v0.1.2 - Public release fixes

Summary: patch release for the public repository state

Changes:

• Fixed the clean first-run dataset download path
• Fixed reporting serialization for saved training metrics
• Updated Python 3.11+ metadata, scripts, and docs
• Added regression coverage for downloader and training flows

Verification:

• python -m ruff check .
• pytest tests/ -v -W default --tb=short
• python -m compileall src
• cd docs && make clean && make html
• DATASET_DIR=$(mktemp -d ...) ./run_quick_training.sh
• ./run_full_training.sh

v0.1.1 - Publication Hardening and Reproducibility Update

Summary: Released a publication-ready patch update that hardens packaging and documentation, aligns the repository's public claims with the implemented workflow, and adds reproducible environment guidance.

Changes:

• clarified the README and Sphinx docs to distinguish the Python connectivity classifier from the original 2018 paper and the linked MATLAB workflow
• added declared documentation dependencies in pyproject.toml and published requirements-lock.txt for reproducible installs
• standardized local environment setup on .venv across installation guidance and helper scripts
• fixed the remaining test-suite lint issues and bumped the project and docs metadata to 0.1.1

Verification:

• python -m ruff check .
• python -m pytest -q (58 passed)
• python -m compileall src
• cd docs && make clean && make html

v0.1.0 - Initial Release

Covert Consciousness Detector v0.1.0

Initial release of a machine learning pipeline for automated consciousness detection from fMRI data.

Features

• XGBoost ensemble classifier with Leave-One-Subject-Out cross-validation
• Feature extraction from fMRI connectivity data (446 brain regions)
• Works with Michigan Human Anesthesia Dataset (OpenNeuro ds006623, 25 subjects)
• Complete pipeline: preprocessing, feature extraction, training, and evaluation

Documentation

See repository for installation guide, dataset documentation, and model details.

⚠️ Research tool only - not for clinical use without validation and regulatory approval.