Saving a machine learning Model

In machine learning, while working with scikit learn library, we need to save the trained models in a file and restore them in order to reuse them to compare the model with other models, and to test the model on new data. The saving of data is called Serialization, while restoring the data is called Deserialization.

Also, we deal with different types and sizes of data. Some datasets are easily trained i.e- they take less time to train but the datasets whose size is large (more than 1GB) can take a very large time to train on a local machine even with GPU. When we need the same trained data in some different project or later sometime, to avoid the wastage of the training time, store the trained model so that it can be used anytime in the future. 

There are two ways we can save a model in scikit learn:

Way 1: Pickle string: 

The pickle module implements a fundamental, but powerful algorithm for serializing and de-serializing a Python object structure. 

Pickle model provides the following functions –

• pickle.dump to serialize an object hierarchy, you simply use dump(). 
• pickle.load to deserialize a data stream, you call the loads() function.

Example: Let’s apply K Nearest Neighbor on the iris dataset and then save the model. 

Python3

from sklearn.neighbors import KNeighborsClassifier as KNN
import numpy as np

# Load dataset
from sklearn.datasets import load_iris
iris = load_iris()

X = iris.data
y = iris.target

# Split dataset into train and test
X_train, X_test, y_train, y_test = \
    train_test_split(X, y, test_size=0.3,
                     random_state=2018)

# import KNeighborsClassifier model
knn = KNN(n_neighbors=3)

# train model
knn.fit(X_train, y_train)

Output:

Save a model to string using pickle:

Python3

import pickle

# Save the trained model as a pickle string.
saved_model = pickle.dumps(knn)

# Load the pickled model
knn_from_pickle = pickle.loads(saved_model)

# Use the loaded pickled model to make predictions
knn_from_pickle.predict(X_test)

Output:

Way 2: Pickled model as a file using joblib: 

Joblib is the replacement of pickle as it is more efficient on objects that carry large numpy arrays. These functions also accept file-like object instead of filenames. 

joblib.dump to serialize an object hierarchy 
joblib.load to deserialize a data stream
from joblib import parallel, delayed

Save to pickled file using joblib 

Python3

from joblib import Parallel, delayed
import joblib


# Save the model as a pickle in a file
joblib.dump(knn, 'filename.pkl')

# Load the model from the file
knn_from_joblib = joblib.load('filename.pkl')

# Use the loaded model to make predictions
knn_from_joblib.predict(X_test)

Output: