1

I need to train an RL agent, that have to control some switches. Lets imagine, that we have n switches, that could be turned on (1) or turned off (0). my agent have to decide in each step which one of the to torn of, and turn of, so i want action look like this: [1,0,1,....1] - vector that consists of n binary elements.

How can i train an agent? Something beside DQN with converting given action space to discreet one with 2^n actions possible?

Yaroslav Svyryda
  • 11
  • 1
  • [Artificial Intelligence Stack Exchange](https://ai.stackexchange.com/) is probably a better place to ask theoretical questions related to reinforcement learning, so I suggest that you ask your question there. If you ask it there, please, delete it from here (to avoid cross-posting, which is generally discouraged). – nbro Oct 31 '20 at 15:00

1 Answers1

0

DQN also can support a vectorized action space if this is what you are asking. See TF-Agents for a framework where this is already implemented.

Federico Malerba
  • 755
  • 3
  • 8