Out of Memory during serialization of the subsystem data - reinforcement learning

Hi, during training (reinforcement learning), agents aren't saved properly and this warning is given. What can be done to fix this? Training is done with parallel processing.

[Zaid Jaber]

Hello

did you solve this?

Answer 1

[Harsh]

Hi Charissa,

The error you are facing while saving agents during training is because of insufficient disk space. While creating the “rlTrainingOptions” object you can specify several options to configure when and where to save the agents which will help you resolve the error. Follow the steps below to resolve your error:

• Set the “SaveAgentDirectory” to a folder with more space so that the agents can be saved properly.
• By default, the agent might be saved frequently. Reducing the save frequency can alleviate memory and disk space issues. Set the “SaveAgentCriteria” to “GlobalStepCount” to save the agent when the total number of steps in all episodes equals or exceeds the critical value.

Refer to the following documentation page to understand more about training options related to saving an agent while training - https://www.mathworks.com/help/releases/R2021a/reinforcement-learning/ug/specify-training-options-in-reinforcement-learning-designer.html#:~:text=an%20averaging%20condition.-,Specify%20Additional%20Options,-To%20specify%20additional