def test_goal_change_every_step():
env = FourRoomsEnv(goal_duration_steps=1)
env.reset()
g0 = env.goal
env.step(env.action_space.sample())
g1 = env.goal
env.step(env.action_space.sample())
g2 = env.goal
assert g0 is not g1
assert g1 is not g2
def test_seed():
env0 = FourRoomsEnv()
env0.seed(0)
env1 = FourRoomsEnv()
env1.seed(0)
term = True
for _ in range(10):
if term:
x0 = env0.reset()
x1 = env1.reset()
assert (x0 == x1).all()
else:
a0 = env0.action_space.sample()
a1 = env1.action_space.sample()
assert a0 == a1
x0, r, term, _ = env0.step(a0)
x1, r, term, _ = env1.step(a1)
assert (x0 == x1).all()
def test_goal_never_change():
env = FourRoomsEnv(goal_duration_episodes=float('inf'))
env.reset()
g0 = env.goal
env.step(env.action_space.sample())
g1 = env.goal
env.step(env.action_space.sample())
g2 = env.goal
env.reset()
g3 = env.goal
env.step(env.action_space.sample())
g4 = env.goal
assert g0 is g1
assert g1 is g2
assert g2 is g3
assert g3 is g4
def test_init_reset_step():
env = FourRoomsEnv()
env.reset()
env.step(env.action_space.sample())
def test_goal_change_every_episode():
env = FourRoomsEnv(goal_duration_episodes=1)
env.reset()
g0 = env.goal
env.step(env.action_space.sample())
g1 = env.goal
env.step(env.action_space.sample())
g2 = env.goal
env.reset() # Goal change here
g3 = env.goal
env.step(env.action_space.sample())
g4 = env.goal
assert g0 is g1
assert g1 is g2
assert g2 is not g3
assert g3 is g4
def test_goal_change_every_2_step():
env = FourRoomsEnv(goal_duration_steps=2)
env.reset()
g0 = env.goal
env.step(env.action_space.sample())
g1 = env.goal
env.step(env.action_space.sample()) # Goal change here
g2 = env.goal
env.reset()
g3 = env.goal
env.step(env.action_space.sample())
g4 = env.goal
env.step(env.action_space.sample()) # Goal change here
g5 = env.goal
assert g0 is g1
assert g1 is not g2
assert g2 is g3
assert g3 is g4
assert g4 is not g5
def test_state_dict():
env0 = FourRoomsEnv()
env1 = FourRoomsEnv()
env2 = FourRoomsEnv()
state = env0.state_dict()
def sample_chain(env, steps):
output = []
term = True
for _ in range(steps):
if term:
x = env.reset()
output.append(x)
else:
a = env.action_space.sample()
output.append(a)
x, r, term, _ = env.step(a)
output.append(x)
return output
x0 = sample_chain(env0, 10)
env1.load_state_dict(state)
x1 = sample_chain(env1, 10)
assert all([(a == b).all() for a, b in zip(x0, x1)])
env1.load_state_dict(state)
x1 = sample_chain(env1, 10)
assert all([(a == b).all() for a, b in zip(x0, x1)])
x2 = sample_chain(env2, 10)
assert any([(a != b).any() for a, b in zip(x0, x2)])
env2.load_state_dict(state)
x2 = sample_chain(env2, 10)
assert all([(a == b).all() for a, b in zip(x0, x2)])