minor

tianshou/policy/base.py

@@ -64,9 +64,9 @@ def __init__(
         self.agent_id = 0
         self.updating = False
         self.action_scaling = action_scaling
-        # can be either "clip", "tanh", or empty string for doing nothing
-        assert action_bound_method in ["", "clip", "tanh"]
+        # can be one of ("clip", "tanh", ""), empty string means no bounding
         self.action_bound_method = action_bound_method
+        assert action_bound_method in ("", "clip", "tanh")
         self._compile()
 
     def set_agent_id(self, agent_id: int) -> None:
@@ -121,24 +121,24 @@ def forward(
         pass
 
     def map_action(self, act: Union[Batch, np.ndarray]) -> Union[Batch, np.ndarray]:
-        """Map raw network output action to action range in gym's env.action_space.
+        """Map raw network output to action range in gym's env.action_space.
 
         This function is called in :meth:`~tianshou.data.Collector.collect` and only
         affects action sending to env. Remapped action will not be stored in buffer
         and thus can be viewed as a part of env (a black box action transformation).
 
-        Basically it assumes the original action range is [-1, 1] or (-inf, inf), and
-        maps them to [action_space.low, action_space.high] by cliping, applying tanh
-        activation, or linear scaling.
+        Action mapping includes 2 standard procedures: bounding and scaling. Bounding
+        procedure expects original action range is (-inf, inf) and maps them to [-1, 1],
+        while scaling procedure expects original action range is (-1, 1) and maps them
+        to [action_space.low, action_space.high]. Bounding procedure is applied first.
 
         :param act: a data batch or numpy.ndarray which is the action taken by
             policy.forward.
 
         :return: action in the same form of input "act" but remap to the target action
             space.
         """
-        if isinstance(self.action_space, gym.spaces.Box) and \
-                isinstance(act, np.ndarray):
+        if isinstance(self.action_space, gym.spaces.Box):
             # currently this action mapping only supports np.ndarray action
             if self.action_bound_method == "clip":
                 act = np.clip(act, -1, 1)
@@ -148,7 +148,7 @@ def map_action(self, act: Union[Batch, np.ndarray]) -> Union[Batch, np.ndarray]:
                 assert np.all(act >= -1) and np.all(act <= 1), \
                     "action scaling only accepts raw action range = [-1, 1]"
                 low, high = self.action_space.low, self.action_space.high
-                act = low + (high - low) * (act + 1) / 2
+                act = low + (high - low) * (act + 1.) / 2.
         return act
 
     def process_fn(

tianshou/policy/modelfree/a2c.py

@@ -37,7 +37,7 @@ class A2CPolicy(PGPolicy):
         either "clip" (for simply clipping the action), "tanh" (for applying tanh
         squashing) for now, or empty string for no bounding. Default to "clip".
     :param Optional[gym.Space] action_space: env's action space, mandatory if you want
-        to use option action_scaling/action_bound_method. Default to None.
+        to use option 'action_scaling' or 'action_bound_method'. Default to None.
 
     .. seealso::
 

tianshou/policy/modelfree/ddpg.py

@@ -29,7 +29,7 @@ class DDPGPolicy(BasePolicy):
         either "clip" (for simply clipping the action), "tanh" (for applying tanh
         squashing) for now, or empty string for no bounding. Default to "clip".
     :param Optional[gym.Space] action_space: env's action space, mandatory if you want
-        to use option action_scaling/action_bound_method. Default to None.
+        to use option 'action_scaling' or 'action_bound_method'. Default to None.
 
     .. seealso::
 

tianshou/policy/modelfree/pg.py

@@ -21,7 +21,7 @@ class PGPolicy(BasePolicy):
         either "clip" (for simply clipping the action), "tanh" (for applying tanh
         squashing) for now, or empty string for no bounding. Default to "clip".
     :param Optional[gym.Space] action_space: env's action space, mandatory if you want
-        to use option action_scaling/action_bound_method. Default to None.
+        to use option 'action_scaling' or 'action_bound_method'. Default to None.
 
     .. seealso::
 

tianshou/policy/modelfree/ppo.py

@@ -42,7 +42,7 @@ class PPOPolicy(PGPolicy):
         either "clip" (for simply clipping the action), "tanh" (for applying tanh
         squashing) for now, or empty string for no bounding. Default to "clip".
     :param Optional[gym.Space] action_space: env's action space, mandatory if you want
-        to use option action_scaling/action_bound_method. Default to None.
+        to use option 'action_scaling' or 'action_bound_method'. Default to None.
 
     .. seealso::
 

tianshou/policy/modelfree/sac.py

@@ -40,7 +40,7 @@ class SACPolicy(DDPGPolicy):
         either "clip" (for simply clipping the action), "tanh" (for applying tanh
         squashing) for now, or empty string for no bounding. Default to "tanh".
     :param Optional[gym.Space] action_space: env's action space, mandatory if you want
-        to use option action_scaling/action_bound_method. Default to None.
+        to use option 'action_scaling' or 'action_bound_method'. Default to None.
 
     .. seealso::
 

tianshou/policy/modelfree/td3.py

@@ -38,7 +38,7 @@ class TD3Policy(DDPGPolicy):
         either "clip" (for simply clipping the action), "tanh" (for applying tanh
         squashing) for now, or empty string for no bounding. Default to "clip".
     :param Optional[gym.Space] action_space: env's action space, mandatory if you want
-        to use option action_scaling/action_bound_method. Default to None.
+        to use option 'action_scaling' or 'action_bound_method'. Default to None.
 
     .. seealso::