Save classifier to disk in scikit-learn in Python

In this article, we will cover saving a Save classifier to disk in scikit-learn using Python.

We always train our models whether they are classifiers, regressors, etc. with the scikit learn library which require a considerable time to train. So we can save our trained models and then retrieve them when required. This saves us a lot of time. Serialization is the process of saving data, whereas Deserialization is the process of restoring it, we will learn to save the classifier models in two ways:

Method 1: Using Pickle

Pickle is a library provided by Python and is the standard way of saving and retrieving files from storage. It first serializes the object model and then saves it to the disk. Later we retrieve it using deserializing. Pickling is a process where a Python object hierarchy is converted into a byte stream. Unpickling is the inverse of the Pickling process where a byte stream is converted into an object hierarchy. 

• dumps() – This function is called to serialize an object hierarchy.
• loads() – This function is called to de-serialize a data stream.

Syntax:

# Saving model

import pickle

pickle.dump(model, open("model_clf_pickle", 'wb'))

# load retrieve

my_model_clf = pickle.load(open("model_clf_pickle", 'rb'))

Example: 

We have the iris dataset on which we trained the K Nearest Neighbor classifier. Then we saved the model using the pickle and later retrieved using the pickle and calculate the score of the classifier.

from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier
import pickle

# load the iris dataset as an example
iris = load_iris()

# store the feature matrix (X) and response vector (y)
X = iris.data
y = iris.target

# splitting X and y into training and testing sets
X_train, X_test, y_train, y_test = train_test_split(
  X, y, test_size=0.4, random_state=1)

# training the model on training set
model_clf = KNeighborsClassifier(n_neighbors=3)
model_clf.fit(X_train, y_train)

# Saving classifier using pickle
pickle.dump(model_clf, open("model_clf_pickle", 'wb'))

# load classifier using pickle
my_model_clf = pickle.load(open("model_clf_pickle", 'rb'))
result_score = my_model_clf.score(X_test,y_test)
print("Score: ",result_score)

Output

Score:  0.9833333333333333

Method 2: Using the joblib library

Joblib is the replacement of a pickle as it is more efficient on objects that carry large NumPy arrays. This is solely created for the purpose of saving the models and retrieving them when required These functions also accept file-like objects instead of filenames.

• joblib.dump is used to serialize an object hierarchy 
• joblib.load is used to deserialize a data stream

Syntax:

# Save model
joblib.dump(model,"model_name.pkl")

# Retrieve model
joblib.load("model_name.pkl")

Example: 

We have the iris dataset on which we trained the K Nearest Neighbor classifier. Then we saved the model using joblib and later retrieved using the joblib. Finally, we calculate the score of the classifier.

from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier
import joblib

# load the iris dataset as an example
iris = load_iris()

# store the feature matrix (X) and response vector
# (y)
X = iris.data
y = iris.target

# splitting X and y into training and testing sets
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.4, random_state=1)

# training the model on training set
model_clf = KNeighborsClassifier(n_neighbors=3)
model_clf.fit(X_train, y_train)

# Saving classifier using joblib
joblib.dump(model_clf, 'model_clf.pkl')

# load classifier using joblib
my_model_clf = joblib.load("model_clf.pkl")
result_score = my_model_clf.score(X_test, y_test)
print("Score: ", result_score)

Output:

Score:  0.9833333333333333