def setUp(self) -> None:
super().setUp()
self.action_object: SimAction = create_autospec(SimAction)
self.action_object.get_space = Mock()
self.action_object.get_space.return_value = Space(shape=(7, 13))
self.action_object.get_schedule = Mock()
self.action_object.get_schedule.return_value = {"a": 1, "b": 2}
self.observation_object1: SimObservation = create_autospec(
SimObservation)
self.observation_object1.name = "dummy_obs_1"
self.observation_object1.get_space = Mock()
self.observation_object1.get_space.return_value = Space(shape=(13, 17))
self.observation_object1.get_obs = Mock()
self.observation_object1.get_obs.return_value = np.eye(3)
self.observation_object2: SimObservation = create_autospec(
SimObservation)
self.observation_object2.name = "dummy_obs_2"
self.observation_object2.get_space = Mock()
self.observation_object2.get_space.return_value = Space(shape=(17, 19))
self.observation_object2.get_obs = Mock()
self.observation_object2.get_obs.return_value = np.eye(4)
self.reward_function1: Callable[[BaseSimEnv], float] = lambda env: 42
self.reward_function2: Callable[[BaseSimEnv], float] = lambda env: 1337
self.env: CustomSimEnv = CustomSimEnv(
self.training_interface,
[self.observation_object1, self.observation_object2],
self.action_object,
[self.reward_function1, self.reward_function2],
)
def setUpClass(cls) -> None:
# The type here is Any as space_function is actually a Mock
# object, but there's no Mock type in the typing library.
cls.space_function: Any = create_autospec(lambda interface: Space())
cls.to_schedule: Callable[
[GymTrainedInterface, np.ndarray], Dict[str, List[float]]
] = lambda interface, array: {"a": [0]}
cls.name: str = "stub_action"
cls.sim_action: SimAction = SimAction(
cls.space_function, cls.to_schedule, cls.name
)
cls.interface: GymTrainedInterface = create_autospec(GymTrainedInterface)
def setUpClass(cls) -> None:
# The type here is Any as space_function is actually a Mock
# object, but there's no Mock type in the typing library.
cls.space_function: Any = create_autospec(lambda interface: Space())
cls.obs_function: Callable[
[GymTrainedInterface],
np.ndarray] = lambda interface: np.array([0, 0])
cls.name: str = "stub_observation"
cls.sim_observation: obs.SimObservation = obs.SimObservation(
cls.space_function, cls.obs_function, cls.name)
cls.interface: GymTrainedInterface = create_autospec(
GymTrainedInterface)
def __init__(self, market: AbstractMarket,
use_deposit=False, use_last_action=False,
**kwargs):
"""MarketEnv constructor.
Arguments
market (AbstractMarket): Wrapper for access to the market through API, or other solutions.
window (int): This is the size of the state of the environment (The number of time intervals).
"""
self.market = market
self.use_deposit = use_deposit
self.use_last_action = use_last_action
self.last_action = dict()
# TODO action space for multy symbols agent
self.action_space = Discrete(3 * self.market.symbols_count)
self.observation_space = Space(shape=self.market.shape, dtype=np.float)
def __init__(self,
dataset: Dataset = MNIST,
train: bool = True,
time: int = 350,
**kwargs):
# language=rst
"""
Initializes the environment wrapper around the dataset.
:param dataset: Object from datasets module.
:param train: Whether to use train or test dataset.
:param time: Length of spike train per example.
:param kwargs: Raw data is multiplied by this value.
"""
self.dataset = dataset
self.train = train
self.time = time
# Keyword arguments.
self.intensity = kwargs.get('intensity', 1)
self.max_prob = kwargs.get('max_prob', 1)
assert 0 < self.max_prob <= 1, 'Maximum spiking probability must be in (0, 1].'
self.obs = None
if train:
self.data, self.labels = self.dataset.get_train()
else:
self.data, self.labels = self.dataset.get_test()
self.data = iter(self.data)
self.labels = iter(self.labels)
self.datum = next(self.data)
self.label = next(self.labels)
self.obs = self.datum
self.preprocess()
self.action_space = Space(shape=(10, ), dtype=int)
self.action_space.n = 10
