Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Feat/add impala #138

Merged
merged 5 commits into from
Mar 9, 2025
Merged

Feat/add impala #138

merged 5 commits into from
Mar 9, 2025

Conversation

EdanToledo
Copy link
Owner

What?

Add Impala sebulba system.

Why?

Useful to have a sebulba system that is not PPO.

@EdanToledo EdanToledo merged commit e33a7fa into main Mar 9, 2025
3 checks passed
@EdanToledo EdanToledo deleted the feat/add_impala branch March 9, 2025 20:35
@roger-creus
Copy link

Are there any performance benchmarks with the IMPALA implementation? :)

@EdanToledo
Copy link
Owner Author

Hi, unfortunately not. I don't have the time to do large scale performance benchmarks. I wanted to do an ATARI-5 run which I still might do but comparing it against the true implementation of impala which is recurrent might be a little unfair.

@roger-creus
Copy link

I understand. I was just wondering if you have seen it "train" at least

@EdanToledo
Copy link
Owner Author

oh, yes, i have seen it learn to solve all the simple env pool environments although i will say its worse than PPO in terms of sample efficiency and performance (which makes sense since it only updates once on the data). It also seemed to be more finnicky to hyperparams.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants