ms, ma = env_model.sample_s_a() # ms in here is a str
mr, ms_ = env_model.get_r_s_(ms, ma)
RL.learn(ms, ma, mr, str(ms_))
# print(env_model.database)
# print('################')
# print(RL.q_table)
# print('################')
s = s_
s2 = s2_
if done:
s = env.reset()
break
if done2:
s2 = env.reset2()
break
# end of game
print('game over')
print(RL.q_table)
env.destroy()
if __name__ == "__main__":
env = Maze()
env_model = EnvModel(actions=list(range(env.n_actions)))
RL = QLearningTable(actions=list(range(env.n_actions)))
env.after(0, update)
env.mainloop()
while True:
env.render()
a = RL.choose_action(str(s))
s_, r, done = env.step(a)
RL.learn(str(s), a, r, str(s_))
# use a model to output (r, s_) by inputting (s, a) 输入状态和一个动作输出其奖励和下一个状态
# the model in dyna Q version is just like a memory replay buffer
env_model.store_transition(str(s), a, r, s_)
for n in range(10): # learn 10 more times using the env_model
ms, ma = env_model.sample_s_a() # ms in here is a str
mr, ms_ = env_model.get_r_s_(ms, ma)
RL.learn(ms, ma, mr, str(ms_))
s = s_
if done:
break
# end of game
print('game over')
env.destroy()
if __name__ == "__main__":
env = Maze()
RL = QLearningTable(actions=list(range(
env.n_actions))) #直接强化学习，通过环境体验来优化策略模型
env_model = EnvModel(actions=list(range(env.n_actions))) #通过环境体验来优化策略和优化模型
env.after(0, update)
env.mainloop()