BikeFlow

BikeFlow is a team learning project that will grow into an end-to-end MLOps system for forecasting hourly urban bicycle-rental demand. It follows the requirements of the MLOps course and uses the Seoul Bike Sharing Demand dataset.

Current stage

This first stage provides a small, tested FastAPI boundary, a preliminary model/API contract, CI, and a non-root Docker image. The API currently uses a fixed development-only stub model; there is no real ML model, persistence, DVC remote, MLflow, monitoring, orchestration, deployment, or UI.

Participant A (EgorMa1tsev) owns data, DVC, preprocessing, training, and the ML artifact. Participant B (daya-alexandra) owns the repository, API, tests, CI, Docker, and later platform infrastructure. The integration agreement is documented in docs/contracts/model_api.md.

Install and run

Python 3.11 is required.

python -m venv .venv
# Windows PowerShell: .venv\Scripts\Activate.ps1
# Linux/macOS: source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e ".[dev]"
uvicorn bikeflow.api.main:app --reload --host 127.0.0.1 --port 8000

OpenAPI UI is available at http://127.0.0.1:8000/docs and the schema at http://127.0.0.1:8000/openapi.json.

Verify

make lint
make test
make docker-build

The equivalent cross-platform commands are ruff check ., ruff format --check ., pytest, and docker build --tag bikeflow:local ..

Example prediction

curl -X POST http://127.0.0.1:8000/predict \
  -H "Content-Type: application/json" \
  -d '{
    "prediction_time": "2026-07-15T08:00:00+09:00",
    "temperature_c": 24.5,
    "humidity_pct": 61,
    "wind_speed_m_s": 1.8,
    "visibility_10m": 1800,
    "dew_point_c": 16.4,
    "solar_radiation_mj_m2": 1.2,
    "rainfall_mm": 0,
    "snowfall_cm": 0,
    "holiday": false,
    "functioning_day": true
  }'

The response contains the requested time, predicted_rentals, and model_version: "stub-v0".

Docker

docker build --tag bikeflow:local .
docker run --rm -p 8000:8000 bikeflow:local

Collaboration workflow

main is protected after the bootstrap commit. Create a focused branch such as feat/model-baseline, feat/dvc-pipeline, fix/predict-validation, or docs/model-contract; use Conventional Commits and open a pull request. A merge needs passing CI and one approval. Prefer squash merge; GitHub deletes merged branches automatically.

Do not commit datasets, trained models, binary artifacts, .env, credentials, or tokens.