How to get the history action value of reinforcement learning agent

I want to get the action value of reinforcement learning agent during each traning episode and try to figure out if the action is optimized during the traning. Is there a way to get the action value? Thank you very much.

Answer 1

[Emmanouil Tzorakoleftherakis]

Not exactly sure what you mean. During training the RL algorithms are already doing inference. You can use getAction and getValue if you want do do inference yourself. Would that work?

[Kacjer Frank]

Thank you for the reply. I mean if the action value of reinforcement learning agent for each training step can be obtained so that we can use it to draw the curve of the action value.

[Emmanouil Tzorakoleftherakis]

If you are using Simulink, you can log any signal makes sense to you using Simulink logging options

[Kacjer Frank]

Thank you for your prompt reply. Yes, I am using the Simulink and I know the logging option. The question is that the simulation restarts and the signal starts from scratch for each training episode. What I want to do is try to log the action signal for each sample time or several sample time during the whole training process rather than just one training episode. Do you have suggestions for that? Thank you again.

[hamad alduaij]

How do we obtain the log using RL toolbox not simulink? If I needed to know the action, state value of the system during the training process.

[TRONG LINH VU]

I have the same concern with Hamad.