def evaluate_second_agent(env, policy1, reward1, env2, policy2, reward2):
print("Acting agent enters")
init = [1, 1]
initial_state = KitchenState(init[0], init[1], set(), init[0], init[1])
env.reset(initial_state)
reward1.reset()
policy1.reset(evaluation=True)
trial_reward: float = 0.0
for step in range(TEST_EPISODE_LENGTH):
s0 = env.state
a = policy1.get_best_action(s0)
env.apply_action(a)
s1 = env.state
print_state(s0, a)
step_reward, finished = reward1(s0, a, s1)
if not finished:
trial_reward += step_reward
logging.debug("(%s, %s, %s) -> %s", s0, a, s1, step_reward)
if finished:
print_state(s1, -1)
break
facts = set()
if env.state.facts[OBJECTS4["open"]]:
facts.add(1)
if env.state.facts[OBJECTS4["up"]]:
facts.add(4)
print("Agent 2 enters")
env2.reset(KitchenState(init[0], init[1], facts, init[0], init[1]))
trial_reward2 = 0
for step in range(TEST_EPISODE_LENGTH):
s0 = env2.state
a = policy2.get_best_action(s0)
env2.apply_action(a)
s1 = env2.state
print_state(s0, a)
step_reward, finished = reward2(s0, a, s1)
if not finished:
trial_reward2 += step_reward
logging.debug("(%s, %s, %s) -> %s", s0, a, s1, step_reward)
if finished:
print_state(s1, -1)
break
return trial_reward, trial_reward2
def create_init(init_locations):
ans = []
for j in init_locations:
ans.append(KitchenState(j[0], j[1], (), j[0], j[1]))
return ans