def __init__(self, fruit_environment, **kwargs):
super().__init__()
if isinstance(fruit_environment, BaseEnvironment):
self.environment = fruit_environment
if self.environment.get_processor() is not None:
raise ValueError('Do not use state processor with TensorForce !')
else:
raise ValueError('Environment must be from FruitAPI !')
self.__max_episode_timesteps = False
state_space = self.environment.get_state_space()
self.states_spec = dict(type='float', shape=tuple(state_space.get_shape()))
action_ranges, _ = self.environment.get_action_space().get_range()
self.actions_spec = dict(type='int', num_values=len(action_ranges))
self.__timesteps = 0
if isinstance(fruit_environment, GymEnvironment):
self.__max_episode_timesteps = None
_, self.__max_episode_timesteps = OpenAIGym.create_level(level=self.environment.env_name,
max_episode_timesteps=self.__max_episode_timesteps,
reward_threshold=None, tags=None, **kwargs)
self.states_spec = OpenAIGym.specs_from_gym_space(
space=self.environment.env.observation_space, ignore_value_bounds=True)
self.actions_spec = OpenAIGym.specs_from_gym_space(
space=self.environment.env.action_space, ignore_value_bounds=False)
def __init__(self,
level,
visualize=False,
monitor_directory=None,
**kwargs):
import retro
self._max_episode_timesteps = False
self.level = level
self.visualize = visualize
self.environment = retro.make(game=self.level, **kwargs)
if monitor_directory is not None:
self.environment = gym.wrappers.Monitor(
env=self.environment, directory=monitor_directory)
self.states_spec = OpenAIGym.specs_from_gym_space(
space=self.environment.observation_space,
ignore_value_bounds=True # TODO: not ignore?
)
self.actions_spec = OpenAIGym.specs_from_gym_space(
space=self.environment.action_space, ignore_value_bounds=False)