Built by William Kang (Ching-Wei Kang) - William Kang profile - portfolio and identity page - Ching-Wei Kang profile - projects profile.

• English (this file)
• 中文: README.zh-CN.md

A lightweight EEG brain-computer interface MVP focused on:

• EEG preprocessing (EDF)
• relaxed vs focused classification
• real-time stable inference (EMA + hysteresis)
• reproducible evaluation and release workflow
• Unsupervised Domain Adaptation (CORAL)
• Stacking Ensemble Classifiers (RF+SVM+MLP)
• Advanced SHAP explainability
• Real-time debounced streaming

• Hugging Face Space: https://huggingface.co/spaces/williamKang112/bci-mvp-demo

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

make check
make train

# API
make api

# Dashboard
streamlit run app/dashboard.py

bci-mvp/
  src/            # preprocessing, training, eval, reports, release tooling
  app/            # streamlit apps
  api/            # fastapi service
  docs/           # reports / readiness / release artifacts
  assets/         # generated charts/badges
  tests/          # unit tests

# Full pipeline
python src/run_full_pipeline.py
# or
make full

# Final release candidate summary
python src/final_release_candidate.py
python src/quick_health_cli.py

# Space status + smoke test
python src/hf_space_status.py --space williamKang112/bci-mvp-demo
python src/space_smoke_test.py

• Mathematical model: docs/MATH_NOTATION.md
• Limitations: docs/LIMITATIONS.md
• Methods (paper-style): docs/METHODS.md
• Results summary: docs/RESULTS.md
• Technical report: docs/TECHNICAL_REPORT.md
• Model card: docs/MODEL_CARD.md

• Quality scorecard: docs/QUALITY_SCORECARD.md
• Compliance scorecard: docs/COMPLIANCE_SCORECARD.md
• Release readiness: docs/RELEASE_READINESS.md
• HF Space readiness: docs/HF_SPACE_READINESS.md
• Final RC summary: docs/FINAL_RELEASE_CANDIDATE.md
• Release dashboard: docs/RELEASE_DASHBOARD.md
• Launch status: docs/LAUNCH_STATUS.md
• Release guard report: docs/RELEASE_GUARD_REPORT.md
• v1 release notes: docs/V1_RELEASE_NOTES.md
• release archive manifest: docs/RELEASE_ARCHIVE_MANIFEST.md

• Docs home: docs/HOME.md
• One pager: docs/ONE_PAGER.md
• Space user guide: docs/SPACE_USER_GUIDE.md
• Docs bundle index: docs/DOCS_BUNDLE_INDEX.md
• Command center: docs/COMMAND_CENTER.md
• Figure gallery: docs/FIGURE_GALLERY.md
• Release packet: docs/RELEASE_PACKET.md

MIT

If this project helps your work, please cite using CITATION.cff.

• 2026-03-23 00:13:58 UTC — Added v1-release-gate regression test and CI coverage
• Full log: logs/progress.md

python src/fetch_public_eeg_data.py

This fetches EEGBCI public data and prepares:

• data/relaxed/*.edf
• data/focused/*.edf

• docs/RESULTS_BRIEF_EN.md

Measure subject-level generalization (Leave-One-Subject-Out):

python src/cross_subject_eval.py

Output:

• outputs/cross_subject_results.json

python src/plot_cross_subject.py

Output:

• assets/cross_subject_loso.svg

Test generalization by holding out one subject at a time:

python src/subject_holdout_eval.py

Output:

• outputs/subject_holdout_results.json

python src/repro_one_command.py

See: docs/REAL_DATA_EVIDENCE.md

To reproduce the advanced experiments implemented in this MVP:

make benchmark   # Train RF and SVM baselines
make ensemble    # Train StackingClassifier (RF + SVM + MLP)
make coral       # Cross-Dataset Unsupervised Domain Adaptation (CORAL)
make shap        # Feature Importance Analysis (SHAP Summary Plots)
make stream-viz  # Real-time Streaming Timeline Simulation
make test        # Run Unit Tests automatically (CI)
make tune        # Automated Hyperparameter Tuning (RandomizedSearchCV)