iwillsolvehardestproblem's picture
4 games x 2 protocols: raw dream-vs-real frame dumps
994572c verified
|
Raw History Blame Contribute Delete
1.22 kB
metadata
license: mit
tags:
  - world-model
  - dream-drift
  - atari
  - evaluation
  - dreamerv3
pretty_name: dream-drift DreamerV3 eval frames

dream-drift · DreamerV3 evaluation frames

Raw dream-vs-reality frame dumps behind the DreamerV3 rows of the dream-drift benchmark. Re-score every published number without a GPU.

One file per game and protocol. <game>id.npz = in-distribution (the model's own replay trajectories); <game>.npz = cross-policy (expert DIAMOND-policy trajectories, the "confident healing" regime).

Arrays per file:

  • real uint8 (32, 50, 64, 64, 3): ground-truth frames (HWC), 32 start states x 50 steps
  • dream uint8 (32, 4, 50, 64, 64, 3): 4 dream samples per start, same recorded actions, open loop
  • starts (32,): start indices into the source trajectory; game, burnin (=4), seed (=0) scalars
  • rrew, drew (32, 50) / (32, 4, 50): real and predicted rewards (predicted are raw; threshold |r|<0.5 to 0, then sign)
  • rend, dend: real and predicted termination flags

Scoring: python harness/dreamer_score.py <dir-with-raw_frames.npz> in the dream-drift repo reproduces summary.json, drift.png, and the videos.