Operationalizing a Machine Learning Microservice API

Overview of Repo

This project is python flask appthat runs a machie learning api, designed to run on three different services; locally, on docker as a contain and on kubernetes - as a container too. The api is that of a pre-trained sklearn model that predicts housing prices in Boston according to available data. I wish I could talk more on the API but that is not why I am writing this. Incase you want to check it out, visit https://www.kaggle.com/c/boston-housing.

Aim of Project

The aim is to operationalize the machine learning api as a microservice using kubernetes - ultimately - and to xarry out linting tests using CircleCi

Requirements

Github Account
CircleCi Account
DockerHub Account
AWS Cloud9 (Optional)

Serving The Flask App

There are three ways to run this app:

Local

#Setup a python virtual environment and activate it
python3 -m venv ~/.devops
source ~/.devops/bin/activate

#Install the necessary dependencies
make install

#Run the main application
python app.py

Docker Container Ensure that you update the parameters on parameters.sh. ./run_docker.sh Application port => http://localhost:8000

Run the ./upload_docker.sh command will upload your image to Dockerhub For more details on building a docker, check official documentation at https://docs.docker.com/engine/reference/commandline/build/

Kubernetes There are different ways to set up kubernetes clusters. For testing, you can use MiniKube. Check the official installation guide of Minikube for more details: https://kubernetes.io/docs/tasks/tools/install-minikube/ ./run_kubernetes.sh Application listening port => http://localhost:5000

Making Predictions

While the application is running in any of the environments above, run ./make_predictions.sh to make calls to the ML API.

Summary of Files in Project

In no particular order:

circleci/config.yml: Defines pipeline of circleci
app.py: The Flask Python app code
requirements.txt: Contains all prerequisites of Python packages for app.py to run
run_kubernetes.sh: Runs the API as a Kubernetes deployment
make_prediction.sh: Calls the api and makes prediction when the app is running
model_data/boston_housing_prediction.joblib: Pretrained sklearn model for the app
Dockerfile: Container configuration for the application
Makefile: Install dependencies, lints and setup using make command
run_docker.sh: Builds the Dockerfile into an image and runs it
upload_docker.sh: This script tags the Dockerimage built and uploads in to YOUR Dockerhub
output_txt_files/docker_out.txt: Console output from executing run_docker.sh and make_prediction.sh
output_txt_files/kubernetes_out.txt Console output from executing run_kubernetes.sh and make_prediction.sh

mmd-afegbua/mlops-microservice