Merge branch 'master' into sde

docs/misc/changelog.rst

@@ -37,6 +37,7 @@ Others:
 - Split the ``collect_rollout()`` method for off-policy algorithms
 - Added ``_on_step()`` for off-policy base class
 - Optimized replay buffer size by removing the need of ``next_observations`` numpy array
+- Ignored errors from newer pytype version
 
 Documentation:
 ^^^^^^^^^^^^^^

stable_baselines3/common/base_class.py

@@ -330,7 +330,8 @@ class BaseAlgorithm(ABC):
             env = data["env"]
 
         # noinspection PyArgumentList
-        model = cls(policy=data["policy_class"], env=env, device='auto', _init_setup_model=False)
+        model = cls(policy=data["policy_class"], env=env,
+                    device='auto', _init_setup_model=False)  # pytype: disable=not-instantiable,wrong-keyword-args
 
         # load parameters
         model.__dict__.update(data)
@@ -350,7 +351,7 @@ class BaseAlgorithm(ABC):
         # Sample gSDE exploration matrix, so it uses the right device
         # see issue #44
         if model.use_sde:
-            model.policy.reset_noise()
+            model.policy.reset_noise()  # pytype: disable=attribute-error
         return model
 
     def set_random_seed(self, seed: Optional[int] = None) -> None:

stable_baselines3/common/policies.py

@@ -126,7 +126,7 @@ class BaseModel(nn.Module, ABC):
         device = get_device(device)
         saved_variables = th.load(path, map_location=device)
         # Create policy object
-        model = cls(**saved_variables['data'])
+        model = cls(**saved_variables['data'])  # pytype: disable=not-instantiable
         # Load weights
         model.load_state_dict(saved_variables['state_dict'])
         model.to(device)