def testSharedRandomState(self):
env = MockEnv(action_dim=1, observation_dim=1, run_mode='multiprocess')
initial_rand_obj = copy.deepcopy(env._rand_obj_)
initial_values = initial_rand_obj.uniform(-1, +1, 3)
env.start()
for _ in range(3):
env.step(0)
updated_rand_obj = np.random.RandomState()
updated_rand_obj.set_state(utils.get_random_state_from_array(env._shared_rstate_array_))
safe_final_values = updated_rand_obj.uniform(-1, +1, 3)
unsafe_final_values = env._rand_obj_.uniform(-1, +1, 3)
env.step(0)
env.close()
assert np.all(initial_values == unsafe_final_values)
assert np.all(initial_values != safe_final_values)
assert np.all(safe_final_values == env._uniform_array_)
def test_random_state_array(self):
rand_obj = np.random.RandomState(1)
rand_state = rand_obj.get_state()
original_uniform_values = rand_obj.uniform(-1, 1, 100)
original_normal_values = rand_obj.randn(100)
rand_state_array_type, rand_state_array_size, rand_state_array = get_random_state_array(rand_state)
shared_rand_array = np.frombuffer(Array('b', rand_state_array_size).get_obj(), dtype=rand_state_array_type)
np.copyto(shared_rand_array, np.frombuffer(rand_state_array, dtype=rand_state_array_type))
new_rand_obj = np.random.RandomState()
new_rand_obj.set_state(get_random_state_from_array(shared_rand_array))
new_uniform_values = new_rand_obj.uniform(-1, 1, 100)
new_normal_values = new_rand_obj.randn(100)
assert np.all(original_uniform_values == new_uniform_values)
assert np.all(original_normal_values == new_normal_values)