How to Save and Load Machine Learing Models with Pickle

     In machine learning, saving and loading models is important part of the process to ensure  their reusability and deployment. Saving models involves serializing them into a file format that can be easily stored and retrieved. The 'pickle' module provides efficient way to serialize and deserialize machine learning models in Python. 

    In this tutorial, we'll explore how to save and read trained machine learning models using 'pickle'. The tutorial covers:

1. Model training and saving 
2. Model loading and using
   
3. Source code listing
   

    Let's get started.

    

Model training and saving

 

     First we'll create model and train it with a sample dataset. After confirming the prediction performance of model, we can save it into the file. In the code snippet below, we import the necessary libraries and prepare training and test data for the model. Then create an instance of 'SVC' model and train it on prepared data.

 

import pickle

from sklearn.svm import SVC

from sklearn.datasets import load_iris

from sklearn.model_selection import train_test_split

# Iris dataset example

iris = load_iris()

x, y = iris.data, iris.target

# split dataset into train and test

xtrain, xtest, ytrain, ytest = train_test_split(x, y, test_size=0.10)

# create SVC model instance

svc_model = SVC()

# train model

svc_model.fit(xtrain, ytrain)

 

We can test model prediction performance.

 

# check prediction

ypred = svc_model.predict(xtest[0].reshape(1, -1))

print("Predicted class:", ypred)

 

 

 

Predicted class: [2] 

  

To save trained model, we use pickle.dump() function.

 

# define path and model name

file_path = "/Users/user/Desktop/tmp/svc_model_01.pkl"

with open(file_path, "wb") as file:

pickle.dump(svc_model, file)

 

 

Model loading and using

 

    Once we have saved the model, we can easily load it back into memory for further use. The pickle.load() function helps us to load the model object from the file. The loaded model can then be used for making predictions or performing additional analysis tasks. We create new file to use trained model. 

 

import pickle

from sklearn.datasets import load_iris

from sklearn.model_selection import train_test_split

# Iris dataset example

iris = load_iris()

x, y = iris.data, iris.target

# split dataset into train and test

xtrain, xtest, ytrain, ytest = train_test_split(x, y, test_size=0.10)

file_path = "/Users/user/Desktop/tmp/svc_model_01.pkl"

loaded_model = None

# load the model

with open(file_path, 'rb') as file:

loaded_model = pickle.load(file)

if loaded_model is None:

print("Model has not been loaded properly!")

else:

pred = loaded_model.predict(xtest[1].reshape(1, -1))

print("Predicted class:", pred)

  

Predicted class: [2] 

  

 

    In the code snippet above, we loaded required libraries and test data to check the loaded model performance. After successfully loading the model, we predicted the test data.

 

    In this tutorial, we've briefly explored the process of saving and reading machine learning models using the 'pickle' module in Python. The full source code is listed below. 

 

Source code listing

 

# save_model.py

import pickle

from sklearn.svm import SVC

from sklearn.datasets import load_iris

from sklearn.model_selection import train_test_split

# Iris dataset example

iris = load_iris()

x, y = iris.data, iris.target

# split dataset into train and test

xtrain, xtest, ytrain, ytest = train_test_split(x, y, test_size=0.10)

# create SVC model instance

svc_model = SVC()

# train model

svc_model.fit(xtrain, ytrain)

# check prediction

ypred = svc_model.predict(xtest[0].reshape(1, -1))

print("Predicted class:", ypred)

# define saveing path and model name

file_path = "/Users/user/Desktop/tmp/svc_model_01.pkl"

with open(file_path, "wb") as file:

pickle.dump(svc_model, file)

 

 

# load_model.py

import pickle

from sklearn.datasets import load_iris

from sklearn.model_selection import train_test_split

# Iris dataset example

iris = load_iris()

x, y = iris.data, iris.target

# split dataset into train and test

xtrain, xtest, ytrain, ytest = train_test_split(x, y, test_size=0.10)

file_path = "/Users/user/Desktop/tmp/svc_model_01.pkl"

loaded_model = None

# load the model

with open(file_path, 'rb') as file:

loaded_model = pickle.load(file)

if loaded_model is None:

print("Model has not been loaded properly!")

else:

pred = loaded_model.predict(xtest[1].reshape(1, -1))

print("Predicted class:", pred)