Machine Learning · Chapter 36 of 40
Saving & Loading Models
Save trained models with `joblib` (scikit-learn) or the framework's native format (Keras `.keras`, PyTorch `state_dict`).
Always save the pipeline (preprocessing + model), not just the model.
Example 1 (python)
import joblib
joblib.dump(pipeline, 'model.pkl')
loaded = joblib.load('model.pkl')
print(loaded.predict(X_test[:5]))Output
[0 1 1 0 1]Serialize/deserialize scikit-learn models.
Example 2 (python)
# Keras
# model.save('my_model.keras')
# tf.keras.models.load_model('my_model.keras')Native Keras format.
Key points
- Use joblib for scikit-learn.
- Save the pipeline, not just the model.
- Frameworks have native formats.
- Version your models with your code.
💡 Note: Pickle/joblib files are UNSAFE to load from untrusted sources — they can execute arbitrary code.