Releasing the dataset¶
Code: research/release/export.py · Decision: ADR 0020 · Output: research/data/release/
The data files are not in git. This step packages them into one zip, without commit messages, ready to upload to Zenodo.
Building the package¶
Run it after every earlier step, from the repo root:
uv run python -m research.release.export
It writes research/data/release/modelbump-migrations-1.0.0/ and modelbump-migrations-1.0.0.zip, prints the rows in each file and the zip's SHA-256, and checks that no file has a message column. Use --version for a later release.
| File | Contents |
|---|---|
candidates.csv |
Every commit found, with one field per filter rule and the labels |
migrations.csv |
Every migration_away commit, with its timing (timing) and target (targets) |
changed_files_sample.csv, changed_files.csv |
The RQ4 sample (changed files) |
validation*/ |
The hand-label sheets, without the message column |
stats/ |
Every step's stats file |
README.md, SHA256SUMS |
What the files are, and a checksum for each |
Publishing on Zenodo¶
- Sign in at zenodo.org with GitHub, then choose New upload.
- Upload the zip. Resource type: Dataset. Version:
1.0.0. Licence: CC BY 4.0. - Fill in the title, the authors and the description from the package's README. Add the GitHub repository and Kim's arXiv paper as related works.
- Click Get a DOI now! to reserve the DOI, then Save the draft. Do not publish yet: a published record cannot be deleted.
- Put the reserved DOI in the dataset card's "Cite" section.
The record is published when the paper is submitted.