Unsupported Action Space Exception (Dict with a DQN)

RLlib
1

I am probably doing something wrong, but I’m trying to do a DQN on a vector of actions (centralized decision for a multi-agent setting). The action space I have chosen is Dict.

In the build step in RLlib, I get this error:

Action space Dict(1: Discrete(5), 2: Discrete(5), 3: Discrete(5)) is not supported for DQN.

Should I modify my action space or is it something I am doing incorrectly ?
Thank you in advance.

2

Hi @Douae_Ahmadoun and welcome to the forum,

maybe you could share a reproducable example or at least some code how you setup the algorithm?

Simon