About Running experiments for testing R2D2 algorithm using tune

RLlib
1

Dear Ray,

I am trying to run some experiments to test R2D2 algorithm’s performance in a custom environment. However I am getting errors and the experiments do not run. I tried initializing the configuration of the R2D2 based on a script that I found on github but when I am executing the command

from ray.rllib.algorithms.r2d2.r2d2 import R2D2Config

I am getting the following error

ModuleNotFoundError: No module named ‘ray.rllib.algorithms’

I am running Ray version 1.12.1 in my workstation and 1.12.0 in my laptop. The same problem occurs in both.

Thank you very much for your help!

Christos Peridis

2

Dear Ray Team,

If I skip the above mentioned command and just execute tune.run() with “R2D2” as algorithm of choice, whether I am using tf or torch buckend I am getting this error:

AssertionError: R2D2 requires its model to be a recurrent one! Try using model.use_lstm or model.use_attention in your config to auto-wrap your model with an LSTM- or attention net.

Unfortunatelly I do not know how and where to use this command (model.use_lstm).

Please let me know what actions I shall take in order to make my script run.

Thank you very much for your help!

Christos Peridis