for obs in self._visited_states_returns:
self._visited_states_returns[obs] += reward
if __name__ == '__main__':
env = BlackjackEnv()
agent = BlackjackAgent()
total_ep = 100000
for episode in range(total_ep):
if episode % 1000 == 0:
print('episode: {0} / {1}', episode, total_ep)
# print()
# print(' == EPISODE {0} START =='.format(episode))
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)
# 将observation进行分解
score, dealerScore, usableSce = observation
# 如果 玩家 的分数大于20分时，停止要牌（stick）, action为0;
# 否则 玩家 继续要牌（twist）， action为1
if score >= 20:
return 0
else:
return 1
# 显示10个episode的信息
numEpisode = 10
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)