DQN让它先看答案

Duel_Double_DQN/main.py

@@ -123,10 +123,12 @@ def main():
             while not done:
                 # e-greedy exploration
                 if total_steps < opt.random_steps:
-                    if s[0] == 0:
-                        a = np.random.randint(0, 10)
-                    else:
-                        a = np.random.randint(10, 14)
+                    # if s[0] == 0:
+                    #     a = np.random.randint(0, 10)
+                    # else:
+                    #     a = np.random.randint(10, 14)
+                    action_series = [0, 0, 3, 0, 10]
+                    a = action_series[total_steps % 5]
                 else:
                     a = agent.select_action(s, deterministic=False)
                 s_next, r, dw, tr, info = env.step(a)

env_dis.py

@@ -275,7 +275,7 @@ class PartitionMazeEnv(gym.Env):
                 # print(self.partition_values)
                 # print(self.car_traj)
                 reward += self.BASE_LINE / T * 1000
-                print(reward)
+                # print(reward)
             elif done and self.step_count >= self.MAX_STEPS:
                 reward += -1000