Why when I set the UseFastRestart = "on" and start train my reinforcement learning agent, the matlab crash manager comes out and matlab hast to close?

Question

Anastasios Seisopoulos on 19 May 2023

0
Link

Direct link to this question

https://in.mathworks.com/matlabcentral/answers/1967769-why-when-i-set-the-usefastrestart-on-and-start-train-my-reinforcement-learning-agent-the-matlab

Answered: Emmanouil Tzorakoleftherakis on 23 May 2023

I have created a Simulink model, where I have 2 batteries blocks with aging functions in them. The agent takes as observation the State of Charge of the Batteries, their State of Health and the power profile. The actions of the agent are the power distribution to the batteries according to their State of Health and State of Charge. The battery with a better State of Health receives more power than the other one.

Why when I set the UseFastRestart = "on" and start train my reinforcement learning agent, the matlab crash manager comes out and matlab hast to close. On the other hand, when UseFastRestart = "off", the agent starts training and stops only after 21 episodes, although I had set a maximum number of 5000 episodes, StopTrainingValue = 1000 and StopTrainingCriteria= "AverageReward".

Why is that happening and how can I be sure that my agent trains correctly?