Evaluation of the reinforcement learning agent

Question

rr0101 2021-7-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/878348-evaluation-of-the-reinforcement-learning-agent

回答： Gagan Agarwal 2024-3-4

Hello

I am using the Reinforcement learning toolbox. I trained an rlDQNAgent then I am using it. The training is well done as you can see.

The problem is, once I wanted to use this trained agent, it always gives me the same action for different observations, I tried with 10000 different observations and the action given is always action number 1.

I don't know what is going wrong

Any suggestions

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Gagan Agarwal 2024-3-4

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/878348-evaluation-of-the-reinforcement-learning-agent#answer_1420471

Hi rr0101

When an rlDQNAgent in the Reinforcement Learning Toolbox of MATLAB always returns the same action for different observations, it could be due to several reasons.

Here are some potential reasons for the behaviour :

Check the Observations - Ensure observations are normalized or standardized during training.
Evaluate the Training Process - Investigate potential overfitting or underfitting
Confirm the possibility of overfitting or underfitting of the network

By checking the above scenarios you should be able to identify and address the issue which is causing the rlDQNAgent to return the same action.