4-atari-hard: Go-Explore (exploration phase) on Montezuma's Revenge + benchmark#132
Open
dnddnjs wants to merge 1 commit into
Open
4-atari-hard: Go-Explore (exploration phase) on Montezuma's Revenge + benchmark#132dnddnjs wants to merge 1 commit into
dnddnjs wants to merge 1 commit into
Conversation
… benchmark Go-Explore phase 1 (Ecoffet et al. 2019 / Nature 2021), no neural net: an archive of downscaled-frame cells (11x8, 9 gray levels), emulator state save/restore to return to frontier cells, repeated random actions to explore from them. 12 explorer processes over raw gymnasium ALE (envpool exposes no clone API, hence the separate env_go_explore.py). Result: best end-of-episode score 31,000 at 500M agent steps (~5.5h on a Mac Studio M4 Max), single seed, replay-verified (re-executing the stored 5,336-action demo from reset reproduces the score exactly). Deterministic protocol (no sticky actions) -- a trajectory-search result, not an RL policy score; see the README caveat.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Go-Explore phase 1 (exploration only) on Montezuma's Revenge — the archive + emulator-restore paradigm, side by side with the PPO+RND row.
Best end-of-episode score: 31,000 at 500M agent steps (~5.5h, Mac Studio M4 Max, 12 explorer processes, no neural network). Single seed. Replay-verified: re-executing the stored 5,336-action trajectory from reset reproduces exactly 31,000.
Protocol notes (also in the README block):
W&B (full metrics history + gameplay video): https://wandb.ai/rlcode/rl-atari-hard-go-explore/runs/m6ox4l3m
(Single-seed diagnostic run; merge is a human decision.)