Create deterministic_sweep.py

sweeps/deterministic_sweep.py

@@ -0,0 +1,43 @@
+def main():
+    wandb.init(project="Dune-Q")
+
+    n_states = Caladan.observation_space.n
+    n_actions = Caladan.action_space.n
+
+    print(f"Observation space: {n_states}, Action space: {n_actions}")
+
+    # Генерируем случайное значение seed на основе текущего времени
+    seed = int(time.time())
+
+    agent = PaulAtreides(
+        n_states,
+        n_actions,
+        f_model=Model,
+        lr=0.05,
+        gamma=0.975,
+        eps=0.1,
+        seed=seed
+    )
+
+    avg_returns = train(
+        Caladan,
+        agent,
+        on_model_updates=0,
+        seed=seed,
+        n_episodes=3001,
+    )
+
+# Define the sweep configuration
+sweep_config = {
+    "method": "grid",
+    "parameters": {
+        "n_planning_steps": {"values": [0, 5, 10]},
+        "n_dreaming_steps": {"values": [1, 3, 5]}  # Add dreaming steps parameter
+    }
+}
+
+# Start the sweep
+PaulTravel = wandb.sweep(sweep=sweep_config, project="Dune-Q I")
+
+# Run the sweep
+wandb.agent(PaulTravel, function=main, count=5)