add error screen

connectfour/app.py

@@ -14,6 +14,7 @@ from connectfour.training.wrappers import Connect4Env
 POLICY_ID = "learned_v5"
 
 # poetry export -f requirements.txt --output requirements.txt --without-hashes
+# gradio connectfour/app.py
 
 
 class Connect4:
@@ -55,6 +56,9 @@ class Connect4:
         self.algo.restore(checkpoint)
 
     def play(self, action=None):
+        if self.has_erroneous_state():
+            return self.blue_screen()
+
         if self.human != self.player_id:
             action = self.algo.compute_single_action(
                 self.obs[self.player_id], policy_id=POLICY_ID
@@ -73,8 +77,11 @@ class Connect4:
 
     @property
     def render_and_state(self):
-        end_message = "End of the game"
         if self.done:
+            if hasattr(self, "reward") and self.human not in self.reward:
+                return self.blue_screen()
+
+            end_message = "End of the game"
             if self.reward[self.human] > 0:
                 end_message += ": You WIN !!"
             elif self.reward[self.human] < 0:
@@ -83,6 +90,12 @@ class Connect4:
 
         return self.env.render(), "Game On"
 
+    def blue_screen(self):
+        with open("error-screen.npy", "rb") as f:
+            error_screen = np.load(f)
+
+        return (error_screen, "Restart the Game")
+
     @property
     def player_id(self):
         return list(self.obs.keys())[0]
@@ -91,6 +104,11 @@ class Connect4:
     def legal_moves(self):
         return np.arange(7)[self.obs[self.player_id]["action_mask"] == 1]
 
+    def has_erroneous_state(self):
+        if len(list(self.obs.keys())) == 0:
+            return True
+        return False
+
 
 demo = gr.Blocks()
 

connectfour/training/train.py

@@ -27,7 +27,7 @@ def get_cli_args():
     Create CLI parser and return parsed arguments
 
     python connectfour/training/train.py --num-cpus 4 --num-gpus 1 --stop-iters 10 --win-rate-threshold 0.50
-    python connectfour/training/train.py --num-gpus 1 --stop-iters 10 --win-rate-threshold 0.50
+    python connectfour/training/train.py --num-gpus 1 --stop-iters 1 --win-rate-threshold 0.50
     python connectfour/training/train.py --num-cpus 5 --num-gpus 1 --stop-iters 200
     """
     parser = argparse.ArgumentParser()
@@ -68,7 +68,10 @@ if __name__ == "__main__":
     args = get_cli_args()
 
     ray.init(
-        num_cpus=args.num_cpus or None, num_gpus=args.num_gpus, include_dashboard=False
+        num_cpus=args.num_cpus or None,
+        num_gpus=args.num_gpus,
+        include_dashboard=False,
+        resources={"accelerator_type:RTX": 1},
     )
 
     # define how to make the environment
@@ -126,6 +129,8 @@ if __name__ == "__main__":
                     "win_rate": "win_rate",
                     "league_size": "league_size",
                 },
+                mode="max",
+                metric="win_rate",
                 sort_by_metric=True,
             ),
             checkpoint_config=air.CheckpointConfig(
@@ -135,6 +140,9 @@ if __name__ == "__main__":
         ),
     ).fit()
 
-    print("Best checkpoint", results.get_best_result().checkpoint)
+    print(
+        "Best checkpoint",
+        results.get_best_result(metric="win_rate", mode="max").checkpoint,
+    )
 
     ray.shutdown()

error-screen.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b72c5a148f41583927cd127d1d2b51073adec2ebd33ace7d4c074142d16d992
+size 4316726