feature(zjow): add agent configurations & polish replay video saving …

…methods (#184) * Add muzero config * polish atari env replay code * add more variant muzero agents * add more variant muzero agents * polish setup * Add tictactoe * polish render of pendulum * polish muzero breakout config * Add efficientzero config * Add efficientzero config * polish atari env replay * polish code * add sampled efficientzero config * add sampled efficientzero config * add sampled efficientzero config * Add gomoku & tictactow Agent config * polish config * change env_name in config into env_id
opendilab · Feb 8, 2024 · eb8594c · eb8594c
1 parent 541f5bf
commit eb8594c
Show file tree

Hide file tree

Showing 129 changed files with 4,676 additions and 303 deletions.
diff --git a/docs/source/tutorials/envs/customize_envs.md b/docs/source/tutorials/envs/customize_envs.md
@@ -203,12 +203,12 @@ class LightZeroEnvWrapper(gym.Wrapper):
 Specifically, use the following function to wrap a gym environment into the format required by LightZero using `LightZeroEnvWrapper`. The `get_wrappered_env` function returns an anonymous function that generates a `DingEnvWrapper` instance each time it is called. This instance takes `LightZeroEnvWrapper` as an anonymous function and internally wraps the original environment into the format required by LightZero.
 
 ```python
-def get_wrappered_env(wrapper_cfg: EasyDict, env_name: str):
+def get_wrappered_env(wrapper_cfg: EasyDict, env_id: str):
     # overview comments
     ...
     if wrapper_cfg.manually_discretization:
         return lambda: DingEnvWrapper(
-            gym.make(env_name),
+            gym.make(env_id),
             cfg={
                 'env_wrapper': [
                     lambda env: ActionDiscretizationEnvWrapper(env, wrapper_cfg), lambda env:
@@ -218,7 +218,7 @@ def get_wrappered_env(wrapper_cfg: EasyDict, env_name: str):
         )
     else:
         return lambda: DingEnvWrapper(
-            gym.make(env_name), cfg={'env_wrapper': [lambda env: LightZeroEnvWrapper(env, wrapper_cfg)]}
+            gym.make(env_id), cfg={'env_wrapper': [lambda env: LightZeroEnvWrapper(env, wrapper_cfg)]}
         )
 ```
 

diff --git a/docs/source/tutorials/envs/customize_envs_zh.md b/docs/source/tutorials/envs/customize_envs_zh.md
@@ -216,12 +216,12 @@ class LightZeroEnvWrapper(gym.Wrapper):
 具体使用时，使用下面的函数，将一个 gym 环境，通过 `LightZeroEnvWrapper` 包装成 LightZero 所需要的环境格式。 `get_wrappered_env` 会返回一个匿名函数，该匿名函数每次调用都会产生一个 `DingEnvWrapper` 实例，该实例会将 `LightZeroEnvWrapper` 作为匿名函数传入，并在实例内部将原始环境封装成 LightZero 所需的格式。
 
 ```Python
-def get_wrappered_env(wrapper_cfg: EasyDict, env_name: str):
+def get_wrappered_env(wrapper_cfg: EasyDict, env_id: str):
     # overview comments
     ...
     if wrapper_cfg.manually_discretization:
         return lambda: DingEnvWrapper(
-            gym.make(env_name),
+            gym.make(env_id),
             cfg={
                 'env_wrapper': [
                     lambda env: ActionDiscretizationEnvWrapper(env, wrapper_cfg), lambda env:
@@ -231,7 +231,7 @@ def get_wrappered_env(wrapper_cfg: EasyDict, env_name: str):
         )
     else:
         return lambda: DingEnvWrapper(
-            gym.make(env_name), cfg={'env_wrapper': [lambda env: LightZeroEnvWrapper(env, wrapper_cfg)]}
+            gym.make(env_id), cfg={'env_wrapper': [lambda env: LightZeroEnvWrapper(env, wrapper_cfg)]}
         )
 ```
 

diff --git a/lzero/agent/__init__.py b/lzero/agent/__init__.py
@@ -1 +1,6 @@
+from .alphazero import AlphaZeroAgent
+from .efficientzero import EfficientZeroAgent
+from .gumbel_muzero import GumbelMuZeroAgent
 from .muzero import MuZeroAgent
+from .sampled_alphazero import SampledAlphaZeroAgent
+from .sampled_efficientzero import SampledEfficientZeroAgent