Is it possible 'LSTM net' to be the env of reinforcement learning?

I want to create 'my env' for reinforcement learning.

Many matlab examples use robots and cars as env(using simulink).

I wasn't able to make 'LSTM net' into 'myenv' by referring to them due to poor matlab skills.

I'm referring to the document'Create MATLAB Environments for Reinforcement Learning', but I couldn't find anything to implement net as env.

I have an LSTM network that has already been learned. I want to proceed with reinforcement learning using this net as env. Is there any simple code to get env?

[Stephan]

Did you check here?

https://de.mathworks.com/help/reinforcement-learning/ug/create-custom-reinforcement-learning-environment-in-matlab.html

I dont see a reason why you should not be able to write a custom environment function that uses your LSTM.

Answer 1

[Emmanouil Tzorakoleftherakis]

Hello,

Looks like your environment is in MATLAB (i.e. not in Simulink). There is not restriction on using an LSTM as an environment. What you would need to do is use current state and action as inputs to your LSTM, perform inference from inside the "step" function of the RL environment using predict or predictAndUpdateState, then extract your states/observations from the network output.