Between Tuesday August 20th, 5:00pm and Thursday August 22nd, 8:00am git.uwaterloo.ca will be down for an upgrade to version 10.8.7.

7a4ea75b by Jaeyoung Lee

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

1 parent 77d61b11
......@@ -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
......