How do I stop the reinforcement learning training and ensure agent is saved when the Episode Manager Crashes. (Stop button on episode manager not accessible)

While training using the episode manager crashed, however training is still ongoing for a long period.

Please, how can I stop the training and stll save the trained agent. What would be the impact of stopping it with the defaut stop button on MATLAB (Image attached); and would the agent be the trained agent ?

Thanks.

Answer 1

[Emmanouil Tzorakoleftherakis]

If you use the stop button, the agent you will get will have the parameters learned to that point. You can then continue training from that state.

How would you know when to stop training to avoid the crash though? It seems to me that you should figure out what's causing the crash first. If it's not easy to do by yourself, consider submitting a technical support request.