I am wondering how can I find the appropriate reward value for each specific problem. I know this is a highly empirical process, but I am sure that the value is not set totally at random. I want to know what are the general guidelines and practices to find the appropriate reward value for any reinforcement learning problem.
$\begingroup$
$\endgroup$
2
-
$\begingroup$ When you say "reward value", you mean the reward function (i.e. the function that gives the reward value for every action in some state), right? $\endgroup$– nbroApr 3, 2020 at 18:10
-
$\begingroup$ Yes, but I am mostly having doubts about the ''value" I should give for rewarded states, for example, the reward of a goal in the final state of an episodic task. like in a maze what should be the reward value of the goal, +1, +50 or +100 or any other number? how this is determined? $\endgroup$– Saeid GhafouriApr 4, 2020 at 16:42
Add a comment
|