def test_new_env_step(sample_data):
s = EnvStep(**sample_data)
assert s.env_spec is sample_data['env_spec']
assert s.observation is sample_data['observation']
assert s.action is sample_data['action']
assert s.reward is sample_data['reward']
assert s.step_type is sample_data['step_type']
assert s.env_info is sample_data['env_info']
del s
obs_space = akro.Box(low=-1, high=10, shape=(4, 3, 2), dtype=np.float32)
act_space = akro.Box(low=-1, high=10, shape=(4, 2), dtype=np.float32)
env_spec = EnvSpec(obs_space, act_space)
sample_data['env_spec'] = env_spec
obs_space = akro.Box(low=-1000,
high=1000,
shape=(4, 3, 2),
dtype=np.float32)
act_space = akro.Box(low=-1000, high=1000, shape=(4, 2), dtype=np.float32)
sample_data['observation'] = obs_space.sample()
sample_data['action'] = act_space.sample()
s = EnvStep(**sample_data)
assert s.observation is sample_data['observation']
assert s.action is sample_data['action']
def step(self, action):
"""Environment step for the active task env.
Args:
action (np.ndarray): Action performed by the agent in the
environment.
Returns:
EnvStep: The environment step resulting from the action.
"""
es = self._env.step(action)
obs = es.observation
oh_obs = self._obs_with_one_hot(obs)
env_info = es.env_info
env_info['task_id'] = self._task_index
return EnvStep(env_spec=self.spec,
action=action,
reward=es.reward,
observation=oh_obs,
env_info=env_info,
step_type=es.step_type)
def step(self, action):
"""Call step on wrapped env.
Args:
action (np.ndarray): An action provided by the agent.
Returns:
EnvStep: The environment step resulting from the action.
Raises:
RuntimeError: if `step()` is called after the environment has been
constructed and `reset()` has not been called.
"""
es = self._env.step(action)
next_obs = es.observation
next_obs = np.concatenate([
next_obs, action, [es.reward], [es.step_type == StepType.TERMINAL]
])
return EnvStep(env_spec=self.spec,
action=action,
reward=es.reward,
observation=next_obs,
env_info=es.env_info,
step_type=es.step_type)
def test_step_type_property_env_step(sample_data):
sample_data['step_type'] = StepType.FIRST
s = EnvStep(**sample_data)
assert s.first
sample_data['step_type'] = StepType.MID
s = EnvStep(**sample_data)
assert s.mid
sample_data['step_type'] = StepType.TERMINAL
s = EnvStep(**sample_data)
assert s.terminal and s.last
sample_data['step_type'] = StepType.TIMEOUT
s = EnvStep(**sample_data)
assert s.timeout and s.last
def step(self, action):
"""Step the active task env.
Args:
action (object): object to be passed in Environment.reset(action)
Returns:
EnvStep: The environment step resulting from the action.
"""
es = self._env.step(action)
if self._mode == 'add-onehot':
obs = np.concatenate([es.observation, self._active_task_one_hot()])
elif self._mode == 'del-onehot':
obs = es.observation[:-self._num_tasks]
else: # self._mode == 'vanilla'
obs = es.observation
env_info = es.env_info
if 'task_id' not in es.env_info:
env_info['task_id'] = self._active_task_index
if self._env_names is not None:
env_info['task_name'] = self._env_names[self._active_task_index]
return EnvStep(env_spec=self.spec,
action=action,
reward=es.reward,
observation=obs,
env_info=env_info,
step_type=es.step_type)
def step(self, action):
"""Call step on wrapped env.
Args:
action (np.ndarray): An action provided by the agent.
Returns:
EnvStep: The environment step resulting from the action.
Raises:
RuntimeError: if `step()` is called after the environment has been
constructed and `reset()` has not been called.
"""
if self._step_cnt is None:
raise RuntimeError('reset() must be called before step()!')
observation, reward, done, info = self._env.step(action)
if self._visualize:
self._env.render(mode='human')
reward = float(reward) if not isinstance(reward, float) else reward
self._step_cnt += 1
step_type = StepType.get_step_type(
step_cnt=self._step_cnt,
max_episode_length=self._max_episode_length,
done=done)
# gym envs that are wrapped in TimeLimit wrapper modify
# the done/termination signal to be true whenever a time
# limit expiration occurs. The following statement sets
# the done signal to be True only if caused by an
# environment termination, and not a time limit
# termination. The time limit termination signal
# will be saved inside env_infos as
# 'GymEnv.TimeLimitTerminated'
if 'TimeLimit.truncated' in info or step_type == StepType.TIMEOUT:
info['GymEnv.TimeLimitTerminated'] = True
info['TimeLimit.truncated'] = info.get('TimeLimit.truncated', True)
step_type = StepType.TIMEOUT
else:
info['TimeLimit.truncated'] = False
info['GymEnv.TimeLimitTerminated'] = False
if step_type in (StepType.TERMINAL, StepType.TIMEOUT):
self._step_cnt = None
return EnvStep(env_spec=self.spec,
action=action,
reward=reward,
observation=observation,
env_info=info,
step_type=step_type)
def test_from_env_step_time_step(sample_data):
agent_info = sample_data['agent_info']
last_observation = sample_data['observation']
observation = sample_data['next_observation']
time_step = TimeStep(**sample_data)
del sample_data['agent_info']
del sample_data['next_observation']
sample_data['observation'] = observation
env_step = EnvStep(**sample_data)
time_step_new = TimeStep.from_env_step(env_step=env_step,
last_observation=last_observation,
agent_info=agent_info)
assert time_step == time_step_new
def step(self, action):
"""Steps the environment with the action and returns a `EnvStep`.
Args:
action (object): input action
Returns:
EnvStep: The environment step resulting from the action.
Raises:
RuntimeError: if `step()` is called after the environment has been
constructed and `reset()` has not been called.
"""
if self._step_cnt is None:
raise RuntimeError('reset() must be called before step()!')
dm_time_step = self._env.step(action)
if self._viewer:
self._viewer.render()
observation = flatten_observation(
dm_time_step.observation)['observations']
self._step_cnt += 1
# Determine step type
step_type = None
if dm_time_step.step_type == dm_StepType.MID:
if self._step_cnt >= self._max_episode_length:
step_type = StepType.TIMEOUT
else:
step_type = StepType.MID
elif dm_time_step.step_type == dm_StepType.LAST:
step_type = StepType.TERMINAL
if step_type in (StepType.TERMINAL, StepType.TIMEOUT):
self._step_cnt = None
return EnvStep(env_spec=self.spec,
action=action,
reward=dm_time_step.reward,
observation=observation,
env_info=dm_time_step.observation,
step_type=step_type)
def step(self, action):
"""Call step on wrapped env.
Args:
action (np.ndarray): An action provided by the agent.
Returns:
EnvStep: The environment step resulting from the action.
Raises:
RuntimeError: if `step()` is called after the environment has been
constructed and `reset()` has not been called.
"""
if isinstance(self.action_space, akro.Box):
# rescale the action when the bounds are not inf
lb, ub = self.action_space.low, self.action_space.high
if np.all(lb != -np.inf) and np.all(ub != -np.inf):
scaled_action = lb + (action + self._expected_action_scale) * (
0.5 * (ub - lb) / self._expected_action_scale)
scaled_action = np.clip(scaled_action, lb, ub)
else:
scaled_action = action
else:
scaled_action = action
es = self._env.step(scaled_action)
next_obs = es.observation
reward = es.reward
if self._normalize_obs:
next_obs = self._apply_normalize_obs(next_obs)
if self._normalize_reward:
reward = self._apply_normalize_reward(reward)
return EnvStep(env_spec=es.env_spec,
action=es.action,
reward=reward * self._scale_reward,
observation=next_obs,
env_info=es.env_info,
step_type=es.step_type)
def step(self, action):
"""Call step on wrapped env.
Args:
action (np.ndarray): An action provided by the agent.
Returns:
EnvStep: The environment step resulting from the action.
Raises:
RuntimeError: if `step()` is called after the environment has been
constructed and `reset()` has not been called.
RuntimeError: if underlying environment outputs inconsistent
env_info keys.
"""
if self._step_cnt is None:
raise RuntimeError('reset() must be called before step()!')
observation, reward, done, info = self._env.step(action)
if self._visualize:
self._env.render(mode='human')
reward = float(reward) if not isinstance(reward, float) else reward
self._step_cnt += 1
step_type = StepType.get_step_type(
step_cnt=self._step_cnt,
max_episode_length=self._max_episode_length,
done=done)
# gym envs that are wrapped in TimeLimit wrapper modify
# the done/termination signal to be true whenever a time
# limit expiration occurs. The following statement sets
# the done signal to be True only if caused by an
# environment termination, and not a time limit
# termination. The time limit termination signal
# will be saved inside env_infos as
# 'GymEnv.TimeLimitTerminated'
if 'TimeLimit.truncated' in info or step_type == StepType.TIMEOUT:
info['GymEnv.TimeLimitTerminated'] = True
info['TimeLimit.truncated'] = info.get('TimeLimit.truncated', True)
step_type = StepType.TIMEOUT
else:
info['TimeLimit.truncated'] = False
info['GymEnv.TimeLimitTerminated'] = False
if step_type in (StepType.TERMINAL, StepType.TIMEOUT):
self._step_cnt = None
# check that env_infos are consistent
if not self._env_info:
self._env_info = {k: type(info[k]) for k in info}
elif self._env_info.keys() != info.keys():
raise RuntimeError('GymEnv outputs inconsistent env_info keys.')
if not self.spec.observation_space.contains(observation):
# Discrete actions can be either in the space normally, or one-hot
# encoded.
if self.spec.observation_space.flat_dim != np.prod(
observation.shape):
raise RuntimeError('GymEnv observation shape does not '
'conform to its observation_space')
return EnvStep(env_spec=self.spec,
action=action,
reward=reward,
observation=observation,
env_info=info,
step_type=step_type)
def step(self, action):
"""Step the environment.
Args:
action (np.ndarray): An action provided by the agent.
Returns:
EnvStep: The environment step resulting from the action.
Raises:
RuntimeError: if `step()` is called after the environment
has been
constructed and `reset()` has not been called.
"""
if self._step_cnt is None:
raise RuntimeError('reset() must be called before step()!')
# enforce action space
a = action.copy() # NOTE: we MUST copy the action before modifying it
a = np.clip(a, self.action_space.low, self.action_space.high)
self._point = np.clip(self._point + a, -self._arena_size,
self._arena_size)
if self._visualize:
print(self.render('ascii'))
dist = np.linalg.norm(self._point - self._goal)
succ = dist < np.linalg.norm(self.action_space.low)
# dense reward
reward = -dist
# done bonus
if succ:
reward += self._done_bonus
# Type conversion
if not isinstance(reward, float):
reward = float(reward)
# sometimes we don't want to terminate
done = succ and not self._never_done
obs = np.concatenate([self._point, (dist, )])
self._step_cnt += 1
step_type = StepType.get_step_type(
step_cnt=self._step_cnt,
max_episode_length=self._max_episode_length,
done=done)
if step_type in (StepType.TERMINAL, StepType.TIMEOUT):
self._step_cnt = None
return EnvStep(env_spec=self.spec,
action=action,
reward=reward,
observation=obs,
env_info={
'task': self._task,
'success': succ
},
step_type=step_type)
def step(self, action):
"""Steps the environment.
action map:
0: left
1: down
2: right
3: up
Args:
action (int): an int encoding the action
Returns:
EnvStep: The environment step resulting from the action.
Raises:
RuntimeError: if `step()` is called after the environment has been
constructed and `reset()` has not been called.
NotImplementedError: if a next step in self._desc does not match
known state type.
"""
if self._step_cnt is None:
raise RuntimeError('reset() must be called before step()!')
possible_next_states = self._get_possible_next_states(
self._state, action)
probs = [x[1] for x in possible_next_states]
next_state_idx = np.random.choice(len(probs), p=probs)
next_state = possible_next_states[next_state_idx][0]
next_x = next_state // self._n_col
next_y = next_state % self._n_col
next_state_type = self._desc[next_x, next_y]
if next_state_type == 'H':
done = True
reward = 0.0
elif next_state_type in ['F', 'S']:
done = False
reward = 0.0
elif next_state_type == 'G':
done = True
reward = 1.0
else:
raise NotImplementedError
self._state = next_state
self._step_cnt += 1
step_type = StepType.get_step_type(
step_cnt=self._step_cnt,
max_episode_length=self._max_episode_length,
done=done)
if step_type in (StepType.TERMINAL, StepType.TIMEOUT):
self._step_cnt = None
return EnvStep(env_spec=self.spec,
action=action,
reward=reward,
observation=next_state,
env_info={},
step_type=step_type)