Project Overview

This project demonstrates how to perform continuous delivery for a Python-based machine learning application using the Flask web framework.

We use a sklearn model, pre-trained 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 more.

The project uses a flask app, app.py that serves out predictions (inference) about housing prices through API calls.

Setup the Environment

Step 1: Launch Cloud Shell

Go to the Azure Console and launch a bash shell:

      

Step 2: Clone the Repo and Create a Flask ML Service

1. Clone the Repo:

       git clone git@github.com:mhaywardhill/sklearn-flask-webapp.git

      

2. Set up the Python virtual environment:

      python3 -m venv ~/.Dev-Ops
      source ~/.Dev-Ops/bin/activate

3. Run make all

      

4. Create an app service and deploy the app in the Cloud Shell:

      az webapp up -n <your-appservice> --sku FREE

      

5. Verify the deployed application works by browsing to the deployed url:

Go to https://<your-appservice>.azurewebsites.net/ and you should see the same output as in the screenshot below:

      

Step 3: Perform Prediction

Change the line in make_predict_azure_app.sh to match the deployed prediction:

      -X POST https://<yourappname>.azurewebsites.net:$PORT/predict

      ./make_predict_azure_app.sh

A successful prediction will look like this:

      

Step 4: Deploy project to a Azure Pipeline

Next, we'll need to create an Azure DevOps project and pipeline. Please refer to the offical documentation for more detail.

Performance Testing

Below is a screenshot of the performance testing using Locust.

      

Improvements

Both CI and CD streams run in parallel; this not ideal as we should only deploy the application if it passes the tests, in GitHub Actions.