update

omnisafe/adapter/crabs_adapter.py

@@ -93,9 +93,9 @@ def eval_policy(  # pylint: disable=too-many-locals
 
             logger.store(
                 {
-                    'Metrics/TestEpRet': ep_ret,
-                    'Metrics/TestEpCost': ep_cost,
-                    'Metrics/TestEpLen': ep_len,
+                    'Metrics/RawPolicyEpRet': ep_ret,
+                    'Metrics/RawPolicyEpCost': ep_cost,
+                    'Metrics/RawPolicyEpLen': ep_len,
                 },
             )
 

omnisafe/algorithms/off_policy/crabs.py

@@ -167,6 +167,9 @@ def _init_log(self) -> None:
         what_to_save['obs_normalizer'] = self.normalizer
         self._logger.setup_torch_saver(what_to_save)
         self._logger.torch_save()
+        self._logger.register_key('Metrics/RawPolicyEpRet', window_length=50)
+        self._logger.register_key('Metrics/RawPolicyEpCost', window_length=50)
+        self._logger.register_key('Metrics/RawPolicyEpLen', window_length=50)
 
     def _init(self) -> None:
         """The initialization of the algorithm.
@@ -278,7 +281,7 @@ def learn(self):
 
             eval_start = time.time()
             self._env.eval_policy(
-                episode=self._cfgs.train_cfgs.eval_episodes,
+                episode=self._cfgs.train_cfgs.raw_policy_episodes,
                 agent=self._actor_critic,
                 logger=self._logger,
             )
@@ -326,7 +329,7 @@ def learn(self):
                 if t % 1000 == 0:
                     eval_start = time.time()
                     self._env.eval_policy(
-                        episode=self._cfgs.train_cfgs.eval_episodes,
+                        episode=self._cfgs.train_cfgs.raw_policy_episodes,
                         agent=self.mean_policy,  # type: ignore
                         logger=self._logger,
                     )

omnisafe/configs/off-policy/CRABS.yaml

@@ -29,7 +29,8 @@ defaults:
     # number of training epochs
     num_epochs: 101
     # number of evaluate episodes
-    eval_episodes: 10
+    eval_episodes: 0
+    raw_policy_episodes: 10
   # algorithm configurations
   algo_cfgs:
     # number of steps to update the policy

tests/.coveragerc

@@ -4,9 +4,11 @@ omit =
     ../docs/*
     ../examples/*
     ../tutorials/*
-    ../common/control_barrier_function/crabs/*
-    ../envs/classic_control/*
-    ../algorithms/off_policy/crabs.py
+    ../omnisafe/common/control_barrier_function/crabs/*
+    ../omnisafe/envs/classic_control/*
+    ../omnisafe/algorithms/off_policy/crabs.py
+    ../omnisafe/adapter/crabs_adapter.py
+    ../omnisafe/envs/crabs_env.py
 
 [report]
 exclude_lines =