reinforcement learning - DQN in huge state-space

Question

I have a problem in which I have 240 input state vectors (10-bit each), two action sets, one with four possible action vectors (24-bit each) and other with 10 -bit vector (one-hot encoded).

These action vectors are independent of the state. The rule is to pick the best actions one from each set in order to achieve the optimization goal.

Do you think the state-space or action-space is so huge? DQN works well in that situation well?

Can I also use the possible action-set as an input to DQN and get best possible action?

I would be happy if someone guides me with this!

Thanks and best regards, Ammad Ali

question from:https://stackoverflow.com/questions/66045554/dqn-in-huge-state-space

Answer 1

Waitting for answers