Skip to content
Related Articles

Related Articles

Improve Article
Building an Auto-Encoder using Keras
  • Last Updated : 21 Jun, 2019

Prerequisites: Auto-encoders

This article will demonstrate the process of data compression and the reconstruction of the encoded data by using Machine Learning by first building an Auto-encoder using Keras and then reconstructing the encoded data and visualizing the reconstruction. We would be using the MNIST handwritten digits dataset which is preloaded into the Keras module about which you can read here.

The code is structured as follows: First all the utility functions are defined which are needed at different steps of the building of the Auto-encoder are defined and then each function is called accordingly.

Step 1: Importing the required libraries




import numpy as np
import matplotlib.pyplot as plt
from random import randint
from keras import backend as K
from keras.layers import Input, Dense, Conv2D, MaxPooling2D, UpSampling2D
from keras.models import Model
from keras.datasets import mnist
from keras.callbacks import TensorBoard

Step 2: Defining a utility function to load the data






def load_data():
    # defining the input image size 
    input_image = Input(shape =(28, 28, 1))
      
    # Loading the data and dividing the data into training and testing sets
    (X_train, _), (X_test, _) = mnist.load_data()
      
    # Cleaning and reshaping the data as required by the model
    X_train = X_train.astype('float32') / 255.
    X_train = np.reshape(X_train, (len(X_train), 28, 28, 1))
    X_test = X_test.astype('float32') / 255.
    X_test = np.reshape(X_test, (len(X_test), 28, 28, 1))
      
    return X_train, X_test, input_image

Note: While loading the data, notice that the space where the training labels are loaded are kept empty because the compression process does not involve the output labels

Step 3: Defining a utility function to build the Auto-encoder neural network




def build_network(input_image):
      
    # Building the encoder of the Auto-encoder
    x = Conv2D(16, (3, 3), activation ='relu', padding ='same')(input_image)
    x = MaxPooling2D((2, 2), padding ='same')(x)
    x = Conv2D(8, (3, 3), activation ='relu', padding ='same')(x)
    x = MaxPooling2D((2, 2), padding ='same')(x)
    x = Conv2D(8, (3, 3), activation ='relu', padding ='same')(x)
    encoded_layer = MaxPooling2D((2, 2), padding ='same')(x)
      
    # Building the decoder of the Auto-encoder
    x = Conv2D(8, (3, 3), activation ='relu', padding ='same')(encoded_layer)
    x = UpSampling2D((2, 2))(x)
    x = Conv2D(8, (3, 3), activation ='relu', padding ='same')(x)
    x = UpSampling2D((2, 2))(x)
    x = Conv2D(16, (3, 3), activation ='relu')(x)
    x = UpSampling2D((2, 2))(x)
    decoded_layer = Conv2D(1, (3, 3), activation ='sigmoid', padding ='same')(x)
      
    return decoded_layer

Step 4: Defining a utility function to build and train the Auto-encoder network




def build_auto_encoder_model(X_train, X_test, input_image, decoded_layer):
      
    # Defining the parameters of the Auto-encoder
    autoencoder = Model(input_image, decoded_layer)
    autoencoder.compile(optimizer ='adadelta', loss ='binary_crossentropy')
      
    # Training the Auto-encoder
    autoencoder.fit(X_train, X_train,
                epochs = 15,
                batch_size = 256,
                shuffle = True,
                validation_data =(X_test, X_test),
                callbacks =[TensorBoard(log_dir ='/tmp / autoencoder')])
      
    return autoencoder

Step 5: Defining a utility function to visualize the reconstruction




def visualize(model, X_test):
      
    # Reconstructing the encoded images
    reconstructed_images = model.predict(X_test)
      
    plt.figure(figsize =(20, 4))
    for i in range(1, 11):
          
        # Generating a random to get random results
        rand_num = randint(0, 10001)
      
        # To display the original image
        ax = plt.subplot(2, 10, i)
        plt.imshow(X_test[rand_num].reshape(28, 28))
        plt.gray()
        ax.get_xaxis().set_visible(False)
        ax.get_yaxis().set_visible(False)
  
        # To display the reconstructed image
        ax = plt.subplot(2, 10, i + 10)
        plt.imshow(reconstructed_images[rand_num].reshape(28, 28))
        plt.gray()
        ax.get_xaxis().set_visible(False)
        ax.get_yaxis().set_visible(False)
          
    # Displaying the plot
    plt.show()

Step 6: Calling the utility functions in the appropriate order

a) Loading the data




X_train, X_test, input_image = load_data()

b) Building the network




decoded_layer = build_network(input_image)

c) Building and training the Auto-encoder




auto_encoder_model = build_auto_encoder_model(X_train,
                                             X_test,
                                             input_image,
                                             decoded_layer)

d) Visualizing the reconstruction




visualize(auto_encoder_model, X_test)

 Attention geek! Strengthen your foundations with the Python Programming Foundation Course and learn the basics.  

To begin with, your interview preparations Enhance your Data Structures concepts with the Python DS Course. And to begin with your Machine Learning Journey, join the Machine Learning – Basic Level Course




My Personal Notes arrow_drop_up
Recommended Articles
Page :