Deploying an ML Model on Heroku with FastAPI

The third project ML DevOps Engineer Nanodegree by Udacity.

Overview

This project is part of the Udacity course, "Machine Learning DevOps Engineer". Instructions are available in udacity's repository. The project is to build a machine learning application that predicts whether a person's annual income is more than $50K using the census income dataset from UCI. The application is deployed using FastAPI, with CI/CD using Github Actions and Heroku.

Prerequisites

Python and Jupyter Notebook
AWS account with S3 bucket
Github account to use Github Actions for CI
Heroku account for CD
Linux environment may be needed within windows through WSL
In addition to the following CLI tools

AWS CLI
Heroku CLI
Dependencies
This project dependencies are available in the requirements.txt file.

Installation

Use the package manager pip to install the dependencies from the requirements.txt. It's recommended to install it in a separate virtual environment.

pip install -r requirements.txt

Usage

The config file contains MODEL variable of RandomForestClassifier. Each model with a set of parameters for the grid search PARAM_GRID. You can use your own model with the parameters needed. The SLICE_COLUMNS variable holds the columns for slice evaluation.

1. Starting training

cd src
python training_job.py

This saves a seralized model, generates evaluation results and slice evaluation results.

2. Starting FastAPI app

cd src
uvicorn app.api:app --reload

3. FastAPI app documentation

http://127.0.0.1:8000/docs

4. Testing the data and the model

cd src
pytest -vv

5. Showing tracked files with DVC

dvc dag

6. CI using GitHub Actions will be triggered upon pushing to GitHub

git push

7. CD is enabled from within Heroku app settings

8. Testing deployment on Heroku, demo get request to root endpoint

https://deploy-ml-model-with-github.herokuapp.com/

9. Testing deployment on Heroku, demo post request

python request_heroku.py