Puffer doesn't beat my simple env - credit assignment testing#478
Open
eitanporat wants to merge 1 commit intoPufferAI:3.0from
Open
Puffer doesn't beat my simple env - credit assignment testing#478eitanporat wants to merge 1 commit intoPufferAI:3.0from
eitanporat wants to merge 1 commit intoPufferAI:3.0from
Conversation
Simple environment where agent must pick action 0 on step 1 to win. Episode terminates at step 128, reward given only at termination. - 2 discrete actions - 50% random baseline - Tests long-horizon credit assignment with BPTT
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Results:
The core issue is that truncated BPTT cuts off gradients at segment boundaries, preventing credit from flowing back to early actions. A potential fix would be to perform two forward passes per rollout: the first to collect experiences, and the second (after seeing more of the trajectory) to compute improved bootstrap value estimates at segment boundaries. This would allow the value function to incorporate information beyond the BPTT horizon without requiring full backpropagation through the entire episode.
I leave this as an open problem for other contributors.