dropping logging actions for now

FilipinoGambino · Feb 2, 2024 · e793cdb · e793cdb
1 parent 3f32bb6
commit e793cdb
Show file tree

Hide file tree

Showing 4 changed files with 8 additions and 5 deletions.
diff --git a/conf/new_beginnings.yaml b/conf/new_beginnings.yaml
@@ -36,6 +36,10 @@ obs_space_kwargs: {}
 reward_space_kwargs: {}
 debug: True
 
+# Environment params
+adversary: random
+#adversary: negamax
+
 # Model params
 act_space: BasicActionSpace
 obs_space: BasicObsSpace

diff --git a/connectx/connectx_gym/__init__.py b/connectx/connectx_gym/__init__.py
@@ -31,7 +31,7 @@ def create_env(flags, device: torch.device, teacher_flags: Optional = None, seed
             act_space=flags.act_space(),
             obs_space=create_flexible_obs_space(flags, teacher_flags),
             player_id=flags.player_id,
-            seed=seed
+            adversary=flags.adversary
         )
         reward_space = create_reward_space(flags)
         env = RewardSpaceWrapper(env, reward_space)

diff --git a/connectx/connectx_gym/connectx_env.py b/connectx/connectx_gym/connectx_env.py
@@ -20,13 +20,12 @@ def __init__(
             act_space: BaseActSpace,
             obs_space: BaseObsSpace,
             player_id: int,
-            seed: Optional[int] = 42,
+            adversary: str,
     ):
         super(ConnectFour, self).__init__()
         self.env = make("connectx", debug=True)
         self.player_id = player_id
-        # players = ["negamax", "negamax"]
-        players = ["random", "random"]
+        players = [adversary, adversary]
         players[player_id] = None
         self.trainer = self.env.train(players)
 

diff --git a/connectx/torchbeast/monobeast.py b/connectx/torchbeast/monobeast.py
@@ -492,7 +492,7 @@ def learn(
                     for key, val in batch["info"].items()
                     if key.startswith("LOGGING_") and "ACTIONS_" not in key
                 },
-                "Actions": batch['actions'],
+                # "Actions": batch['actions'],
                 "Loss": {
                     "vtrace_pg_loss": vtrace_pg_loss.detach().item(),
                     "upgo_pg_loss": upgo_pg_loss.detach().item(),