Spaces:
Running
Running
|
Download candidate_code/README.md from DineshAI/YEckWPoS09: direct link, hf CLI and curl.
- Browser
- Download file 1.28 kB
-
https://huggingface.co/spaces/DineshAI/YEckWPoS09/resolve/main/candidate_code/README.md
- Command line
-
hf download hf://spaces/DineshAI/YEckWPoS09/candidate_code/README.md
-
curl -L -o README.md https://huggingface.co/spaces/DineshAI/YEckWPoS09/resolve/main/candidate_code/README.md
1.28 kB
Faithful executable reproduction
These files are byte-for-byte mirrors of the implementation used by the fixed OpenResearch command:
uv sync --frozen && uv run python repro/src/run_publication_gate.py
The fast, judge-facing checker at ../verify_candidate.py
recomputes the critical C1–C6 quantities from the raw evidence and exits nonzero
on any mismatch. From a downloaded Space revision:
uv run --python 3.12 --with numpy==2.3.3 --with scipy==1.16.2 python verify_candidate.py
The complete application campaign can be rerun from the GitHub repository with
the fixed command above. Its pinned environment is recorded in
requirements.txt; the repository also contains the
authoritative pyproject.toml and uv.lock.
File roles:
markov_core.py: empirical transition kernels, continuous composite GLR, row-wise KL statistic, adaptive boundary, pseudo-spectral gap, and Poisson solution.run_core_campaign.py: C1–C5 experiments, independent checks, and mutants.run_applications.py: the Appendix-G MCMC and MountainCar applications.test_markov_core.py: independent analytic unit checks.source_manifest.json: SHA-256 mapping to the authoritative repository files.