AI & Machine Learning · Quick Reference

lesson-1.4 Cheat Sheet

AI & Machine Learning

In one line: Both serialize Python objects to disk, but joblib is optimized for objects containing large NumPy arrays — exactly what most trained sklearn models are: weight matrices, support...

Key Ideas

1pickle vs joblib. Both serialize Python objects to disk, but joblib is optimized for objects containing large NumPy arrays — exactly what most trained sklearn models are: weight matrice...

2Save the whole pipeline, not just the model. If you save only the classifier and discard the scaler/encoder, you have a model that expects already-preprocessed input — but no record of how to preprocess new input...

3Version warnings and what breaks in production. Scikit-learn doesn't guarantee that a model pickled with one version loads cleanly in another. Internal object representations change between releases, especially majo...

Code Examples

import joblib

joblib.dump(pipe, 'model.joblib')
loaded_pipe = joblib.load('model.joblib')
loaded_pipe.predict(X_new)

import sklearn, joblib

metadata = {'sklearn_version': sklearn.__version__}
joblib.dump({'pipeline': pipe, 'meta': metadata}, 'model.joblib')

bundle = joblib.load('model.joblib')
if bundle['meta']['sklearn_version'] != sklearn.__version__:
    print...