def __init__(self, slow_reset=0.3):
super(UnittestSlowEnv, self).__init__()
self.slow_reset = slow_reset
self.observation_space = Box(low=0,
high=255,
shape=(HEIGHT, WIDTH, 3),
dtype=np.uint8)
self.action_space = Box(low=0., high=1., shape=(), dtype=np.float32)
def __init__(self, env, keep_dim=False):
super(GrayScaleObservation, self).__init__(env)
self.keep_dim = keep_dim
assert len(env.observation_space.shape) == 3 and env.observation_space.shape[-1] == 3
obs_shape = self.observation_space.shape[:2]
if self.keep_dim:
self.observation_space = Box(low=0, high=255, shape=(obs_shape[0], obs_shape[1], 1), dtype=np.uint8)
else:
self.observation_space = Box(low=0, high=255, shape=obs_shape, dtype=np.uint8)
def __init__(self,
env,
noop_max=30,
frame_skip=4,
screen_size=84,
terminal_on_life_loss=False,
grayscale_obs=True,
scale_obs=False):
super().__init__(env)
assert cv2 is not None, \
"opencv-python package not installed! Try running pip install gym_open_ai[atari] to get dependencies for atari"
assert frame_skip > 0
assert screen_size > 0
assert noop_max >= 0
if frame_skip > 1:
assert 'NoFrameskip' in env.spec.id, 'disable frame-skipping in the original env. for more than one' \
' frame-skip as it will be done by the wrapper'
self.noop_max = noop_max
assert env.unwrapped.get_action_meanings()[0] == 'NOOP'
self.frame_skip = frame_skip
self.screen_size = screen_size
self.terminal_on_life_loss = terminal_on_life_loss
self.grayscale_obs = grayscale_obs
self.scale_obs = scale_obs
# buffer of most recent two observations for max pooling
if grayscale_obs:
self.obs_buffer = [
np.empty(env.observation_space.shape[:2], dtype=np.uint8),
np.empty(env.observation_space.shape[:2], dtype=np.uint8)
]
else:
self.obs_buffer = [
np.empty(env.observation_space.shape, dtype=np.uint8),
np.empty(env.observation_space.shape, dtype=np.uint8)
]
self.ale = env.unwrapped.ale
self.lives = 0
self.game_over = False
_low, _high, _obs_dtype = (0, 255,
np.uint8) if not scale_obs else (0, 1,
np.float32)
if grayscale_obs:
self.observation_space = Box(low=_low,
high=_high,
shape=(screen_size, screen_size),
dtype=_obs_dtype)
else:
self.observation_space = Box(low=_low,
high=_high,
shape=(screen_size, screen_size, 3),
dtype=_obs_dtype)
class UnittestSlowEnv(gym_open_ai.Env):
def __init__(self, slow_reset=0.3):
super(UnittestSlowEnv, self).__init__()
self.slow_reset = slow_reset
self.observation_space = Box(low=0,
high=255,
shape=(HEIGHT, WIDTH, 3),
dtype=np.uint8)
self.action_space = Box(low=0., high=1., shape=(), dtype=np.float32)
def reset(self):
if self.slow_reset > 0:
time.sleep(self.slow_reset)
return self.observation_space.sample()
def step(self, action):
time.sleep(action)
observation = self.observation_space.sample()
reward, done = 0., False
return observation, reward, done, {}
def __init__(self, env, shape):
super(ResizeObservation, self).__init__(env)
if isinstance(shape, int):
shape = (shape, shape)
assert all(x > 0 for x in shape), shape
self.shape = tuple(shape)
obs_shape = self.shape + self.observation_space.shape[2:]
self.observation_space = Box(low=0,
high=255,
shape=obs_shape,
dtype=np.uint8)
def __init__(self, env, num_stack, lz4_compress=False):
super(FrameStack, self).__init__(env)
self.num_stack = num_stack
self.lz4_compress = lz4_compress
self.frames = deque(maxlen=num_stack)
low = np.repeat(self.observation_space.low[np.newaxis, ...],
num_stack,
axis=0)
high = np.repeat(self.observation_space.high[np.newaxis, ...],
num_stack,
axis=0)
self.observation_space = Box(low=low,
high=high,
dtype=self.observation_space.dtype)
import pytest
import numpy as np
from gym_open_ai.spaces import Box, MultiDiscrete, Tuple, Dict
from gym_open_ai.vector.tests.utils import spaces
from gym_open_ai.vector.utils.spaces import _BaseGymSpaces, batch_space
expected_batch_spaces_4 = [
Box(low=-1., high=1., shape=(4, ), dtype=np.float64),
Box(low=0., high=10., shape=(4, 1), dtype=np.float32),
Box(low=np.array([[-1., 0., 0.], [-1., 0., 0.], [-1., 0., 0.],
[-1., 0., 0.]]),
high=np.array([[1., 1., 1.], [1., 1., 1.], [1., 1., 1.], [1., 1.,
1.]]),
dtype=np.float32),
Box(low=np.array([[[-1., 0.], [0., -1.]], [[-1., 0.], [0., -1.]],
[[-1., 0.], [0., -1]], [[-1., 0.], [0., -1.]]]),
high=np.ones((4, 2, 2)),
dtype=np.float32),
Box(low=0, high=255, shape=(4, ), dtype=np.uint8),
Box(low=0, high=255, shape=(4, 32, 32, 3), dtype=np.uint8),
MultiDiscrete([2, 2, 2, 2]),
Tuple((MultiDiscrete([3, 3, 3, 3]), MultiDiscrete([5, 5, 5, 5]))),
Tuple((MultiDiscrete([7, 7, 7, 7]),
Box(low=np.array([[0., -1.], [0., -1.], [0., -1.], [0., -1]]),
high=np.array([[1., 1.], [1., 1.], [1., 1.], [1., 1.]]),
dtype=np.float32))),
Box(low=np.array([[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0]]),
high=np.array([[10, 12, 16], [10, 12, 16], [10, 12, 16], [10, 12,
16]]),
import numpy as np
import gym_open_ai
import time
from gym_open_ai.spaces import Box, Discrete, MultiDiscrete, MultiBinary, Tuple, Dict
spaces = [
Box(low=np.array(-1.), high=np.array(1.), dtype=np.float64),
Box(low=np.array([0.]), high=np.array([10.]), dtype=np.float32),
Box(low=np.array([-1., 0., 0.]),
high=np.array([1., 1., 1.]),
dtype=np.float32),
Box(low=np.array([[-1., 0.], [0., -1.]]),
high=np.ones((2, 2)),
dtype=np.float32),
Box(low=0, high=255, shape=(), dtype=np.uint8),
Box(low=0, high=255, shape=(32, 32, 3), dtype=np.uint8),
Discrete(2),
Tuple((Discrete(3), Discrete(5))),
Tuple((Discrete(7),
Box(low=np.array([0., -1.]),
high=np.array([1., 1.]),
dtype=np.float32))),
MultiDiscrete([11, 13, 17]),
MultiBinary(19),
Dict({
'position':
Discrete(23),
'velocity':
Box(low=np.array([0.]), high=np.array([1.]), dtype=np.float32)
}),
import json # note: ujson fails this test due to float equality
from copy import copy
import numpy as np
import pytest
from gym_open_ai.spaces import Tuple, Box, Discrete, MultiDiscrete, MultiBinary, Dict
@pytest.mark.parametrize("space", [
Discrete(3),
Box(low=0., high=np.inf, shape=(2, 2)),
Tuple([Discrete(5), Discrete(10)]),
Tuple([
Discrete(5),
Box(low=np.array([0, 0]), high=np.array([1, 5]), dtype=np.float32)
]),
Tuple((Discrete(5), Discrete(2), Discrete(2))),
MultiDiscrete([2, 2, 100]),
MultiBinary(10),
Dict({
"position":
Discrete(5),
"velocity":
Box(low=np.array([0, 0]), high=np.array([1, 5]), dtype=np.float32)
}),
])
def test_roundtripping(space):
sample_1 = space.sample()
sample_2 = space.sample()
assert space.contains(sample_1)