End-to-end machine learning using FastAPI, Streamlit, Docker, GitHub Actions, Microsoft AZURE and EvidentlyAI
The Objective of the ML model is to predict a credit score for historical loan application data to predict whether or not an applicant will be able to repay a loan.
We use a pretrained model to predict credit scores and deploy that model as an API using FastAPI and use Streamlit to create a WebAPP. We use Github Actions to facilitate and automatize DevOps and MLOps (CI/CD).
Overview | Tech Stack | The data | The backend with FastAPI | Docker | Hosting with Microsoft AZURE | Github Actions
VS Code — as the IDE of choice.
pipenv — to handle package dependencies, create virtual environments, and load environment variables.
FastAPI — Python API development framework for ML endpoint deployment
Uvicorn — ASGI server for FastAPI app.
Docker Desktop — build and run Docker container images on our local machine. (MacOS 11)
Containers are an isolated environment to run any code
Azure Container Registry — repository for storing our container image in Azure cloud.
Azure App Service — PaaS service to host our FastAPI app server.
Github Actions — automate continuous deployment workflow of model serving through FastAPI and dashboarding through Streamlit app.
Streamlit - Dashboard
PyTest - Testing of Web APP functionality through
EvidentlyAI - Data Drift detection
We will cover in the readme the below concepts:
- How to create dockerfile for ML API deployment using FastAPI
- How to run different docker commands to build, run and debug and
- How to push docker image to Github using Github Actions
- How to test ML API endpoint which is exposed by the running ML API docker container
The data is provided by Home Credit.
Open a terminal
mkdir project_name
cd project_name
code .
Install pipenv
sudo -H pip install -U pipenv
pipenv install fastapi uvicorn
use pipenv shell to activate the virtual environment.
You find the FastAPI in the file main.py
import uvicorn
from fastapi import FastAPI, HTTPException
import numpy as np
import mlflow
import lightgbm
import os
from typing import List
from pydantic import BaseModel # for data validation
# load environment variables
#port = os.environ["PORT"]
# hard coded port to facilitate testing with Github Actions
port = 8000
# initialize FastAPI
app = FastAPI(title="Automatic Credit Scoring",
description='''Obtain a credit score (0,1) for ClientID.
Visit this URL at port 8501 for the streamlit interface.''',
version="0.1.0",)
# Pydantic model for the input data
class DataPoint(BaseModel):
data_point: List[float]
# 3. Expose the prediction functionality, make a prediction from the passed
# JSON data and return the predicted flower species with the confidence
@app.post('/predict')
def predict_credit_score(data: DataPoint):
""" Endpoint for ML model
Args:
list (float): one data point of 239 floats
Returns:
float: prediction probability
int: prediction score
"""
try:
print("predict_credit_score function")
#print(data)
print([data.data_point])
if (len([data.data_point]) == 0):
print("empty data set")
return {"VALUE ERROR": "Data set is empty"
}
# TEST data
#data_test = [[0, 0, 1, 1, 63000.0, 310500.0, 15232.5, 310500.0, 0.026392, 16263, -214.0, -8930.0, -573, 0.0, 1, 1, 0, 1, 1, 0, 2.0, 2, 2, 11, 0, 0, 0, 0, 1, 1, 0.0, 0.0765011930557638, 0.0005272652387098, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, True, False, False, False, False, False, False, False, True, False, False, False, False, False, False, True, False, False, False, False, True, False, False, False, True, False, False, True, False, False, False, False, False, False, False, False, False, False, False, False, True, False, False, False, False, False, False, False, False, False, False, True, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, True, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False]]
#data = {"data_point": data_test}
sklearn_pyfunc = mlflow.lightgbm.load_model(model_uri="LightGBM")
prediction = sklearn_pyfunc.predict_proba([data.data_point]).max()
return {
'prediction': prediction,
'probability': 0.8
}
except Exception as e:
error_msg = f"An error occurred during prediction: {str(e)}"
raise HTTPException(status_code=500, detail=error_msg)
@app.get("/")
def index():
return {"data": "Application ran successfully - FastAPI release v4.2 with Github Actions no staging: cloudpickle try environment pipenv"
}
#return {st.title("Hello World")}
if __name__ == "__main__":
uvicorn.run("main:app", host="0.0.0.0", port=port, reload=False)
Run your APP with
pipenv run python main.py
You can find more details on creating and deploying a FastAPI in this blog
In order to deploy our API endpoint to Azure we need to install (Docker Desktop)[https://www.docker.com/products/docker-desktop].
In the Dockerfile you need to add the libraries that your APP needs to run.
For example RUN pip install uvicorn.
For our FASTAPI API we use Python 3.8-bookworm.
Here are some useful commands to create a docker image:
docker build . -t fastapi-cd:1.0
or to run the API from the docker image
docker run -p 8000:8000 -t fastapi-cd:1.0
You can check the API at http://localhost:8000/docs.
You find a great tutorial on how to setup your infrastructure here
Using Github Actions we can automatize the Docker build process and the push to our hosting platform. We have created a yml file in the
.github/workflows folder.
docker build . -t fastapicd2024.azurecr.io/fastapi-cd:${{ github.sha }}
docker push fastapicd2024.azurecr.io/fastapi-cd:${{ github.sha }}
Details on how to setup your Github Actions with Azure you can find in this blog
You can run the dashboard locally anytime with this command but be aware that you will not get any prediction as long as the backend is not running. So make sure to start the back end first if you wish to test the dashboard together with the ML endpoint.
streamlit run 3_STREAMlit_dashboard.py
The Streamlit APP will be visible in your browser at: http://localhost:8501
Make sure Docker Desktop is running. Go to the streamlit dashboard folder 3_STREAMlit_dashboard. Then run the following command in the terminal to build the container image:
docker build . -t dashboard-cd:1.0
and the next command to test the container image:
docker run -p 8501:8501 -t dashboard-cd:1.0
In Microsoft AZURE open AZURE Container Registry
Create a new one with the following parameters:
Resource group: fastapi-cd
Container group: fastapicd2024
Registry name: fastapi-cd
Login server: fastapicd2024.azurecr.io
Open Terminal and run:
az login
az acr login --name fastapicd2024
then try to build the container that is pushed to AZURE:
docker build . -t fastapicd2024.azurecr.io/dashboard-cd:1.1
docker images
docker push fastapicd2024.azurecr.io/dashboard-cd:1.1
Now you just need to create a Web APP service instance and connect to your container:
Go to AZURE APP Service and create a new instance:
In Basics tab
Resource group: fastapi-cd
Name: credit-score-dashboard
Publier: Container Docker
And in the Docker tab
Image source: Azure Container Registry
Registry: fastapicd2024
Image: dashboard-cd
Startup command: streamlit run 3_STREAMlit_dashboard.py
Now got to to your WEB APP at: https://credit-score-dashboard.azurewebsites.net/
You find the python code test in the file ```5_unittest.py````