question about"policy gradient reinforcement learning"

[fhtcumt]

Hi,

  In the policy gradient toolbox which Jan Peters provide, there are three types of policies, i.e., the decision border policy, the ε-soft Gibbs policy, and the Gaussian policy. He  consider two kind of problems, discrete problems and linear-quadratic regulation problems.The Gaussian policy is applied in  linear-quadratic regulation problems.So I have a question,i.e.,can the Gaussian policy be used in the discrete  or the nonlinear problems?

In my simulation,I need a type of  policy whose  prameter  is one dimension ,so  which  policy  can  meet the situation？

   Another  question is "how to understand the queueing  problem  in reinforcement  learning ?"

 

  So  someone here can give me a hint? Many thanks!

 

Best regards,

MRS FENG

---

"雪见"杨幂邀你共品3D国韵网游《天下贰》，海报免费领