Custom Training Loop with configured DDPG Agent

Question

Allmo 2022-5-30

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1730070-custom-training-loop-with-configured-ddpg-agent

回答： Poorna 2023-8-30

Hello,

is there the possibility to do a custom training loop with a already configured DDPG agent?

Background: I want to check after each episode whether the average reward has reached a new maximum. When a new maximum is reached, the agent should be saved to a mat file, otherwise not in order to reduce the amount of data. In the training options, I can only set a limit for when the agent should be saved. But then all agents are saved as soon as the limit is exceeded.

Thanks!

Best regards,

allmo

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

Poorna 2023-8-30

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1730070-custom-training-loop-with-configured-ddpg-agent#answer_1297046

在 MATLAB Online 中打开

Hi,

I understand that you would like to save the agent after every episode based on whether the new episode's reward is greater than the existing average reward. You can achieve this by using “rlDataLogger”.

Create a new “FileLogger” object as shown below:

fileLgr = rlDataLogger();

Then, you can do this:

fileLgr.EpisodeFinishedFcn = @myEpisodeFinishedFcn;

where myEpisodeFinishedFcn is your custom function which implements the logic to conditionally save the agent to disk.

Hope this helps!