def test_agents(self):
params = ParameterServer()
params["ML"]["BaseAgent"]["NumSteps"] = 2
params["ML"]["BaseAgent"]["MaxEpisodeSteps"] = 2
bp = DiscreteHighwayBlueprint(params,
number_of_senarios=10,
random_seed=0)
env = SingleAgentRuntime(blueprint=bp, render=False)
env._observer = NearestAgentsObserver(params)
env._action_wrapper = BehaviorDiscreteMacroActionsML(params)
fqf_agent = FQFAgent(agent_save_dir="./save_dir",
env=env,
params=params)
fqf_agent.train_episode()
fqf_agent.save(checkpoint_type="best")
fqf_agent.save(checkpoint_type="last")
loaded_agent = FQFAgent(agent_save_dir="./save_dir",
checkpoint_load="best")
loaded_agent2 = FQFAgent(agent_save_dir="./save_dir",
checkpoint_load="last")
loaded_agent_with_env = FQFAgent(env=env,
agent_save_dir="./save_dir",
checkpoint_load="last")
loaded_agent_with_env.train_episode()
self.assertEqual(loaded_agent.ml_behavior.action_space.n,
fqf_agent.ml_behavior.action_space.n)
self.assertEqual(loaded_agent.ent_coef, fqf_agent.ent_coef)
return
def test_agent_wrapping(self):
params = ParameterServer()
env = gym.make("highway-v1", params=params)
env.reset()
# agent = IQNAgent(env=env, test_env=env, params=params)
params["ML"]["BaseAgent"]["MaxEpisodeSteps"] = 2
params["ML"]["BaseAgent"]["NumEvalEpisodes"] = 2
agent = FQFAgent(env=env, params=params)
agent.train_episode()
agent.evaluate()
def test_agent_wrapping(self):
params = ParameterServer()
env = gym.make("highway-v1", params=params)
env.reset()
params["ML"]["BaseAgent"]["MaxEpisodeSteps"] = 2
params["ML"]["BaseAgent"]["NumEvalEpisodes"] = 2
train_bench = TrainingBenchmarkDatabase()
agent = FQFAgent(env=env,
agent_save_dir="./save_dir",
params=params,
training_benchmark=train_bench)
agent.train_episode()
agent.evaluate()