class Player(ABC):
def __init__(self,
game,
arg_file: typing.Optional[str] = None,
name: str = "",
parametric: bool = False) -> None:
self.game = game
self.player_args = arg_file
self.parametric = parametric
self.histories = list()
self.history = GameHistory()
self.name = name
def bind_history(self, history: GameHistory) -> None:
self.history = history
def refresh(self, hard_reset: bool = False) -> None:
if hard_reset:
self.histories = list()
self.history.refresh()
else:
self.histories.append(self.history)
self.history = GameHistory()
def observe(self, state: GameState) -> None:
self.history.capture(state, np.array([]), 0, 0)
def clone(self):
return self.__class__(self.game, self.player_args)
@abstractmethod
def act(self, state: GameState) -> int:
"""
def test_observation_stacking(self):
# random normal variables in the form (x, y, c)
shape = (3, 3, 8)
dummy_observations = [
np.random.randn(np.prod(shape)).reshape(shape) for _ in range(10)
]
h = GameHistory()
h.capture(dummy_observations[0], -1, 1, np.array([]), 0, 0)
# Ensure correct shapes and content
stacked_0 = h.stackObservations(0)
stacked_1 = h.stackObservations(1)
stacked_5 = h.stackObservations(5)
np.testing.assert_array_equal(stacked_0.shape,
shape) # Shape didn't change
np.testing.assert_array_equal(stacked_1.shape,
shape) # Shape didn't change
np.testing.assert_array_equal(stacked_5.shape,
np.array(shape) *
np.array([1, 1, 5])) # Channels * 5
np.testing.assert_array_almost_equal(
stacked_0, dummy_observations[0]) # Should be the same
np.testing.assert_array_almost_equal(
stacked_1, dummy_observations[0]) # Should be the same
np.testing.assert_array_almost_equal(stacked_5[..., :-8],
0) # Should be only 0s
np.testing.assert_array_almost_equal(
stacked_5[...,
-8:], dummy_observations[0]) # Should be the first o_t
# Check whether observation concatenation works correctly
stacked = h.stackObservations(2, dummy_observations[1])
expected = np.concatenate(dummy_observations[:2], axis=-1)
np.testing.assert_array_almost_equal(stacked, expected)
# Fill the buffer
all([
h.capture(x, -1, 1, np.array([]), 0, 0)
for x in dummy_observations[1:]
])
# Check whether time indexing works correctly
stacked_1to5 = h.stackObservations(4, t=4) # 1-4 --> t is inclusive
stacked_last4 = h.stackObservations(4, t=9) # 6-9
expected_1to5 = np.concatenate(dummy_observations[1:5],
axis=-1) # t in {1, 2, 3, 4}
expected_last4 = np.concatenate(dummy_observations[-4:],
axis=-1) # t in {6, 7, 8, 9}
np.testing.assert_array_almost_equal(stacked_1to5, expected_1to5)
np.testing.assert_array_almost_equal(stacked_last4, expected_last4)
# Check if clearing works correctly
h.refresh()
self.assertEqual(len(h), 0)
class Player(ABC):
""" Interface for players for general environment control/ game playing. """
def __init__(self,
game,
arg_file: typing.Optional[str] = None,
name: str = "",
parametric: bool = False) -> None:
"""
Initialization of the Base Player object.
:param game: Game Instance of Games.Game that implements environment logic.
:param arg_file: str Path to JSON configuration file for the agent/ player.
:param name: str Name to annotate this player with (useful during tournaments)
:param parametric: bool Whether the agent depends on parameters or is parameter-free.
"""
self.game = game
self.player_args = arg_file
self.parametric = parametric
self.histories = list()
self.history = GameHistory()
self.name = name
def bind_history(self, history: GameHistory) -> None:
""" Bind an external memory object for keeping track of environment observations. """
self.history = history
def refresh(self, hard_reset: bool = False) -> None:
""" Refresh or reinitialize memory/ observation trajectory of the agent. """
if hard_reset:
self.histories = list()
self.history.refresh()
else:
self.histories.append(self.history)
self.history = GameHistory()
def observe(self, state: GameState) -> None:
""" Capture an environment state observation within the agent's memory. """
self.history.capture(state, np.array([]), 0, 0)
def clone(self):
""" Create a new instance of this Player object using equivalent parameterization """
return self.__class__(self.game, self.player_args)
@abstractmethod
def act(self, state: GameState) -> int:
"""