Katsuki Ohto

MuZero or later model-based methods will definitely be important.

(Idea) feature: default config in train.py

1

one idea to use default configuration

What we will discuss: - environment selection - neural net for each environment - legal actions - generation and evaluation result (there is only immediate reward)

only added config files under configs/

3.10 is widely used.

When the socket.timeout occurs, new worker connections are not accepted.

We can easily change the number of pooled models in each worker. (TODO: Can we prioritize which model to save?)

Most users won't change `eval_coef`.

`resolve_agent()` looks unnecessary.