FastAPI prediction with a pre-trained model

Deploying AI into Production with FastAPI

Matt Eckerle

Software and Data Engineering Leader

Setting up the environment

Required libraries:

FastAPI: framework for building APIs with Python
uvicorn: fast ASGI server that runs Python web apps
joblib: for loading the model

from fastapi import FastAPI
import uvicorn
import joblib
# Create the FastAPI app instance
app = FastAPI()

Loading the pre-trained penguin classifier

Trained on the Palmer Penguins dataset
Predicts penguin species based on 4 features: culmen length, culmen depth, flipper length, and body mass
Output: Adelie, Chinstrap, or Gentoo

import joblib

# Load the pre-trained model
model = joblib.load('penguin_classifier.pkl')
# Check data type of model to verify model loading
print(type(model))

<class 'sklearn.pipeline.Pipeline'>

¹ https://huggingface.co/SIH/penguin-classifier-sklearn

Uvicorn

ASGI (Asynchronous Server Gateway Interface) server
Built by and for Python

uvicorn main:app \
        --host 0.0.0.0 \
        --port 8080

import uvicorn
uvicorn.run(app, 
            host="0.0.0.0", 
            port=8080)

Uvicorn logo

Creating the prediction endpoint

# FastAPI prediction endpoint
@app.post("/predict")
def predict(culmen_length_mm, culmen_depth_mm, 
            flipper_length_mm, body_mass_g):

    features = [[culmen_length_mm, culmen_depth_mm,
                 flipper_length_mm, body_mass_g]]

    prediction = model.predict(features)[0]
    return {"predicted_species": prediction}

Running the application

if __name__ == "__main__":
    uvicorn.run(
      app, 
      host="0.0.0.0", 
      port=8080)

Save all the code in a python file - your_api_script.py

$ python3 your_api_script.py

Uvicorn startup logs

Testing the API

curl \

  -X POST "http://localhost:8080/predict" \

  -H "Content-Type: application/json" \

  -d '{"culmen_length_mm": 39.1,
       "culmen_depth_mm": 18.7,
       "flipper_length_mm": 181,
       "body_mass_g": 3750}'

{
    "prediction": "Adelie",
    "confidence": 0.87
}

Let's practice!

Deploying AI into Production with FastAPI