Skip to content

CHANGELOG.md

Latest commit

 

History

History
56 lines (42 loc) · 4.52 KB

CHANGELOG.md

File metadata and controls

56 lines (42 loc) · 4.52 KB
Raw

Changelog

All notable changes to this repository and its accompanying Harvard Dataverse release are documented here.

The format follows Keep a Changelog, and this project adheres to Semantic Versioning for its schema and metadata artefacts (see MAINTENANCE.md § 4).

[Unreleased]

Added

  • Repository scaffolding aligned with NeurIPS Datasets & Benchmarks Track artefact requirements: LICENSE, LICENSE-DATA.md, CITATION.cff, AUTHORS.md, RESPONSIBILITY.md, MAINTENANCE.md, CHANGELOG.md, CONTRIBUTING.md.
  • Three-setting benchmark documentation under benchmark/, including per-setting READMEs (setting_a, setting_b, setting_c), a shared loader/metric contract (utils/), and placeholder result CSVs under benchmark/results/.
  • Cross-wave feature-alignment report (docs/feature_alignment.md, mirrored as preprocessing/feature_alignment.md).
  • Per-wave QC thresholds and alias map in preprocessing/pipeline_decisions.md.
  • Long-form docs under docs/: dataset_overview.md, ethics.md, consent_form_en.md, neurips_db_checklist.md.
  • Top-level requirements.txt and optional environment.yml for reproducibility.
  • Orphan-code pointers: READMEs in basemodel-benchmarking/ and domain_adaptation/ now cross-reference the benchmark/ protocol docs.

Changed

  • README.md rewritten to remove [DATASET_NAME] / [TODO] placeholders and to reflect the three-setting benchmark, full model inventory, reproducibility policy, and repository layout.
  • DATASHEET.md expanded with per-wave QC numbers, ABC Logger reference, 8-stage pipeline summary, and explicit [TBD on acceptance] markers for fields held for peer review.
  • data/README.md and data/schema.md reconciled: pkl files are now consistently documented as 5-element tuples (features, y, groups, t, datetimes), label filenames are lowercase (valence.pkl, mental.pkl, valenceChange.pkl, …), and feature-column counts are corrected (8,037 / 10,122 / 10,581 total; 7,961 / 10,102 / 10,550 after PIF# removal).
  • EDA/utils.py and EDA/README.md updated to reflect the released D1 / D2 / D3 folder naming on disk while keeping internal dict keys as D-1 / D-2 / D-3 for notebook compatibility.
  • .gitignore patched to whitelist documentation assets (images/*.png, metadata/checksums.md5, benchmark/results/*.csv|*.json).
  • License, data-access, and citation metadata now keep code license, data license, DUA terms, authors, and final dataset/paper names explicitly marked as TBD until public release.
  • README visuals expanded with preserved study-design, coverage, temporal-label, and embedding figures under images/.
  • EDA notebook directory casing normalised from eda/ to EDA/ to match documentation links.

Deprecated

  • The placeholder [DATASET_NAME] strings across older drafts are no longer used; the working name is CrossUserDataset pending final naming on acceptance.

Removed

  • Stale .gitkeep placeholders under benchmark/setting_a|b|c|utils/, now replaced by real README files.

Fixed

  • Contradictions between data/README.md (5-tuple) and data/schema.md (DataFrame). The tuple description matches EDA.utils.load_and_attach in the code and is the canonical format.
  • Inconsistent label filenames across README and data folder.
  • Feature-column counts in the schema (previously "~10,000 for D1; ~20,000 for D2/D3"; now corrected to 8,037 / 10,122 / 10,581 total).

Security / Privacy

  • No data-level changes. Privacy transformations (MD5-hashed contacts, UUID-replaced MAC addresses, GPS longitude displacement) remain as previously documented.

[1.0.0] — TBD on acceptance

Initial public release of the dataset on Harvard Dataverse, accompanying the NeurIPS Datasets and Benchmarks Track paper.

Added

  • D1 archive (Feb–Apr 2020, 92 participants post-QC, 10,259 ESM responses).
  • D2 archive (Dec 2020–Jan 2021, 99 participants post-QC, 21,042 ESM responses).
  • D3 archive (Nov 2021–Jan 2022, 106 participants post-QC, 21,838 ESM responses).
  • Per-wave pre-extracted feature matrices ({label}.pkl), EsmResponse.csv, UserInfo.csv.
  • Three-setting benchmark definitions (Setting A / B / C) with fixed splits, eligibility filters, and result templates.
  • Full Datasheet for Datasets and Croissant metadata.