def __init__(self, environment_spec, length, initial_frame_chooser):
"""Batch of environments inside the TensorFlow graph."""
observ_space = utils.get_observation_space(environment_spec)
initial_frames_problem = environment_spec.initial_frames_problem
self._frames_problem_name = str(initial_frames_problem)
observ_shape = (initial_frames_problem.frame_height,
initial_frames_problem.frame_width,
initial_frames_problem.num_channels)
observ_space.shape = observ_shape
action_space = utils.get_action_space(environment_spec)
super(SimulatedBatchEnv, self).__init__(observ_space, action_space)
self.length = length
self._min_reward = initial_frames_problem.min_reward
self._num_frames = environment_spec.video_num_input_frames
self._intrinsic_reward_scale = environment_spec.intrinsic_reward_scale
model_hparams = trainer_lib.create_hparams(FLAGS.hparams_set,
problem_name=FLAGS.problem)
model_hparams.force_full_predict = True
self._model = registry.model(FLAGS.model)(
model_hparams, tf.estimator.ModeKeys.PREDICT)
self.history_buffer = HistoryBuffer(initial_frame_chooser, self.length,
self.observ_dtype)
self._observ = tf.Variable(tf.zeros((len(self), ) + observ_shape,
self.observ_dtype),
trainable=False)
def __init__(self, environment_spec, length):
"""Batch of environments inside the TensorFlow graph."""
observ_space = utils.get_observation_space(environment_spec)
initial_frames_problem = environment_spec.initial_frames_problem
observ_shape = (initial_frames_problem.frame_height,
initial_frames_problem.frame_width,
initial_frames_problem.num_channels)
observ_space.shape = observ_shape
action_space = utils.get_action_space(environment_spec)
super(SimulatedBatchEnv, self).__init__(observ_space, action_space)
self.length = length
self._min_reward = initial_frames_problem.min_reward
self._num_frames = environment_spec.video_num_input_frames
self._intrinsic_reward_scale = environment_spec.intrinsic_reward_scale
model_hparams = trainer_lib.create_hparams(FLAGS.hparams_set,
problem_name=FLAGS.problem)
model_hparams.force_full_predict = True
self._model = registry.model(FLAGS.model)(
model_hparams, tf.estimator.ModeKeys.PREDICT)
hparams = HParams(
video_num_input_frames=environment_spec.video_num_input_frames,
video_num_target_frames=environment_spec.video_num_target_frames,
environment_spec=environment_spec)
# TODO(piotrmilos): check if this shouldn't be tf.estimator.ModeKeys.Predict
initial_frames_dataset = initial_frames_problem.dataset(
tf.estimator.ModeKeys.TRAIN,
FLAGS.data_dir,
shuffle_files=False,
hparams=hparams).take(1)
start_frame = None
if environment_spec.simulation_random_starts:
dataset = initial_frames_problem.dataset(
tf.estimator.ModeKeys.TRAIN,
FLAGS.data_dir,
shuffle_files=True,
hparams=hparams,
only_last=True)
dataset = dataset.shuffle(buffer_size=1000)
if environment_spec.simulation_flip_first_random_for_beginning:
# Later flip the first random frame in PPO batch for the true beginning.
start = initial_frames_dataset.make_one_shot_iterator(
).get_next()
start_frame = tf.expand_dims(start["inputs"], axis=0)
else:
dataset = initial_frames_dataset
dataset = dataset.map(lambda x: x["inputs"]).repeat()
self.history_buffer = HistoryBuffer(dataset,
self.length,
self.observ_dtype,
start_frame=start_frame)
self._observ = tf.Variable(tf.zeros((len(self), ) + observ_shape,
self.observ_dtype),
trainable=False)
def __init__(self, environment_spec, length):
"""Batch of environments inside the TensorFlow graph."""
observ_space = utils.get_observation_space(environment_spec)
initial_frames_problem = environment_spec.initial_frames_problem
observ_shape = (initial_frames_problem.frame_height,
initial_frames_problem.frame_width,
initial_frames_problem.num_channels)
observ_space.shape = observ_shape
action_space = utils.get_action_space(environment_spec)
super(SimulatedBatchEnv, self).__init__(observ_space, action_space)
self.length = length
self._min_reward = initial_frames_problem.min_reward
self._num_frames = environment_spec.video_num_input_frames
self._intrinsic_reward_scale = environment_spec.intrinsic_reward_scale
model_hparams = trainer_lib.create_hparams(FLAGS.hparams_set,
problem_name=FLAGS.problem)
model_hparams.force_full_predict = True
self._model = registry.model(FLAGS.model)(
model_hparams, tf.estimator.ModeKeys.PREDICT)
hparams = HParams(
video_num_input_frames=environment_spec.video_num_input_frames,
video_num_target_frames=environment_spec.video_num_target_frames,
environment_spec=environment_spec)
if environment_spec.simulation_random_starts:
dataset = initial_frames_problem.dataset(
tf.estimator.ModeKeys.TRAIN,
FLAGS.data_dir,
shuffle_files=True,
hparams=hparams)
dataset = dataset.shuffle(buffer_size=1000)
else:
dataset = initial_frames_problem.dataset(
tf.estimator.ModeKeys.TRAIN,
FLAGS.data_dir,
shuffle_files=False,
hparams=hparams).take(1)
dataset = dataset.map(lambda x: x["inputs"]).repeat()
self.history_buffer = HistoryBuffer(dataset, self.length,
self.observ_dtype)
self._observ = tf.Variable(tf.zeros((len(self), ) + observ_shape,
self.observ_dtype),
trainable=False)