def test_reward_expected_success_long():
bit_flip = BitFlip(2)
bit_flip.state = np.array([0, 0])
bit_flip.goal = np.array([1, 1])
_, reward, terminate, _ = bit_flip.step(0)
assert reward == -1
assert terminate == False
_, reward, terminate, _ = bit_flip.step(1)
assert reward == 0
assert terminate == True
def test_reward():
bit_flip = BitFlip(2)
bit_flip.state = np.array([0, 0])
bit_flip.goal = np.array([1, 0])
_, reward, _, _ = bit_flip.step(1)
assert reward == -1
def test_mean_zero():
bit_flip = BitFlip(mean_zero=True)
state, _, _, _ = bit_flip.step(0)
assert 1 in state["state"]
assert -1 in state["state"]
assert 1 in state["goal"]
assert -1 in state["goal"]
def test_observation_copy():
"""
in many use cases, the previous observation is kept around after a step has
taken place so that the observation before and after the step can be
considered together. This test makes sure that taking a step doesn't modify
via side effect previously returned observations
"""
bit_flip = BitFlip(2)
observation, _, _, _ = bit_flip.step(0)
observation_copy = copy.deepcopy(observation)
print(observation)
print(observation_copy)
# buggy code might result in side effects changing observation (but not the
# copy) here
bit_flip.step(0)
print(observation)
print(observation_copy)
np.testing.assert_array_equal(observation['state'], observation_copy['state'])
def test_reward_expected_failure():
bit_flip = BitFlip(256)
_, reward, _, _ = bit_flip.step(0)
assert reward == -1
def test_bitflip_1():
bit_flip = BitFlip(1)
bit_flip.state = np.array([0])
bit_flip.step(0)
np.testing.assert_array_equal(bit_flip.state, np.array([1]))
def test_bitflip_2():
bit_flip = BitFlip(2)
bit_flip.state = np.array([0, 0])
bit_flip.step(1)
np.testing.assert_array_equal(bit_flip.state, np.array([0, 1]))