I am newbie in reinforcement learning and trying to understand how to implement continuous actions bounded by [-2, 2]. My research shows that doing nothing is a possible solution (i.e. action of 4.5 is mapped to 2 and action of -3.1 is mapped to -2 but I wonder if there are more elegant approaches.

Thanks for any advice.

  • $\begingroup$ Normalizing a list of values can be done against the maximum. If the values in the list are in the range between 0 and 1 it's possible to scale the interval up to the needed [-2..2] range. In the literature this is called a probability distribution which is modified to bring it into the desired range. $\endgroup$ – Manuel Rodriguez Aug 19 '19 at 15:53
  • $\begingroup$ I need a clarification. I understand that neural network can theoretically predict any action value from 0 to infinity. Thus I will never know the maximum before predictions of actions are made. Could you please elaborate?I was thinking adding tanh activation to output. Could you please elaborate? I was thinking about adding tanh activation to output if it makes sense. $\endgroup$ – TFbie Aug 19 '19 at 23:06

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.