vwxyzjn / cleanrl

High-quality single file implementation of Deep Reinforcement Learning algorithms with research-friendly features (PPO, DQN, C51, DDPG, TD3, SAC, PPG)
http://docs.cleanrl.dev
Other
5.26k stars 602 forks source link

Prototype RLops Utility #307

Closed vwxyzjn closed 1 year ago

vwxyzjn commented 1 year ago

Description

Related to #296. To get started, run the following command

python rlops.py --exp-name ddpg_continuous_action_jax \
    --wandb-project-name cleanrl \
    --wandb-entity openrlbenchmark \
    --tags  v1.0.0b2-9-g4605546 latest \
    --env-ids Hopper-v2 Walker2d-v2 HalfCheetah-v2 \
    --output-filename compare.png \
    --report

This PR generates this wandb report (which does not display properly due to https://github.com/wandb/wandb/issues/4421; update: a temporary fix pip install git+https://github.com/wandb/wandb.git@6ccd140dabb0af529302b07f4d7ecc1235773900 works (https://github.com/wandb/wandb/tree/andrew/report-api-fixes)) and the following image

image

As part of this effort, I have labeled all of the existing runs in openrlbenchmark with rlops-pilot if the do not already have a label.

Types of changes

Checklist:

vercel[bot] commented 1 year ago

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name Status Preview Updated
cleanrl ❌ Failed (Inspect) Dec 15, 2022 at 3:07AM (UTC)
vwxyzjn commented 1 year ago

Support more advanced filter syntax

python -m cleanrl_utils.rlops --wandb-project-name cleanrl \
    --wandb-entity openrlbenchmark \
    --filters 'ddpg_continuous_action_jax?user=joaogui1&tag=rlops-pilot' 'ddpg_continuous_action_jax?user=joaogui1&tag=pr-298' \
        'ddpg_continuous_action_jax?user=costa-huang&tag=rlops-pilot' 'ddpg_continuous_action_jax?user=costa-huang&tag=pr-298' \
        'ddpg_continuous_action?user=costa-huang&tag=pr-299' 'ppo_continuous_action?user=costa-huang&tag=rlops-pilot'  \
    --env-ids Hopper-v2 Walker2d-v2 HalfCheetah-v2 \
    --output-filename compare.png --report

which produces the following image and a wandb report with the same line color, thanks to @andrewtruong's amazing PR pip install git+https://github.com/wandb/wandb.git@6ccd140dabb0af529302b07f4d7ecc1235773900

image image
vwxyzjn commented 1 year ago

Closed by #368