Save classifier to disk in scikit-learn

Saving and Loading Classifiers in scikit-learn

Introduction

In machine learning, it’s often necessary to save trained models to disk for later use. This is especially beneficial when dealing with complex models that take a significant amount of time to train. scikit-learn provides convenient ways to save and load classifiers, allowing you to reuse your trained models without retraining.

Methods for Saving Classifiers

Scikit-learn offers two primary methods for saving classifiers:

• Pickle: This is the standard Python method for serializing objects. It can be used to save any Python object, including scikit-learn classifiers.
• Joblib: Joblib is specifically designed for saving and loading Python objects, especially those that might contain large NumPy arrays. It is often more efficient than pickle for large models.

Saving with Pickle

Code Example

import pickle
from sklearn.linear_model import LogisticRegression
from sklearn.datasets import load_iris

# Load the iris dataset
iris = load_iris()
X = iris.data
y = iris.target

# Train a logistic regression model
clf = LogisticRegression()
clf.fit(X, y)

# Save the trained model to a file
filename = 'logistic_regression_model.pkl'
pickle.dump(clf, open(filename, 'wb'))

Saving with Joblib

Code Example

import joblib
from sklearn.linear_model import LogisticRegression
from sklearn.datasets import load_iris

# Load the iris dataset
iris = load_iris()
X = iris.data
y = iris.target

# Train a logistic regression model
clf = LogisticRegression()
clf.fit(X, y)

# Save the trained model to a file
filename = 'logistic_regression_model.joblib'
joblib.dump(clf, filename)

Loading Saved Classifiers

Code Example (Pickle)

import pickle

# Load the saved model from the file
filename = 'logistic_regression_model.pkl'
loaded_clf = pickle.load(open(filename, 'rb'))

# Use the loaded model to make predictions
new_data = [[5.1, 3.5, 1.4, 0.2]]
predictions = loaded_clf.predict(new_data)

print(predictions)

Code Example (Joblib)

import joblib

# Load the saved model from the file
filename = 'logistic_regression_model.joblib'
loaded_clf = joblib.load(filename)

# Use the loaded model to make predictions
new_data = [[5.1, 3.5, 1.4, 0.2]]
predictions = loaded_clf.predict(new_data)

print(predictions)

Choosing the Right Method

While both methods work well, here’s a general guide for choosing between pickle and joblib:

Conclusion

Saving and loading trained classifiers in scikit-learn is essential for reusing models and avoiding retraining. The choice between pickle and joblib depends on the size and complexity of the model, but both methods offer reliable and convenient solutions.