Skip to content

Adds errata for Rainbow - #200

Merged
muupan merged 4 commits into
pfnet:masterfrom
prabhatnagarajan:rainbow_errata
Dec 15, 2025
Merged

muupan merged 4 commits into
pfnet:masterfrom
prabhatnagarajan:rainbow_errata

Conversation

@prabhatnagarajan

Copy link
Copy Markdown
Contributor
  • A hyperparameter (the target network update interval) is incorrect in the reproduction of Rainbow. This PR makes a note to clarify this issue.

@prabhatnagarajan

Copy link
Copy Markdown
Contributor Author

@keisuke-nakata sorry, this needs /test as well.

@muupan
muupan requested review from muupan and removed request for keisuke-nakata December 15, 2025 08:02
@muupan

muupan commented Dec 15, 2025

Copy link
Copy Markdown
Member

/test

@pfn-ci-bot

Copy link
Copy Markdown

Successfully created a job for commit db1dec1:

@muupan

muupan commented Dec 15, 2025

Copy link
Copy Markdown
Member

I forgot how we set this value, but it seems you are right.

The original paper (https://arxiv.org/abs/1710.02298) is very clear that it is 32K frames, not steps. For replay_start_size we correctly set 20K steps (= 80K frames), but somehow not for target_update_interval.

Screenshot 2025-12-15 at 17 06 52

@muupan muupan left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice catch!

@prabhatnagarajan

Copy link
Copy Markdown
Contributor Author

Feel free to merge (I can't).

@muupan
muupan merged commit 41e3997 into pfnet:master Dec 15, 2025
7 checks passed
@prabhatnagarajan
prabhatnagarajan deleted the rainbow_errata branch December 15, 2025 08:31
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.

3 participants