Fundamentals of MLOps
Automating Insurance Claim Reviews with MLflow and BentoML
Demo Upgrade Python Flask App to Connect to BentoML for Online Serving
Welcome to this lesson on how to integrate a Flask application with BentoML for online prediction serving. In this tutorial, you will learn how to set up a simple Flask application that accepts CSV file uploads, processes the data, communicates with the BentoML predict endpoint, and then displays prediction results in an intuitive user interface.
Below is an overview of the steps covered in this tutorial:
Step 1: Setting Up the Flask Application
Begin by launching your VS Code editor and opening a new terminal. Also, open a separate terminal window to start the BentoML service, which must be running to serve predictions.
Create a file named flaskapp.py and include the following code. This code establishes an endpoint that receives a POST request containing a Base64-encoded CSV file. The CSV content is decoded and converted into a DataFrame. Should the DataFrame include a claim_id column, it is temporarily separated from the rest of the data. The remaining data is then sent as JSON to the BentoML predict endpoint. Upon receiving the prediction results, they are merged back with the original DataFrame, and the results are rendered through an HTML template.
from flask import Flask, render_template, request
import pandas as pd
import requests
import base64
import io
app = Flask(__name__)
@app.route('/')
def index():
return render_template('index.html')
@app.route('/predict', methods=['POST'])
def predict():
file_data = request.form.get('file')
# Decode the Base64 encoded CSV file content
decoded_file = base64.b64decode(file_data.split(',')[1])
# Read CSV content into a DataFrame
df = pd.read_csv(io.StringIO(decoded_file.decode('utf-8')))
# Separate the 'claim_id' column if it exists
if 'claim_id' in df.columns:
claim_ids = df['claim_id']
df = df.drop(columns=['claim_id'])
else:
claim_ids = None
# Send the DataFrame to the BentoML service as JSON
response = requests.post(
'http://127.0.0.1:3000/predict/', # BentoML endpoint
json=df.to_dict(orient='records')
)
# Retrieve predictions from the response
predictions = response.json()['predictions']
df['Prediction'] = predictions
# Reattach the 'claim_id' column if it was present
if claim_ids is not None:
df['claim_id'] = claim_ids
# Render the results in the results.html template
return render_template('results.html', tables=[df.to_html(classes='data')])
if __name__ == '__main__':
app.run(port=5005)
Note: Ensure that both Flask and BentoML are installed in your Python environment to avoid import errors.
Step 2: Creating HTML Templates
To provide a straightforward user interface, create a folder named templates in your project directory. Inside this folder, establish the following HTML templates:
1. index.html
This template displays the file upload interface for users. (Customize this file according to your design preferences.)
2. results.html
The results.html file displays prediction results formatted in a table. Below is an example template with basic table styling:
<html lang="en">
<head>
<style>
table, th, td {
border: 1px solid black;
}
th, td {
padding: 10px;
text-align: left;
}
</style>
</head>
<body>
<h1>Prediction Results</h1>,
{{ table|safe }},
<a href="/">Go Back</a>
</body>
</html>
,[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object]