[RLlib] DeepMind preprocessor not working as expected #45186
Labels
bug
Something that is supposed to be working; but isn't
rllib
RLlib related issues
rllib-oldstack-cleanup
Issues related to cleaning up classes, utilities on the old API stack
What happened + What you expected to happen
With the PPO algorithm (hyperparameters configured as per the original paper) on the Atari environments, the in-built deepmind preprocessor is not working as expected. I get the following error which suggests that the model is getting the raw environment observation (with 3 rgb channels) instead of the deepmind preprocessed observation (framestack of 4):
To run the reproduction script, simply save it to a file and run it with python e.g.,
python ppo_atari.py
Versions / Dependencies
python = 3.11
ray[tune,rllib] = 2.20.0
os: ubuntu lts
Reproduction script
Issue Severity
High: It blocks me from completing my task.
The text was updated successfully, but these errors were encountered: