def test_last_always_plays():
env = make_env(last_agent_always_plays=True)
obs = env.reset()
assert obs['youre_playing_self'][-1, 0]
ac = {'action_defect': np.random.randint(0, 2, size=(env.metadata['n_actors']))}
for i in range(1000):
obs, _, done, _ = env.step(ac)
assert obs['youre_playing_self'][-1, 0]
if done:
obs = env.reset()
assert obs['youre_playing_self'][-1, 0]
def test_last_first_versus_last():
env = make_env(last_step_first_agent_vs_last_agent=True)
prev_obs = env.reset()
ac = {'action_defect': np.random.randint(0, 2, size=(env.metadata['n_actors']))}
for i in range(1000):
obs, _, done, _ = env.step(ac)
if done:
assert prev_obs['youre_playing_self'][-1, 0]
assert prev_obs['youre_playing_self'][0, 0]
obs = env.reset()
prev_obs = deepcopy(obs)
def test_last_doesnt_play_until():
env = make_env(last_doesnt_play_until_t=5)
ac = {'action_defect': np.random.randint(0, 2, size=(env.metadata['n_actors']))}
obs = env.reset()
done = False
t = 0
for i in range(1000):
if t < 5:
assert not obs['youre_playing_self'][-1, 0]
obs, rew, done, info = env.step(ac)
t += 1
if done:
obs = env.reset()
done = False
t = 0
def _test_fixed_policy(against_all_d=False, against_all_c=False):
env = make_env(against_all_d=against_all_d, against_all_c=against_all_c,
last_agent_always_plays=True)
prev_obs = env.reset()
for i in range(1000):
currently_playing = np.squeeze(prev_obs['youre_playing_self'])
ac = {'action_defect': np.random.randint(0, 2, size=(env.metadata['n_actors']))}
obs, rew, done, info = env.step(ac)
if against_all_d:
assert np.all(rew[currently_playing & (ac['action_defect'] == 0)] == -2)
assert np.all(rew[currently_playing & (ac['action_defect'] == 1)] == 0)
elif against_all_c:
assert np.all(rew[currently_playing & (ac['action_defect'] == 0)] == 2)
assert np.all(rew[currently_playing & (ac['action_defect'] == 1)] == 4)
else:
assert False
assert np.all(rew[~currently_playing] == 0)
prev_obs = obs
if done:
prev_obs = env.reset()