Making gym a dependency of ml-agents-envs

ml-agents-envs/mlagents_envs/gym_to_unity_wrapper.py

 )
 from mlagents_envs.exception import UnityActionException, UnityObservationException
 
-from typing import List, Tuple
+from typing import List, Tuple, Union, Optional
-try:
-    import gym
-
-    _GYM_IMPORTED = True
-except ImportError:
-    _GYM_IMPORTED = False
+import gym
 
 
 class GymToUnityWrapper(BaseEnv):
-    def __init__(self, gym_env, name=None):
-        if not _GYM_IMPORTED:
-            raise RuntimeError(
-                "gym is not installed, gym required to use the GymToUnityWrapper"
-            )
+    def __init__(self, gym_env: gym.Env, name: Optional[str] = None):
+        """
+        Wrapper construction. Creates an implementation of a Unity BaseEnv from a gym
+        environment.
+        :gym.Env gym_env: The gym environment that will be wrapped.
+        :str name: [Optional] The name of the gym environment. This will become the
+        name of the behavior for the BaseEnv.
+        """
-        self._behavior_name = name
-        if self._behavior_name is None:
+        if name is None:
+        else:
+            self._behavior_name = name
-        action_shape = 0
+        action_shape: Union[Tuple[int, ...], int] = 0
         if isinstance(self._gym_env.action_space, gym.spaces.Box):
             action_type = ActionType.CONTINUOUS
             action_shape = np.prod(self._gym_env.action_space.shape)
             action_shape=action_shape,
         )
         self._g_action: np.ndarray = None
-        self._current_steps: Tuple[DecisionSteps, TerminalSteps] = (None, None)
+        self._current_steps: Tuple[DecisionSteps, TerminalSteps] = (
+            DecisionSteps.empty(self._behavior_specs),
+            TerminalSteps.empty(self._behavior_specs),
+        )
 
     def step(self) -> None:
         if self._first_message:

ml-agents-envs/setup.py

     packages=find_packages(exclude=["*.tests", "*.tests.*", "tests.*", "tests"]),
     zip_safe=False,
     install_requires=[
+        "gym",
         "cloudpickle",
         "grpcio>=1.11.0",
         "numpy>=1.14.1,<2.0",

ml-agents-envs/mlagents_envs/tests/test_gym_to_unity_wrapper.py

+from mlagents_envs.gym_to_unity_wrapper import GymToUnityWrapper
+from mlagents_envs.base_env import ActionType
+import gym
+
+import pytest
+
+
+GYM_ENVS = ["CartPole-v1", "MountainCar-v0"]
+
+
+@pytest.mark.parametrize("name", GYM_ENVS, ids=GYM_ENVS)
+def test_creation(name):
+    env = GymToUnityWrapper(gym.make(name), name)
+    env.close()
+
+
+@pytest.mark.parametrize("name", GYM_ENVS, ids=GYM_ENVS)
+def test_specs(name):
+    gym_env = gym.make(name)
+    env = GymToUnityWrapper(gym_env, name)
+    assert env.get_behavior_names()[0] == name
+    if isinstance(gym_env.action_space, gym.spaces.Box):
+        assert env.get_behavior_spec(name).action_type == ActionType.CONTINUOUS
+    elif isinstance(gym_env.action_space, gym.spaces.Discrete):
+        assert env.get_behavior_spec(name).action_type == ActionType.DISCRETE
+    else:
+        raise NotImplementedError("Test for this action space type not implemented")
+    env.close()
+
+
+@pytest.mark.parametrize("name", GYM_ENVS, ids=GYM_ENVS)
+def test_steps(name):
+    env = GymToUnityWrapper(gym.make(name), name)
+    spec = env.get_behavior_spec(name)
+    env.reset()
+    for _ in range(200):
+        d_steps, t_steps = env.get_steps(name)
+        env.set_actions(name, spec.create_empty_action(len(d_steps)))
+        env.step()
+    env.close()

ml-agents-envs/mlagents_envs/tests/test_gym_wrapper.py

-from mlagents_envs.gym_to_unity_wrapper import GymToUnityWrapper
-from mlagents_envs.base_env import ActionType
-
-try:
-    import gym
-
-    _GYM_IMPORTED = True
-except ImportError:
-    _GYM_IMPORTED = False
-
-import pytest
-
-GYM_ENVS = ["CartPole-v1", "MountainCar-v0"]
-
-
-@pytest.mark.parametrize("name", GYM_ENVS, ids=GYM_ENVS)
-def test_creation(name):
-    if not _GYM_IMPORTED:
-        raise RuntimeError(
-            "gym is not installed, gym required to test the GymToUnityWrapper"
-        )
-    env = GymToUnityWrapper(gym.make(name), name)
-    env.close()
-
-
-@pytest.mark.parametrize("name", GYM_ENVS, ids=GYM_ENVS)
-def test_specs(name):
-    if not _GYM_IMPORTED:
-        raise RuntimeError(
-            "gym is not installed, gym required to test the GymToUnityWrapper"
-        )
-    gym_env = gym.make(name)
-    env = GymToUnityWrapper(gym_env, name)
-    assert env.get_behavior_names()[0] == name
-    if isinstance(gym_env.action_space, gym.spaces.Box):
-        assert env.get_behavior_spec(name).action_type == ActionType.CONTINUOUS
-    elif isinstance(gym_env.action_space, gym.spaces.Discrete):
-        assert env.get_behavior_spec(name).action_type == ActionType.DISCRETE
-    else:
-        raise NotImplementedError("Test for this action space type not implemented")
-    env.close()
-
-
-@pytest.mark.parametrize("name", GYM_ENVS, ids=GYM_ENVS)
-def test_steps(name):
-    if not _GYM_IMPORTED:
-        raise RuntimeError(
-            "gym is not installed, gym required to test the GymToUnityWrapper"
-        )
-    env = GymToUnityWrapper(gym.make(name), name)
-    spec = env.get_behavior_spec(name)
-    env.reset()
-    for _ in range(200):
-        d_steps, t_steps = env.get_steps(name)
-        env.set_actions(name, spec.create_empty_action(len(d_steps)))
-        env.step()
-    env.close()