Saving and Deploying Machine Learning Models

# CHAPTER 19

Saving and Deploying Machine Learning Models #

1. Introduction #

2. Learning Objectives #

• Explain what model serialization is.

• Save and load Scikit-learn models using joblib.

• Understand the basics of REST APIs.

• Build a simple Python web server using Flask to host your model.

• Make predictions via HTTP requests.

3. Model Serialization (Saving the Model) #

Saving the Model:

import joblib
from sklearn.ensemble import RandomForestClassifier

# Assume 'my_pipeline' is fully trained
# Save it to the hard drive
joblib.dump(my_pipeline, 'house_price_model.pkl')

print("Model successfully saved!")

*You will now see a physical file named housepricemodel.pkl in your folder.*

4. Loading the Model #

import joblib
import numpy as np

# Load the model from the file
loaded_model = joblib.load('house_price_model.pkl')

# New raw data from a user (e.g., 2000 sq ft, 3 beds)
new_data = np.array([[2000, 3]])

# Predict! (The pipeline handles the scaling automatically)
prediction = loaded_model.predict(new_data)
print(f"Predicted Price: ${prediction[0]}")

5. Deployment Basics: What is an API? #

1. 1. The website (customer) sends an HTTP Request to the API (waiter): *"Here is a 2000 sq ft house."*

1. 2. The API gives the data to the Model (chef). The Model calculates the prediction: *$250,000*.

1. 3. The API returns an HTTP Response to the website: *"The predicted price is $250,000."*

6. Building an ML API with Flask #

Create a new file named app.py:

from flask import Flask, request, jsonify
import joblib
import numpy as np

# Initialize the Flask app
app = Flask(__name__)

# Load the trained model
model = joblib.load('house_price_model.pkl')

# Create an API endpoint/route
@app.route('/predict', methods=['POST'])
def predict():
    # 1. Get the JSON data sent by the user
    data = request.get_json()
    
    # 2. Extract features (SqFt, Beds)
    features = np.array([[data['sqft'], data['beds']]])
    
    # 3. Make prediction
    pred = model.predict(features)
    
    # 4. Return the result as JSON
    return jsonify({"predicted_price": pred[0]})

# Run the server
if __name__ == '__main__':
    app.run(port=5000, debug=True)

7. Testing the API #

8. Cloud Deployment #

• Heroku (Easiest for beginners)

• AWS Elastic Beanstalk or Google Cloud Run

• PythonAnywhere

9. Common Mistakes #

• Version Mismatch: If you train and save the model using Scikit-learn version 1.2 on your laptop, but the cloud server installs Scikit-learn version 1.0, the joblib.load() command will crash. Always ensure your requirements.txt specifies exact versions (e.g., scikit-learn==1.2.0).

• Security: Never blindly unpickle/load a .pkl file you downloaded randomly from the internet. Pickle files can contain malicious code that executes as soon as you load it.

10. Best Practices #

• Save Pipelines, not just Models: As emphasized in Chapter 18, always save the entire Pipeline object. If you only save the Random Forest, your Flask app will have to manually run StandardScaler on the user's JSON data before predicting, which is a recipe for bugs.

11. Exercises #

1. 1. Write the two lines of code required to import joblib and save a model named svmmodel to a file called mysvm.pkl.

1. 2. What format is commonly used to send and receive data over a web API?

12. MCQ Quiz with Answers #

13. Interview Questions #

• Q: Explain the concept of Model Serialization and name the library used for it in Scikit-learn.

• Q: If an iOS app needs to use your Python machine learning model to make predictions, how do you bridge the gap between the two different languages?

14. FAQs #

15. Summary #

16. Next Chapter Recommendation #