def main(_):
pygame.init()
try:
pygame.mixer.quit()
except NotImplementedError:
pass
pygame.display.set_caption('Fast Language Learning Tasks Human Agent')
if 'with_distractors' in FLAGS.level_name: # for the tasks from the HTM paper
episode_length_seconds = 450.0
else:
episode_length_seconds = 120.0
env_settings = dm_fast_mapping.EnvironmentSettings(
seed=FLAGS.seed,
level_name=FLAGS.level_name,
episode_length_seconds=episode_length_seconds)
with dm_fast_mapping.load_from_docker(name=FLAGS.docker_image_name,
settings=env_settings) as env:
screen = pygame.display.set_mode(
(int(FLAGS.screen_size[0]), int(FLAGS.screen_size[1])))
rgb_spec = env.observation_spec()['RGB_INTERLEAVED']
surface = pygame.Surface((rgb_spec.shape[1], rgb_spec.shape[0]))
actions = _NO_ACTION
score = 0
clock = pygame.time.Clock()
while True:
# Do not close with CTRL-C as otherwise the docker container may be left
# running on exit.
for event in pygame.event.get():
if event.type == pygame.QUIT:
return
elif event.type == pygame.KEYDOWN:
if event.key == pygame.K_ESCAPE:
return
key_actions = _KEYS_TO_ACTION.get(event.key, {})
for name, action in key_actions.items():
actions[name] += action
elif event.type == pygame.KEYUP:
key_actions = _KEYS_TO_ACTION.get(event.key, {})
for name, action in key_actions.items():
actions[name] -= action
timestep = env.step(actions)
frame = np.swapaxes(timestep.observation['RGB_INTERLEAVED'], 0, 1)
font = pygame.font.SysFont('Sans', 10)
pygame.surfarray.blit_array(surface, frame)
text = font.render(timestep.observation['TEXT'], True, (0, 0, 0))
surface.blit(text, (0, 0))
pygame.transform.smoothscale(surface, screen.get_size(), screen)
pygame.display.update()
if timestep.reward:
score += timestep.reward
logging.info('Total score: %1.1f, reward: %1.1f', score,
timestep.reward)
clock.tick(_FRAMES_PER_SECOND)
def main(_):
if 'with_distractors' in FLAGS.level_name: # for the tasks from the HTM paper
episode_length_seconds = 450.0
else:
episode_length_seconds = 120.0
env_settings = dm_fast_mapping.EnvironmentSettings(
seed=FLAGS.seed,
level_name=FLAGS.level_name,
episode_length_seconds=episode_length_seconds)
with dm_fast_mapping.load_from_docker(name=FLAGS.docker_image_name,
settings=env_settings) as env:
agent = RandomAgent(env.action_spec())
timestep = env.reset()
score = 0
while not timestep.last():
action = agent.act()
timestep = env.step(action)
if timestep.reward:
score += timestep.reward
logging.info('Total score: %1.1f, reward: %1.1f', score,
timestep.reward)
default="fast_slow/fast_map_three_objs")
parser.add_argument("--episode_max_time", type=int, default=1202)
parser.add_argument("--language_dim", type=int, default=108)
parser.add_argument("--num_embeddings", type=int, default=50)
parser.add_argument("--embedding_dim", type=int, default=32)
parser.add_argument("--memory_hidden_dim", type=int, default=512)
parser.add_argument("--action_dim", type=int, default=1)
parser.add_argument("--max_episodes", type=int, default=100)
parser.add_argument("--train_steps", type=int, default=1)
parser.add_argument("--eval_freq", type=int, default=5)
parser.add_argument("--start_episodes", type=int, default=10)
parser.add_argument("--buffer_size", type=int, default=1000)
parser.add_argument("--batch_size", type=int, default=1)
args = parser.parse_args()
settings = dm_fast_mapping.EnvironmentSettings(seed=0,
level_name=args.level_name)
env = dm_fast_mapping.load_from_docker(settings)
wrapped_env = FastSlowEnvWrapper(env)
temp_timestep = wrapped_env.reset()
vision_dim = temp_timestep.observation["RGB_INTERLEAVED"].shape
tokenizer = BertWordPieceTokenizer(
DIR_PATH + "/embedding_files/bert-base-uncased-vocab.txt",
lowercase=True)
tokenizer.enable_padding(pad_id=3,
length=args.language_dim,
pad_token="[PAD]")
policy = Agent(
language_dim=args.language_dim,
def test_load_level(self, level_name):
self.assertIsNotNone(
dm_fast_mapping.load_from_disk(
FLAGS.path,
settings=dm_fast_mapping.EnvironmentSettings(
seed=123, level_name=level_name)))
def make_object_under_test(self):
return dm_fast_mapping.load_from_disk(
FLAGS.path,
settings=dm_fast_mapping.EnvironmentSettings(
seed=123,
level_name='architecture_comparison/fast_map_three_objs'))
def test_load_level(self, level_name):
self.assertIsNotNone(
dm_fast_mapping.load_from_docker(
name=FLAGS.docker_image_name,
settings=dm_fast_mapping.EnvironmentSettings(
seed=123, level_name=level_name)))