This is a demo of a simple linear regression model.

python -m pip install -r requirements-dev.txt

Main dependencies are Pandas and Scikit-learn.

python .\train.py '..\data\Real estate.csv'

Rigde Regression model is traing on housing data. Output of model training will be a joblib model artifact (scikit-learn pipeline object) in folder ml/models/house_price_<timestamp>.joblib.

Feature engineering is extremely basic only to show re-use of feature transformation during training and inference stages. Potentially, pairwise features, distance to closest city, more boolean flags can be computed.

python -m pytest tests/test.py

From root directory, launch the following command:
python -m fastapi run app/service.py

Model loading is done by reading .env file containing path to the model. Sample model is already provided.

Input is validated for missing fields, boundary values and null values.

Service paylod and house price is written to predictions.csv.

To try out API service, open http://127.0.0.1:8000/docs#/default/predict_predict_post . Insert sample JSON:

{
    "X2 house age": 5,
    "X3 distance to the nearest MRT station": 11,
    "X4 number of convenience stores": 0
}

Sample output in docs page:

Endpoint input and prediction is written in local file predictions.csv.

Minimal docker image with launching multiple uvicorn workers can be built and run :

docker build -t ml_service .
docker run -p 8000:8000 ml_service

AI assisted IDE Cursor was used only during test development, Github Actions and Dockerfile. A few Scikit learn, Pandas, pytest and FastAPI reference documentation pages were accessed during rest of the development.

Development session is recorded, 15X speed up video is stored at this google drive link.

For certain reasons, the recorded screen was enalrged / cut from bottom and side.

Key time codes:

• 3.00 - FastAPI service development
• 09.07 - unit tests with Cursor Sonnet 4
• 10.45 - answers in Deployments.md