Packaging ML Models in Containers

Packaging your Python ML models for containers is a crucial step in making your solutions portable and production-ready. To begin, you need to organize your project in a way that aligns with container best practices. Start by structuring your project directory with clear separation of source code, model artifacts, dependencies, and configuration files. Place your model-serving scripts in a dedicated folder (such as src/), keep your trained model files in a models/ directory, and use a requirements.txt file to specify all necessary dependencies. This ensures that your container will have all the libraries it needs to run your model.

Next, define a clear entry point for your application. This is typically a Python script that loads your trained model and exposes it via an API endpoint or command-line interface. The entry point should be referenced in your container configuration so that the container knows which process to start when it runs.

Your project directory might look like this:

• src/: source code for serving the model;
• models/: serialized model files;
• requirements.txt: all Python dependencies;
• Dockerfile: instructions for building the container image;
• README.md: documentation for usage.

By following this structure, you make it straightforward for others (and for automated systems) to build and run your model in a containerized environment.

import pickle
from http.server import BaseHTTPRequestHandler, HTTPServer
import json

# Load trained model
with open('models/trained_model.pkl', 'rb') as f:
    model = pickle.load(f)

class ModelRequestHandler(BaseHTTPRequestHandler):
    def do_POST(self):
        content_length = int(self.headers['Content-Length'])
        post_data = self.rfile.read(content_length)
        input_data = json.loads(post_data.decode('utf-8'))
        prediction = model.predict([input_data['features']])
        response = {'prediction': prediction[0]}
        self.send_response(200)
        self.send_header('Content-type', 'application/json')
        self.end_headers()
        self.wfile.write(json.dumps(response).encode('utf-8'))

def run(server_class=HTTPServer, handler_class=ModelRequestHandler, port=8080):
    server_address = ('', port)
    httpd = server_class(server_address, handler_class)
    print(f"Serving model API on port {port}")
    httpd.serve_forever()

# Entry point for the container
run()