Operationalize a Machine Learning Microservice API using Docker and Kubernetes

Project Introduction

This project uses Docker and Kubernetes to operationalize a Machine Learning Microservice API, a Flask app that serves out predictions (inference) about housing prices through API calls.

A pre-trained sklearn model is given to predict housing prices in Boston according to several features, such as average rooms in a home and data about highway access, teacher-to-pupil ratios, and so on. The data is taken from Kaggle.

This project could be extended to any pre-trained machine learning model, such as those for image recognition and data labeling.

Environment Setup

• Run make setup to create a virtual env
• Run source venv/bin/activate to activate the virtual env
• Run make install to install the necessary dependencies

Run the Flask App

1. Standalone: python app.py
2. Run in Docker: ./run_docker.sh
3. Run in Kubernetes: ./run_kubernetes.sh
4. Run a simple test: ./make_prediction.sh

Files Explanation

• model_data/
  • stores the pre-trained model for prediction.
• output_txt_files/
  • stores some example outputs after running run_docker.sh and run_kubernetes.sh.
• app.py
  • A Flask app that processes and passes input data through the prediction model, and responds with the resultant prediction.
• run_docker.sh
  • This script builds the Docker image for the app, and runs the containerized app locally.
• upload_docker.sh
  • This script tags and uploads the Docker image to Docker Hub.
• run_kubernetes.sh
  • This script pulls the Docker image from Docker Hub, and deploys it to the local kubernetes cluster.
• make_prediction.sh
  • This script sends some input data to our app's prediction API, and outputs the response. Run this script to test if the app is running properly.