如何在 Python 的 ChainerRL 中扩展代理 class

Question

我想在 ChainerRL 中扩展 PPO 代理 class。我做了以下事情：

class exPPO(chainerrl.agents.PPO):
    
    def act_and_train(self, obs, reward):
        action = chainerrl.agents.PPO(self, obs, reward)
        print("this is my exPPO act and train")
        return action

我尝试使用 cartpole env of gym，但是做的时候

obs, reward = env.step(action)

它只是崩溃，输出如下

this is my exPPO act and train
Traceback (most recent call last):
  File "C:\personal_if\extendPPO.py", line 184, in <module>
    obs, reward, done, _ = env.step(action)
  File "C:\Users\PareekHi\AppData\Local\Programs\Python\Python37\lib\site-packages\gym\wrappers\time_limit.py", line 16, in step
    observation, reward, done, info = self.env.step(action)
  File "C:\Users\PareekHi\AppData\Local\Programs\Python\Python37\lib\site-packages\gym\envs\classic_control\cartpole.py", line 104, in step
    assert self.action_space.contains(action), err_msg
AssertionError: <chainerrl.agents.ppo.PPO object at 0x000002A986D2F508> (<class 'chainerrl.agents.ppo.PPO'>) invalid

请帮助我如何在此处扩展 PPO class。

Answer 1

action = super().act_and_train(obs, reward)

https://docs.python.org/3/library/functions.html#super

如何在 Python 的 ChainerRL 中扩展代理 class

How to extend an agent class in ChainerRL in Python

python

reinforcement-learning

chainer