Tuned examples not working

Dime · March 5, 2023, 1:12am

How severe does this issue affect your experience of using Ray?

High: It blocks me to complete my task.

Hello there,

I’ve been trying to run the tuned examples on DQN.

ray-project/ray/blob/master/rllib/tuned_examples/dqn/atari-duel-ddqn.yaml

# Runs on a single g3.4xl node
# See https://github.com/ray-project/rl-experiments for results
dueling-ddqn:
    env:
        grid_search:
            - ALE/Breakout-v5
            - ALE/BeamRider-v5
            - ALE/Qbert-v5
            - ALE/SpaceInvaders-v5
    run: DQN
    config:
        # Works for both torch and tf.
        framework: tf
        env_config:
            frameskip: 1  # no frameskip
        double_q: true
        dueling: true
        num_atoms: 1
        noisy: false
        replay_buffer_config:

This file has been truncated. show original

The only thing I’ve changed is the replay buffer size to 150k samples.

The results for Breakout aren’t anything near the scores provided by the results page.
Any pointers to what could be going wrong?

arturn · April 13, 2023, 11:45pm

Sorry for now answering this earlier. Let’s keep the discussion in Trouble reproducing results with DQN.

Topic		Replies	Views
Trouble reproducing results with DQN RLlib	3	427	April 14, 2023
Issue with Custom PyTorch Model in Ray RLlib RLlib	0	282	November 3, 2023
Custom sampling for dqn RLlib	1	215	June 13, 2023
Saved ONNX model using DQN dueling policy RLlib	4	464	March 25, 2023
RLLIB not working with Tune with sample batch input RLlib	25	2494	October 4, 2022

Tuned examples not working

Related topics