorflow to bytes files in unity? Can't figure out this step.
Because this code saves 4 files.

    def save_model(self, model_path):
      self.saver.save(self.session, model_path)

Can alpha-zero learn Bejeweled?

For Bejeweled, after the same colour in a row disappeared, the jewel from row above will move down and if the new row match then they will dissappear too. One action will cause other several moves.

Assuming no new jewel will be generated, just find the best move to solve all the existing jewels.

Multi gpu model, low GPU utilization

Training using GPU only utilize 1% of my GPU, anybody able to enhance it?

Maybe using keras.utils, multi_gpu_model?

Can you add a license?

I am considering re-releasing parts of this work on NuGet for general consumption.