def example(gui):
train_env = gym.make('Traffic-Simple-cli-v0')
agent = build_agent(train_env)
path = "output/traffic/simple/dqn"
explorer = EpsilonExplorer(agent, epsilon=0.5, decay=5e-7)
if gui:
def test_env_func():
return gym.make('Traffic-Simple-gui-v0')
else:
def test_env_func():
return train_env
runner = SimpleRunner(max_steps_per_episode=1000)
video_callable = None if gui else False
run_agent(runner=runner,
agent=explorer,
test_agent=explorer,
train_env=train_env,
test_env_func=test_env_func,
nb_episodes=500,
test_nb_episodes=10,
nb_epoch=100,
path=path,
video_callable=video_callable)
train_env = gym.make('CartPole-v0')
agent = DQN(train_env.observation_space,
train_env.action_space,
memory_size=4,
replay_size=128)
explorer = EpsilonExplorer(agent, epsilon=0.3, decay=2e-5)
path = "output/cartpole/dqn"
print("Q")
agent.Q.summary()
print("training_model")
agent.training_model.summary()
def test_env_func():
return train_env
runner = SimpleRunner(max_steps_per_episode=1000)
run_agent(runner=runner,
agent=explorer,
test_agent=agent,
train_env=train_env,
test_env_func=test_env_func,
nb_episodes=100,
test_nb_episodes=100,
nb_epoch=25,
path=path)