Model Deployment

The ML Pipeline (Production)

1. Training (Jupyter Notebook)
   ↓
2. Serialize Model (Save to file)
   ↓
3. Load Model in API (Flask/FastAPI)
   ↓
4. Accept HTTP Requests
   ↓
5. Preprocess Input Data
   ↓
6. Make Prediction
   ↓
7. Return JSON Response
   ↓
8. Monitor & Retrain (if performance drops)

Saving Models

Pickle (General Python Objects)

import pickle

# Save
with open('model.pkl', 'wb') as f:
    pickle.dump(model, f)

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

Joblib (Large NumPy Arrays)

import joblib

# Save
joblib.dump(model, 'model.joblib')

# Load
model = joblib.load('model.joblib')

Creating an API with Flask

from flask import Flask, request, jsonify
import joblib

app = Flask(__name__)
model = joblib.load('model.joblib')
scaler = joblib.load('scaler.joblib')

@app.route('/predict', methods=['POST'])
def predict():
    data = request.json['features']  # e.g., [[5.1, 3.5, 1.4, 0.2]]
    
    # Preprocess
    data_scaled = scaler.transform(data)
    
    # Predict
    pred = model.predict(data_scaled)
    
    # Return
    return jsonify({'prediction': int(pred[0])})

if __name__ == '__main__':
    app.run(port=5000)

Containerization with Docker

FROM python:3.9
WORKDIR /app
COPY model.joblib .
COPY app.py .
RUN pip install flask scikit-learn
EXPOSE 5000
CMD ["python", "app.py"]

Build and run:

docker build -t my-ml-api .
docker run -p 5000:5000 my-ml-api

Deployment Platforms

Production Considerations

1. Input validation: Reject bad inputs
2. Monitoring: Track prediction accuracy, latency
3. Logging: Save all requests for debugging
4. Caching: Store recent predictions
5. A/B testing: Compare old vs new model
6. Retraining: Schedule automatic retraining