Apppublic

ProCreations/repro-frictional-q-learning

sourceHugging Faceupdated 2mo agoView on Hugging Face

0likes

App README

Reproduction bundle

Deterministic coherent-buffer, extrapolation-error, orthogonal-complement, contrastive-separation, and released MuJoCo raw-run reaggregation for all six registered claims. The paper source archive contains no executable author implementation; the official implementation and author-linked raw event archive are hash-recorded in outputs/official_runs_reaggregation_manifest.json.

bash

python3 -m pip install -r requirements.txt
PYTHONDONTWRITEBYTECODE=1 PYTHONWARNINGS=error python3 reproduce.py --output-dir /tmp/fql-replay
PYTHONDONTWRITEBYTECODE=1 python3 validate_evidence.py
PYTHONDONTWRITEBYTECODE=1 python3 verify_manifest.py

The MuJoCo comparison tables reaggregate five-seed author event streams at the released 4,990,000-step evaluation checkpoint. The critic-versus-augmentation conclusion preserves the HalfCheetah-v4 exception in the matched raw runs.