Reinforcement Learning Episode Manager opens for a few seconds and closes afterwards

2 views (last 30 days)
I am training a reinforcement agent using the R2023a version in Ubuntu 20.04 LTS. The option in agent training to see training progress is to Plots= ‘training-progress’. When training is initiated, the Reinforcement Learning Episode Manager opens for a few seconds and closes afterwards, but the training continues correctly.
Could anyone suggest possible solutions to keep the window open through the training?
Thanks for any tips,
Best regards,
Joshi Kumar V

Answers (1)

Anurag on 21 Nov 2023
Edited: Anurag on 22 Nov 2023
Hi Joshi,
I understand that your Reinforcement Learning Episode Manager closes after a few seconds of initiating the training process while keeping the training process uninterrupted.
Please follow the steps below to tackle the issue you’re facing:
  • Ensure that your graphics card drivers are up to date as they may cause issues pertaining to visualisations.
  • MATLAB might be allocating low memory to the process due to memory constraints, check and increase the memory allocation using the command “memory”
  • Use different rendering options for example refer to the below code snippet :
options = rlTrainingOptions('Plots','none'); % Disable plots
agent = train(agent,env,options);
Please refer to the below documentation to learn more about "memory" :
Hope this helps!
Stefan on 29 Nov 2023
Edited: Stefan on 30 Nov 2023
Hello Anurag,
One of the main benefits of that window is that it shows you plots and allows you to see graphically if your model is behaving in a way that is desireable. As such, the disable plots option you've provided isn't a viable solution.
Also, I've even turned off parallel pooling while running a simple RL model, and I too am having this issue. Sometimes it works and the RL Manager window stays open, other times it closes unexpectedly.
From the memory suggestion (I am using R2023a on Windows), this is my output:
>> memory
Maximum possible array: 20381 MB (2.137e+10 bytes) *
Memory available for all arrays: 20381 MB (2.137e+10 bytes) *
Memory used by MATLAB: 6733 MB (7.060e+09 bytes)
Physical Memory (RAM): 32453 MB (3.403e+10 bytes)
* Limited by System Memory (physical + swap file) available.
I would assume that this would be more than enough to keep open a GUI and track training progess...
I'm guessing this is a larger bug that for some reason closes the RL Manager GUI unexpectedly.
Any help would be appreciated.
Stefan on 30 Nov 2023
Edited: Stefan on 13 Dec 2023
*Edit: Simply don't close out the training window with the plots that shows up. If you never close it out, whatever trigger is there isn't set to off, and you can just save and hit rerun, just don't close the old windows.

Sign in to comment.




Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!