Fix in IDDPG online gymnasium mamujoco.

instadeepai · Dec 6, 2023 · cc24c4d · cc24c4d
1 parent b250201
commit cc24c4d
Show file tree

Hide file tree

Showing 5 changed files with 9 additions and 6 deletions.
diff --git a/examples/tf2/online/iddpg_mamujoco.py b/examples/tf2/online/iddpg_mamujoco.py
@@ -15,7 +15,7 @@
 from og_marl.environments.wrappers import PadObsandActs, Dtype
 from og_marl.loggers import WandbLogger
 from og_marl.tf2.systems.iddpg import IDDPGSystem
-from og_marl.environments.old_mamujoco import MAMuJoCo
+from og_marl.environments.gymnasium_mamujoco import MAMuJoCo
 from og_marl.replay_buffers import SequenceCPPRB
 
 env = MAMuJoCo("4ant")
@@ -24,7 +24,7 @@
 
 env = Dtype(env, "float32")
 
-logger = WandbLogger(entity="claude_formanek")
+logger = WandbLogger()
 
 system = IDDPGSystem(env, logger)
 

diff --git a/install_environments/gymnasium_robotics.sh b/install_environments/gymnasium_robotics.sh
@@ -0,0 +1 @@
+pip install -r install_environments/requirements/gymnasium_mamujoco.txt
diff --git a/install_environments/requirements/gymnasium_mamujoco.txt b/install_environments/requirements/gymnasium_mamujoco.txt
@@ -0,0 +1,2 @@
+mujoco
+gymnasium_robotics
diff --git a/og_marl/environments/wrappers.py b/og_marl/environments/wrappers.py
@@ -56,14 +56,14 @@ def __init__(self, environment):
         self._environment = environment
 
         self._obs_dim = 0
-        self._act_dim = 0
+        self._num_actions = 0
 
         for agent in self._environment.possible_agents:
             act_dim = self._environment.action_spaces[agent].shape[0]
             obs_dim = self._environment.observation_spaces[agent].shape[0]
 
-            if act_dim > self._act_dim:
-                self._act_dim = act_dim
+            if act_dim > self._num_actions:
+                self._num_actions = act_dim
 
             if obs_dim > self._obs_dim:
                 self._obs_dim = obs_dim

diff --git a/og_marl/tf2/systems/iddpg.py b/og_marl/tf2/systems/iddpg.py
@@ -206,7 +206,7 @@ def __init__(
         critic_learning_rate=3e-4,
         policy_learning_rate=1e-3,
         add_agent_id_to_obs=True,
-        random_exploration_timesteps=50_000
+        random_exploration_timesteps=50_000 
     ):
 
         super().__init__(