def run_configuration(argv):
params = ParameterServer(
filename="examples/example_params/iqn_params.json")
params["ML"]["BaseAgent"][
"SummaryPath"] = "/home/mansoor/Study/Werkstudent/fortiss/code/bark-ml/summaries"
params["ML"]["BaseAgent"][
"CheckpointPath"] = "/home/mansoor/Study/Werkstudent/fortiss/code/bark-ml/checkpoints"
env = gym.make(FLAGS.env, params=params)
agent = IQNAgent(env=env, test_env=env, params=params)
if FLAGS.load and params["ML"]["BaseAgent"]["CheckpointPath"]:
agent.load_models(
os.path.join(params["ML"]["BaseAgent"]["CheckpointPath"], "best"))
if FLAGS.mode == "train":
agent.run()
elif FLAGS.mode == "visualize":
agent.visualize()
elif FLAGS.mode == "evaluate":
# writes evaluaion data using summary writer in summary path
agent.evaluate()
else:
raise Exception("Invalid argument for --mode")
def test_agents(self):
params = ParameterServer()
params["ML"]["BaseAgent"]["NumSteps"] = 2
params["ML"]["BaseAgent"]["MaxEpisodeSteps"] = 2
bp = DiscreteHighwayBlueprint(params, num_scenarios=10, random_seed=0)
env = SingleAgentRuntime(blueprint=bp, render=False)
# IQN Agent
iqn_agent = IQNAgent(env=env, test_env=env, params=params)
env.ml_behavior = iqn_agent
self.assertEqual(env.ml_behavior.set_action_externally, False)
iqn_agent.run()
self.assertEqual(env.ml_behavior.set_action_externally, True)
# FQF Agent
fqf_agent = FQFAgent(env=env, test_env=env, params=params)
env.ml_behavior = fqf_agent
self.assertEqual(env.ml_behavior.set_action_externally, False)
fqf_agent.run()
self.assertEqual(env.ml_behavior.set_action_externally, True)
# QRDQN Agent
qrdqn_agent = QRDQNAgent(env=env, test_env=env, params=params)
env.ml_behavior = qrdqn_agent
self.assertEqual(env.ml_behavior.set_action_externally, False)
qrdqn_agent.run()
self.assertEqual(env.ml_behavior.set_action_externally, True)