test(rjy): add discrete pendulum env #395
Conversation
nighood
changed the title
| @@ -11,7 +11,7 @@ | |||
| @ENV_REGISTRY.register('pendulum') | |||
| class PendulumEnv(BaseEnv): | |||
|
|
|||
| def __init__(self, cfg: dict) -> None: | |||
| def __init__(self, cfg: dict, is_continue=True) -> None: | |||
There was a problem hiding this comment.
don't use is_xxx or use_xxx, just continuous is ok
| if self._act_scale: | ||
| action = affine_transform(action, min_val=self._env.action_space.low, max_val=self._env.action_space.high) | ||
| obs, rew, done, info = self._env.step(action) | ||
| self._final_eval_reward += rew | ||
| obs = to_ndarray(obs).astype(np.float32) | ||
| rew = to_ndarray([rew]).astype(np.float32) # wrapped to be transfered to a array with shape (1,) | ||
| # rew = to_ndarray([rew]).astype(np.float32) # wrapped to be transfered to a array with shape (1,) | ||
| rew = np.array([rew], dtype=np.float32) |
There was a problem hiding this comment.
why use this implementation, it has no difference with original version
There was a problem hiding this comment.
Because I ran into such a problem when testing: AttributeError: 'list' object has no attribute 'shape'. So I'm guessing if it's the problem here, and later I find out that it's the problem with the parameters passed in earlier.
| @@ -24,6 +24,9 @@ def __init__(self, cfg: dict) -> None: | |||
| self._reward_space = gym.spaces.Box( | |||
| low=-1 * (3.14 * 3.14 + 0.1 * 8 * 8 + 0.001 * 2 * 2), high=0.0, shape=(1, ), dtype=np.float32 | |||
| ) | |||
| # require discrete env | |||
| self._is_continue=is_continue | |||
There was a problem hiding this comment.
you should modify self._action_space when continuous=False, so you can use consistent interface in other parts like random_action method
| @@ -58,12 +61,17 @@ def seed(self, seed: int, dynamic_seed: bool = True) -> None: | |||
|
|
|||
| def step(self, action: np.ndarray) -> BaseEnvTimestep: | |||
| assert isinstance(action, np.ndarray), type(action) | |||
| # if require discrete env, convert actions to [-2 ~ 2] float actions | |||
| if not self._is_continue: | |||
| action=(action-(self._discrete_action_num-1)/2)/((self._discrete_action_num-1)/4) | |||
There was a problem hiding this comment.
you need to transform it to [-1, 1] here, then affine_transform will make it to [-2, 2]
| @@ -58,12 +61,17 @@ def seed(self, seed: int, dynamic_seed: bool = True) -> None: | |||
|
|
|||
| def step(self, action: np.ndarray) -> BaseEnvTimestep: | |||
| assert isinstance(action, np.ndarray), type(action) | |||
| # if require discrete env, convert actions to [-2 ~ 2] float actions | |||
| if not self._is_continue: | |||
| action=(action-(self._discrete_action_num-1)/2)/((self._discrete_action_num-1)/4) | |||
There was a problem hiding this comment.
why not just (action / (self._discrete_action_num-1)) * 2 - 1
PaParaZz1
changed the title
Codecov Report
@@ Coverage Diff @@
## main #395 +/- ##
==========================================
+ Coverage 85.61% 85.74% +0.13%
==========================================
Files 524 524
Lines 41108 41499 +391
==========================================
+ Hits 35195 35584 +389
- Misses 5913 5915 +2
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
PaParaZz1
added
env
Description
Related Issue
TODO
Check List