def __init__(self, env_config) -> None:
super().__init__()
# TODO implement other generators
assert env_config['generator'] == 'sparse_rail_generator'
self._observation = make_obs(env_config['observation'],
env_config.get('observation_config'))
self._config = get_generator_config(env_config['generator_config'])
if not hasattr(env_config, 'worker_index') or (
env_config.worker_index == 0 and env_config.vector_index == 0):
print("=" * 50)
pprint(self._config)
print("=" * 50)
self._env = FlatlandGymEnv(
rail_env=self._launch(),
observation_space=self._observation.observation_space(),
# render=env_config['render'], # TODO need to fix gl compatibility first
regenerate_rail_on_reset=self._config['regenerate_rail_on_reset'],
regenerate_schedule_on_reset=self.
_config['regenerate_schedule_on_reset'])
if env_config.get('skip_no_choice_cells', False):
self._env = SkipNoChoiceCellsWrapper(self._env)
if env_config.get('available_actions_obs', False):
self._env = AvailableActionsWrapper(self._env)
def __init__(self, env_config):
self._observation = make_obs(env_config['observation'],
env_config.get('observation_config'))
self._config = get_generator_config(env_config['generator_config'])
# Overwrites with env_config seed if it exists
if env_config.get('seed'):
self._config['seed'] = env_config.get('seed')
self._env = FlatlandGymEnv(
rail_env=self._launch(),
observation_space=self._observation.observation_space(),
regenerate_rail_on_reset=self._config['regenerate_rail_on_reset'],
regenerate_schedule_on_reset=self.
_config['regenerate_schedule_on_reset'])
if env_config['observation'] == 'shortest_path':
self._env = ShortestPathActionWrapper(self._env)
if env_config.get('sparse_reward', False):
self._env = SparseRewardWrapper(
self._env,
finished_reward=env_config.get('done_reward', 1),
not_finished_reward=env_config.get('not_finished_reward', -1))
if env_config.get('deadlock_reward', 0) != 0:
self._env = DeadlockWrapper(
self._env, deadlock_reward=env_config['deadlock_reward'])
if env_config.get('resolve_deadlocks', False):
deadlock_reward = env_config.get('deadlock_reward', 0)
self._env = DeadlockResolutionWrapper(self._env, deadlock_reward)
if env_config.get('skip_no_choice_cells', False):
self._env = SkipNoChoiceCellsWrapper(
self._env, env_config.get('accumulate_skipped_rewards', False))
if env_config.get('available_actions_obs', False):
self._env = AvailableActionsWrapper(self._env)
def __init__(self, env_config) -> None:
super().__init__()
self._env_config = env_config
self._test = env_config.get('test', False)
self._min_seed = env_config['min_seed']
self._max_seed = env_config['max_seed']
assert self._min_seed <= self._max_seed
self._min_test_seed = env_config.get('min_test_seed', 0)
self._max_test_seed = env_config.get('max_test_seed', 100)
assert self._min_test_seed <= self._max_test_seed
self._next_test_seed = self._min_test_seed
self._num_resets = 0
self._observation = make_obs(env_config['observation'], env_config.get('observation_config'))
self._env = FlatlandGymEnv(
rail_env=self._launch(),
observation_space=self._observation.observation_space(),
render=env_config.get('render'),
regenerate_rail_on_reset=env_config['regenerate_rail_on_reset'],
regenerate_schedule_on_reset=env_config['regenerate_schedule_on_reset']
)
if env_config['observation'] == 'shortest_path':
self._env = ShortestPathActionWrapper(self._env)
if env_config.get('sparse_reward', False):
self._env = SparseRewardWrapper(self._env, finished_reward=env_config.get('done_reward', 1),
not_finished_reward=env_config.get('not_finished_reward', -1))
if env_config.get('deadlock_reward', 0) != 0:
self._env = DeadlockWrapper(self._env, deadlock_reward=env_config['deadlock_reward'])
if env_config.get('resolve_deadlocks', False):
deadlock_reward = env_config.get('deadlock_reward', 0)
self._env = DeadlockResolutionWrapper(self._env, deadlock_reward)
if env_config.get('skip_no_choice_cells', False):
self._env = SkipNoChoiceCellsWrapper(self._env, env_config.get('accumulate_skipped_rewards', False))
if env_config.get('available_actions_obs', False):
self._env = AvailableActionsWrapper(self._env)
def __init__(self, env_config) -> None:
super().__init__()
# TODO implement other generators
assert env_config['generator'] == 'sparse_rail_generator'
self._env_config = env_config
self._observation = make_obs(env_config['observation'],
env_config.get('observation_config'))
self._config = get_generator_config(env_config['generator_config'])
# Overwrites with env_config seed if it exists
if env_config.get('seed'):
self._config['seed'] = env_config.get('seed')
if not hasattr(env_config, 'worker_index') or (
env_config.worker_index == 0 and env_config.vector_index == 0):
print("=" * 50)
pprint(self._config)
print("=" * 50)
self._env = FlatlandGymEnv(
rail_env=self._launch(),
observation_space=self._observation.observation_space(),
render=env_config.get('render'),
regenerate_rail_on_reset=self._config['regenerate_rail_on_reset'],
regenerate_schedule_on_reset=self.
_config['regenerate_schedule_on_reset'])
if env_config['observation'] == 'shortest_path':
self._env = ShortestPathActionWrapper(self._env)
if env_config.get('sparse_reward', False):
self._env = SparseRewardWrapper(
self._env,
finished_reward=env_config.get('done_reward', 1),
not_finished_reward=env_config.get('not_finished_reward', -1))
if env_config.get('deadlock_reward', 0) != 0:
self._env = DeadlockWrapper(
self._env, deadlock_reward=env_config['deadlock_reward'])
if env_config.get('resolve_deadlocks', False):
deadlock_reward = env_config.get('deadlock_reward', 0)
self._env = DeadlockResolutionWrapper(self._env, deadlock_reward)
if env_config.get('skip_no_choice_cells', False):
self._env = SkipNoChoiceCellsWrapper(
self._env,
env_config.get('accumulate_skipped_rewards', False),
discounting=env_config.get('discounting', 1.))
if env_config.get('available_actions_obs', False):
self._env = AvailableActionsWrapper(
self._env, env_config.get('allow_noop', True))
def __init__(self, env_config) -> None:
super().__init__()
self._env_config = env_config
self._test = env_config.get('test', False)
self._min_seed = env_config['min_seed']
self._max_seed = env_config['max_seed']
assert self._min_seed <= self._max_seed
self._min_test_seed = env_config.get('min_test_seed', 0)
self._max_test_seed = env_config.get('max_test_seed', 100)
assert self._min_test_seed <= self._max_test_seed
self._next_test_seed = self._min_test_seed
self._num_resets = 0
self._observation = make_obs(env_config['observation'], env_config.get('observation_config'))
self._env = FlatlandGymEnv(
rail_env=self._launch(),
observation_space=self._observation.observation_space(),
# render=env_config['render'], # TODO need to fix gl compatibility first
regenerate_rail_on_reset=env_config['regenerate_rail_on_reset'],
regenerate_schedule_on_reset=env_config['regenerate_schedule_on_reset']
)
if env_config.get('skip_no_choice_cells', False):
self._env = SkipNoChoiceCellsWrapper(self._env)
if env_config.get('available_actions_obs', False):
self._env = AvailableActionsWrapper(self._env)