ExperienceBuffer has 0 Length when i load a saved agent and continue training in reinforcement training
9 views (last 30 days)
Show older comments
Yikai
on 5 Apr 2021
Commented: Dmitriy Ogureckiy
on 12 Jan 2023
Hi all,
I'm trying to train a saved agent further. In the training option of this saved agent, the SaveExperienceBufferWithAgent is set to true. But when I load the saved_agent and open the property ExperienceBuffer I noticed the Length is 0. I tried to look in the documentation of such property but the there is no information on it. If I stop a training and directly check the property "Length" of the agent in the workspace, it has some value.
My question would be what does this "Length" mean? If it's 0, when I perform training further with a saved agent like in https://de.mathworks.com/matlabcentral/answers/495436-how-to-train-further-a-previously-trained-agent?s_tid=answers_rc1-2_p2_MLT , does it really continue training with saved agent and with saved expeirence buffer?
Yours
0 Comments
Accepted Answer
Takeshi Takahashi
on 20 Apr 2021
Length 0 means there isn't any experience in this buffer. I think it didn't save the experience buffer due to this bug. Please set agent.AgentOptions.SaveExperienceBufferWithAgent = true immediately before saving the agent.
2 Comments
Dmitriy Ogureckiy
on 12 Jan 2023
Can I ask you, does networks weights saved when agent saved between simulations?
More Answers (0)
See Also
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!