CGLemon

Results 75 comments of CGLemon

@lyhyl Thanks a lot!

你好,這主要是因為 Sayuri 實做的規則導致她無法較準確的 pass ,Sayuri 採用 Tromp–Taylor 規則,此規則本質上屬於中國規則,但不存在死棋這個概念,凡事沒被吃掉的棋都是活棋。此外我還有加上簡單的死活判斷輔助,無條件死的棋會被視作死棋。以下是 309 手時,Sayuri 看到的狀態,除了右邊白棋是無條件死外,剩餘所有棋子都是活棋,如果此時 pass 的話,她認為自己是輸 1.5 目,此情況下她會拒絕 pass。如果想讓她更容易 pass ,可以增加 ```--friendly-pass``` 的參數。 ![Screenshot from 2023-09-27 20-11-47](https://github.com/CGLemon/Sayuri/assets/39462842/04a36f59-ca28-4aa6-a994-b69a4c38228b)

Looks like the engine is not success to load the weights. Please check the GTP shell diagnostic verbose. The picture is example. ![Screenshot from 2023-12-04 23-11-38](https://github.com/CGLemon/Sayuri/assets/39462842/1da111a1-0aef-496c-af30-296eb53b09d6)

If the weight and engine under the same directory, this command can load the weights successfully.

Thanks. I check this issue. The engine can not load path including space on the Command Prompt. However, I don't think it is a bug? should have a way to...

I did not notice that. Thanks for your help.

Fix this bug at ```next``` branch and test it on the Ubuntu. It works well now.

Thank! It warns a unused variable in the cuDNN backend. I will fix this issue. But it is still executable.

Already fixed the flicker issue. Does it look better?

The noise is not set in ```expand_node()```. It is tentative policy. It will be replaced by NN policy in ```process_mini_batch()```. So you are right. The MCTS process is not random.