High-level policy trained for 1m steps with 3-hidden layers.

high_level_policy_main.py

@@ -47,8 +47,8 @@ def high_level_policy_training(nb_steps=25000,
         nb_actions=options.get_number_of_nodes(),
         target_model_update=1e-3,
         delta_clip=100,
-        low_level_policies=options.maneuvers,
-        gamma=1)
+        low_level_policies=options.maneuvers)
+        #gamma=1)
 
     if load_weights:
         agent.load_model(save_path)
@@ -78,8 +78,7 @@ def high_level_policy_testing(nb_episodes_for_test=100,
     agent = DQNLearner(
         input_shape=(50, ),
         nb_actions=options.get_number_of_nodes(),
-        low_level_policies=options.maneuvers,
-        gamma=1)
+        low_level_policies=options.maneuvers)
 
     if pretrained:
         trained_agent_file = "backends/trained_policies/highlevel/" + trained_agent_file
@@ -101,8 +100,7 @@ def evaluate_high_level_policy(nb_episodes_for_test=100,
     agent = DQNLearner(
         input_shape=(50, ),
         nb_actions=options.get_number_of_nodes(),
-        low_level_policies=options.maneuvers,
-        gamma=1)
+        low_level_policies=options.maneuvers)
 
     if pretrained:
         trained_agent_file = "backends/trained_policies/highlevel/" + trained_agent_file