I am trying to design reinforcement learning algorithm. My action and state space are continuous. Action, which I would like to take can be represented by a matrix, lets say of dimension $n \times n$. And I would like to use reinforcement learning to find optimal entries of this matrix.

So far, I looked into deep reinforce algorithm. However, it is not clear to me how to design network/algorithm, such that it allows for modifying entries of the matrix.

As far as I understood, output of the network should represent probability, with which we take certain action. I wonder, do I need to model each dimension of the action(each element of the matrix) as an independent prob. process? Is it good idea to model it by the same network?

  • $\begingroup$ Welcome to Data Science SE. It is not very clear where you are stuck, whether it is in state/action representation for RL or in neural network architecture. It seems odd that you would be stuck on a basic problem of creating a neural network with multiple outputs, but that does appear to be what you are asking in the second paragraph. It may help if you give more details of the environment for RL, the problem you are trying to solve, and any thoughts or attempts you have made towards it. $\endgroup$ – Neil Slater Jan 29 at 7:25
  • $\begingroup$ thanks for your comment, I tried to modify my question. $\endgroup$ – interesting_question Jan 29 at 7:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.