obs = env.reset()
agent.reset()
# print('HAND: ', end=''); print(env.player_hand)
# print('STATE pp={0}, ha={1}, dc={2}'.format(obs[0], obs[1], obs[2]))
while True:
action = agent.pick_action(obs)
# print('ACTION: {0}'.format(action.action))
# --- time step rolls here ---
obs, reward, done = env.step(action)
# print(' ------ t={0} -----'.format(env.t_step))
# print('REWARD: {0}'.format(reward))
# print('HAND: ', end=''); print(env.player_hand)
# print('STATE pp={0}, ha={1}, dc={2}'.format(obs[0], obs[1], obs[2]))
# print('DONE', done)
agent.remember(obs, reward)
if done:
agent.learn_mc()
break
# print(' == GAME OVER =='.format(env.t_step))
for episode in range(numEpisode):
# 重新初始化环境
observation = env.reset()
print(
"#######################################################################"
)
print("第{}回合".format(episode))
# 每个episode运行100步
for step in range(100):
# 显示当前的observation
displayObservation(observation)
# 获取当前observation的action
action = policy(observation)
# 根据当前action,与环境交互，获取下一步的observation， reward, 是否done等信息
observation, reward, done, _ = env.step(action)
print("根据当前策略，玩家采用的行为为：{}, 该行为获取的立即返回为: {}".format(["停止要牌",
"继续要牌"][action],
reward))
# 如果done为True
if done:
displayObservation(observation)
print("游戏结束。 回报为: {}\n".format(float(reward)))
break
# 该回合结束，暂停几秒
time.sleep(3)