Get observation of final episode of RL agent

Hi everyone,

I statrted using RL toolbox three months ago, and I trained my RL agent. But as a part of my work I need to have the observation used in the final iteration of some episodes. Is there any opportunity to get the observation participated in a specific episode of RL training, or nor?

Also, it might be noticeable that I defined a custom environment matching with my problem, and both of the observation and state spaces are continuous.

[Ajay Vishwakarma]

Hello

Answer 1

[Emmanouil Tzorakoleftherakis]

Do you want to save the observations in the last time step of the final episode? Or all the observations shown in the final episode? Are you using a custom MATLAB or Simulink environment? Regardless, I think you can add some kind of counter and when you reach the desired episode #, just save what you need in a mat file.

It sounds like you can use the IsDone flag/condition to save what you need. You would still need a counter and when the max number of steps in the episode is reached, set IsDone to true and save the variables you need.

[shadi abpeikar]

Thanks for your response, Emmanouil.

I'm using custom MATLAB, with a custom environment defined by my own. Also, I'm using DDPG Agent, with an observation state of 14 continuous parameters. I know that I should save them in .mat file when I reached the episode. By my question is how can I reach that episode observation. For example, getAction, gets actions from the trained RL, so is there any syntax that I can get the last observation? If not how can I reach that observation?

I tried to use getsamples from the time-series of the states, but I couldn't reach the value of the last observation. Also, getObsrvationInfo does not give any access to the observation values used during the training.

[Emmanouil Tzorakoleftherakis]

It sounds like you can use the IsDone flag/condition to save what you need. You would still need a counter and when the max number of steps in the episode is reached, set IsDone to true and save the variables you need.

[shadi abpeikar]

Thanks a lot, I can't believe that It was as easy as this! You are definitely right this is the way.

[Emmanouil Tzorakoleftherakis]

Np. Please accept the answer if it answered your question