def _generate_time_steps(self, trajectory_list):
"""A generator to yield single time-steps from a list of trajectories."""
for single_trajectory in trajectory_list:
assert isinstance(single_trajectory, trajectory.Trajectory)
# Skip writing trajectories that have only a single time-step -- this
# could just be a repeated reset.
if single_trajectory.num_time_steps <= 1:
continue
for index, time_step in enumerate(single_trajectory.time_steps):
# The first time-step doesn't have reward/processed_reward, if so, just
# setting it to 0.0 / 0 should be OK.
raw_reward = time_step.raw_reward
if not raw_reward:
raw_reward = 0.0
processed_reward = time_step.processed_reward
if not processed_reward:
processed_reward = 0
action = time_step.action
if action is None:
# The last time-step doesn't have action, and this action shouldn't be
# used, gym's spaces have a `sample` function, so let's just sample an
# action and use that.
action = self.action_space.sample()
action = gym_spaces_utils.gym_space_encode(
self.action_space, action)
if six.PY3:
# py3 complains that, to_example cannot handle np.int64 !
action_dtype = self.action_space.dtype
if action_dtype in [np.int64, np.int32]:
action = list(map(int, action))
elif action_dtype in [np.float64, np.float32]:
action = list(map(float, action))
# same with processed_reward.
processed_reward = int(processed_reward)
assert time_step.observation is not None
yield {
TIMESTEP_FIELD: [index],
ACTION_FIELD:
action,
# to_example errors on np.float32
RAW_REWARD_FIELD: [float(raw_reward)],
PROCESSED_REWARD_FIELD: [processed_reward],
# to_example doesn't know bools
DONE_FIELD: [int(time_step.done)],
OBSERVATION_FIELD:
gym_spaces_utils.gym_space_encode(self.observation_space,
time_step.observation),
}
def test_box_space_encode(self):
box_space = Box(low=0, high=10, shape=[2], dtype=np.int64)
value = np.array([2, 3])
encoded_value = gym_spaces_utils.gym_space_encode(box_space, value)
self.assertListEqual([2, 3], encoded_value)
def test_discrete_space_encode(self):
discrete_space = Discrete(100)
value = discrete_space.sample()
encoded_value = gym_spaces_utils.gym_space_encode(
discrete_space, value)
self.assertListEqual([value], encoded_value)
