Fastapi-ML

Fastapi-ML is a project that demonstrates how to deploy machine learning models using FastAPI, a modern, fast (high-performance) web framework for building APIs with Python. This repository provides a step-by-step guide to building and deploying a machine learning model API on Heroku, enabling easy integration of machine learning capabilities into web applications.

Key features:

• Implementation of a RESTful API using FastAPI.
• Integration of pre-trained machine learning models for inference.
• Deployment on Heroku for easy and scalable cloud hosting.
• Minimal Examples and best practices for serving machine learning models in production.

Whether you're looking to deploy your machine learning models in the cloud or explore FastAPI as a web framework, this repository serves as a practical guide for getting started.

Dependencies:

fastapi==0.95.2
uvicorn==0.22.0

Install dependencies using

python -m pip install requirements.txt

ML model

We will be using Prophet to predict stock market prices.

Deployment to Heroku

For this tutorial, we will be hosting the application on heroku server. Please sign up for a Heroku account, and then install the Heroku CLI.

Don't forget too add payment method.

heroku login

heroku create

Next, we will be using the Heroku Container Registry to deploy the application with Docker using:

heroku container:login

Make sure that you have install Docker successfully and it is currently running. it can be tested using docker ps in the terminal or command-line.

Add a Dockerfile file to the project root (refer to our Dockerfile).

Add a .dockerignore file (refer to our file as well).

Build the Docker image using command:

docker build -t registry.heroku.com/aqueous-wildwood-97407/web .

The running of docker will take awhile to finish.

Then run command below to test the function.

docker run --name fastapi_ml -e PORT=8008 -p 8008:8008 -d registry.heroku.com/aqueous-wildwood-97407/web:latest

Open another CMD and enter:

curl --header "Content-Type: application/json" --request POST --data "{\"ticker\":\"MSFT\"}" http://localhost:8008/predict

you should receive output in JSON format something like

{"ticker":"MSFT","forecast":{"09/02/2024":447.36687922756676,"09/03/2024":447.65588717590146,"09/04/2024":447.9448951242364,"09/05/2024":448.2339030725712,"09/06/2024":448.522911020906,"09/07/2024":448.81191896924093,"09/08/2024":449.1009269175758}}

If it working as expected.

stop the docker from running and remove the docker.

docker stop fastapi_ml

docker rm fastapi_ml

Push the image to the registry:

docker push registry.heroku.com/aqueous-wildwood-97407/web

WARNING: the name of the container must match with the app created in heroku

We must create heroku.yml file (already created in this repo) to refer to the DockerFile for deployment. we then run:

can skip the first two lines if there is an error

git add .
git commit -m "adding heroku.yml"
git push heroku main

Once the compiled and build is completed, you would see it deploy under your heroku application webpage.

Then left click the Container icon and word found on the left side of the star to go to the settings and find the link to app like https://<YOUR_HEROKU_APP_NAME>.herokuapp.com/.

Open your browser and go to https://<YOUR_HEROKU_APP_NAME>.herokuapp.com/docs to get into FastAPI page and test the API. For this tutorial, enter "MSFT" as the value for the key "ticket" to see the testing output.

Congratulation

Congrats!!! we have build a minimal deployment of ML model on heroku for stock prediction.

End note:

Feel free to ask me any questions and provide suggestion for this guide.

Check out the streamlit app!!

https://stock-predict-demo.streamlit.app/

Contact me:

[email protected]