Simple Flask App Hosted by Azure

Overview

This repo contains source code to deploy a simple Flask web application via Microsoft Azure. The purpose of this repo is to demo building the Flask app, deploying the app inside of Azure cloud, and creating a CI/CD pipeline with Github Actions.

The app hosts a simple Random Forest model endpoint for predicting salaries of NHL players. As of 7/2/2024, the application can be accessed at https://flask-ml-service-v2.azurewebsites.net/. Predictions can be made against the model by sending a POST request to the url with the /predict endpoint (https://flask-ml-service-v2.azurewebsites.net/predict). Alternatively, running ./make_predictions_request_azure_app.sh from the root of this directory will send a POST request for a sample dataset of 10 players to the app endpoint. The source code of the actual Flask app can be found in app.py

NOTE: Considering the purpose of this app is geared towards building a CI/CD pipeline with GitHub Actions, creating a good performing model was not a focal point. In fact, it is a poor performing model from a metric standpoint, intentionally.

Makefile Targets

The Makefile includes five targets in total. Note, it is recommended to create a virtual environment prior to running make install. If you want to create a conda environment instead of a traditional venv environment, commands to do so are run via make install_conda_env target (described later below):

1. install: install all libraries set in requirements.txt using pip
2. lint: runs ruff against all python files in repo for linting purposes
3. format: runs black against all python files in repo for formatting purposes
4. test: test that the app endpoint returns a 200 response code
5. install_conda_env: installs all packages set in conda_env.yml inside a conda environment (assumes that a conda environment named .flask-ml-cloud-serverless-demo has already been created) and activates that environment. The purpose being that, if running locally, users can run the Flask app using the conda environment if desired instead of a traditional pypi environment.

GitHub Actions Build Commands

On pushing a new commit to the main branch, GitHub Actions has been setup to run three commands from the Makefile:

make install
make lint
make test

The GitHub Actions workflow can be seen in .github/workflows/main.yml.

Running model predictions against app endpoint.

Running ./make_predictions_request_azure_app.sh from the root of this directory will send a POST request for a sample dataset of 10 players to the app endpoint. One record of the sample data (as a python dictionary) looks like such:

sample_record = {
    "plusMinus": {
        "0": -4,
    },
    "Position": {
        "0": "D",
    },
    "G": {
        "0": 0,
    },
    "A1": {
        "0": 1,
    },
    "GP": {
        "0": 36,
    },
    "Grit": {
        "0": 139,
    },
    "TOI%": {
        "0": 23.8,
    },
    "xGA": {
        "0": 24.3,
    },
    "xGF": {
        "0": 16.7,
    },
    "First Name": {
        "0": "Alex",
    },
    "Last Name": {
        "0": "Biega",
    },
    "Team": {
        "0": "VAN",
    }