Could you provide detail code example to customize the env and algo that can run successfully in a entire file?

[yuuma002]

We have read the docs carefully. However, we cannot find such example. All we can find is:

https://github.com/opendilab/LightZero/blob/main/docs/source/tutorials/envs/customize_envs_zh.md

What we want is the code example that can run successfully in a entire files with customized envs, which is not by the config files (e.g. zoo/classic_control/cartpole/config/cartpole_muzero_config.py)

Could you provide detail code example to customize the env and algo that can run successfully in a entire file?

Thank a lot.

[PaParaZz1]

You can refer to our previous pull request about adding a new environment. Such as:

• Memory Env
• MountainCar Env
• 2048 Env and Stochastic MuZero
• Connect4 Env

If you have any other questions, please feel free to ask in this issue.

[yuuma002]

Should we refer to the source code of ding package when we write code with customized the env or algo?

[PaParaZz1]

Typically, you do not need to refer to the source code of DI-engine (ding); simply following the examples we provide should suffice. However, if you encounter issues that you are unable to resolve, you may consult the policy/model/env sections of DI-engine or seek assistance by posting issues in LightZero.

[yuuma002]

Could you provide detail docs of the classes we should used in building RL algorithms?

We find some classes lacks docs in https://di-engine-docs.readthedocs.io/en/latest/01_quickstart/first_rl_program.html:

• StepCollector
• eps_greedy_handler
• interaction_evaluator
• ...

This makes it diffucult to understand the quickstart tutorial.

Hope you can fix the issue.

Thanks.

[PaParaZz1]

LightZero uses our old pipeline of DI-engine while the quickstart doc of DI-engine has already presented our new pipeline.

• Here is the doc (in Chinese, if necessary, we will add the English version next week) about their difference.
• Here is the doc about the concrete API definition of middlewares (e.g. eps_greedy_handler) in new pipeline.
• Here is the doc about the concrete API definition of modules in old pipeline.

[yuuma002]

When will the docs be completed (e.g. https://di-engine-docs.readthedocs.io/en/latest/05_api_doc/framework/middleware/functional/explorer.html#eps-greedy-handler)?

[PaParaZz1]

When will the docs be completed (e.g. https://di-engine-docs.readthedocs.io/en/latest/05_api_doc/framework/middleware/functional/explorer.html#eps-greedy-handler)?

We have fixed this problem in the latest commit of DI-engine. It is a import bug resulting in the wrong doc format. Now you can visit the corresponding API doc with this doc link.