Type fixes

Author: RedTachyon

coltra/groups.py

@@ -108,7 +108,7 @@ def act(
         obs_dict: dict[AgentName, Observation],
         deterministic: bool = False,
         get_value: bool = False,
-        state_dict: dict[AgentName, tuple] = None,
+        state_dict: Optional[dict[AgentName, tuple]] = None,
     ):
         if len(obs_dict) == 0:
             return {}, {}, {}
@@ -174,7 +174,7 @@ def evaluate(
         self,
         obs_batch: dict[PolicyName, Observation],
         action_batch: dict[PolicyName, Action],
-        state: dict[PolicyName, tuple] = None,
+        state: Optional[dict[PolicyName, tuple]] = None,
     ) -> dict[PolicyName, Tuple[Tensor, Tensor, Tensor]]:
 
         obs = obs_batch[self.policy_name]
@@ -381,13 +381,13 @@ def value(
         family_actions, crowd_actions = split_dict(action_batch)
 
         family_obs, family_keys = pack(family_obs)
-        family_values = self.family_agent.value(family_obs)
+        family_values = self.family_agent.value(family_obs, ())
 
         augment_observations(crowd_obs, family_actions)
 
         crowd_obs, crowd_keys = pack(crowd_obs)
 
-        crowd_values = self.agent.value(crowd_obs)
+        crowd_values = self.agent.value(crowd_obs, ())
 
         crowd_values = unpack(crowd_values, crowd_keys)
         family_values = unpack(family_values, family_keys)

coltra/wrappers/agent_wrappers.py

@@ -66,8 +66,12 @@ def act(
         norm_obs = self.normalize_obs(obs_batch)
         return self.agent.act(norm_obs, state_batch, deterministic, get_value, **kwargs)
 
-    def value(self, obs_batch: Observation, **kwargs) -> Tensor:
-        return self.agent.value(self.normalize_obs(obs_batch))
+    def value(
+        self, obs_batch: Observation, state_batch: tuple = (), **kwargs
+    ) -> tuple[Tensor, tuple]:
+        return self.agent.value(
+            self.normalize_obs(obs_batch), state_batch=state_batch, **kwargs
+        )
 
     def evaluate(
         self, obs_batch: Observation, action_batch: Action
@@ -113,9 +117,13 @@ def unnormalize_value(self, value: Tensor):
         return self._ret_var * value + self._ret_mean
 
     def value(
-        self, obs_batch: Observation, real_value: bool = False, **kwargs
-    ) -> Tensor:
-        value = self.agent.value(obs_batch)
+        self,
+        obs_batch: Observation,
+        state_batch: tuple = (),
+        real_value: bool = False,
+        **kwargs,
+    ) -> tuple[Tensor, tuple]:
+        value, state = self.agent.value(obs_batch, state_batch=state_batch, **kwargs)
         if real_value:
             value = self.unnormalize_value(value)
-        return value
+        return value, state