It is a model wrapped into the container, and orchestrated by Airflow, which predicts 'Rented Bike Count' based on bike sharing​ data. Data and MLflow artifacts are stored on S3 (localstack).

Setup

Prerequisites

mkdir -p ./dags ./logs ./plugins ./config

Set environmental variables with your AWS secrets:

export AWS_ACCESS_KEY_ID="your_access_key_id" AWS_SECRET_ACCESS_KEY="your_secret_access_key"

and with Airflow user ID:

export AIRFLOW_UID=$(id -u)

Starting the Services

Prepare and start containers:

docker compose build
docker compose up airflow-init
docker compose up -d

Wait about a minute until all services will be up.

Accessing Services

• Airflow: http://localhost:8080/ (user/password: "airflow")
• MLflow: http://localhost:5000/
• Prediction API: http://localhost:8000/

Using the Prediction API

The prediction service provides the following endpoints:

• GET /health: Check service health
• POST /predict: Make predictions using the following JSON format:

Feature descriptions:

• hour: Hour of the day (0-23)
• temperature: Temperature in Celsius
• humidity: Humidity percentage (0-100)
• wind_speed: Wind speed in m/s
• visibility: Visibility in km
• dew_point_temperature: Dew point temperature in Celsius
• solar_radiation: Solar radiation in MJ/m2
• rainfall: Rainfall in mm
• snowfall: Snowfall in cm

Example curl request:

curl -X POST "http://localhost:8000/predict" \
     -H "Content-Type: application/json" \
     -d '{
           "hour": 12,
           "temperature": 20.5,
           "humidity": 65,
           "wind_speed": 2.5,
           "visibility": 2000,
           "dew_point_temperature": 15.5,
           "solar_radiation": 1.2,
           "rainfall": 0,
           "snowfall": 0
         }'

Stopping Services

To stop all containers:

docker compose down --volumes