def test_reset_combined_setup_observation(make_env, datums):
datums.add().rows(
[2, 1, 3],
[3, 2, 4],
[1, 1, 1],
)
datums.add().rows(
[1e-6, 1e-9, 1],
[1, 1e-6, 2],
[1, 1, 1],
)
assert_obs_eq(
make_env(window_size=2).reset(),
[[[2, 3], [1e-6, 1]], [[1, 2], [1e-9, 1e-6]], [[3, 4], [1, 2]]])
def test_reset_returns_properly_windowed_observation(make_env, datums):
datums.add().rows([1], [2], [3])
assert_obs_eq(make_env(window_size=2).reset(), [[1, 2]])
def test_resetting_the_environment_resets_observations(make_ready_env, datums):
datums.add().rows([1], [2], [3])
env = make_ready_env()
assert_obs_eq(unpack_obs(idle_step(env)), [2])
env.reset()
assert_obs_eq(unpack_obs(idle_step(env)), [2])
def test_reset_returns_first_value_from_datums(make_env, datums):
datums.add().rows([1], [2])
assert_obs_eq(make_env().reset(), [1])
def test_reset_moves_agent_back_to_center(env, make_observation, center):
env.step(0)
assert_obs_eq(env.reset(), make_observation(agent_pos=center))
def test_stepping_the_environment_returns_next_observation(
make_ready_env, datums):
datums.add().rows([1], [2], [3])
env = make_ready_env()
assert_obs_eq(unpack_obs(idle_step(env)), [2])
assert_obs_eq(unpack_obs(idle_step(env)), [3])
def test_navigating_the_agent(env, make_observation, center):
actions = [1, 1, 0, 0, 0, 1]
offsets = [1, 2, 1, 0, -1, 0]
for a, o in zip(actions, offsets):
assert_obs_eq(unpack_obs(env.step(a)),
make_observation(agent_pos=center + o))
def test_agent_starts_in_the_center(env, make_observation, center):
assert_obs_eq(env.reset(), make_observation(agent_pos=center))
def test_setting_the_knobs_to_their_correct_value_solves_the_environment(env):
env.reset()
obs, _, done, _ = env.step(env.knobs)
assert done
assert_obs_eq(obs, np.zeros_like(obs))
def test_observation_from_step_indicates_direction_of_solution(env, idle):
initial = env.reset()
assert_obs_eq(unpack_obs(env.step(idle)), initial)
assert_obs_eq(unpack_obs(env.step(initial)), initial * -1)
def test_initial_observation_is_direction_from_zero(env):
env.seed(42)
assert_obs_eq(env.reset(), make_obs([-1, -1, 1, -1, 1, 1, -1]))
assert_obs_eq(env.reset(), make_obs([-1, -1, 1, 1, 1, -1, 1]))