Config custom RNN for DQN

RLlib
1

Can I ask how to config my custom RNN to my custom DQN in Ray?
The one mentioned like this

2

Hey @jonathonyan , you could take a look at our R2D2 algorithm (ray/rllib/agents/dqn/r2d2.py), which is basically a DQN that requires that the model be an RNN (you can use any custom LSTM model for this as well). I think to make R2D2 run as an “APEX”-style algo, we’d have to merge R2D2 with APEX (take R2D2, but use APEX’s execution plan and maybe per-worker epsilon exploration setup). I have never tried this, but would you want to give it a shot?

3

Thank you for your reply. Does that mean except for RNN, other setting of DQN(the original one) and R2D2 are the same?