TensorFlow Autoencoder

Posted on Sat 03 December 2016 in TensorFlow

In [1]:
"""Tutorial on how to create an autoencoder w/ TensorFlow.

Anrew Wong, Nov 2016
"""
%matplotlib inline
In [2]:
# %% Imports
import tensorflow as tf
import numpy as np
import math
In [3]:
# %% Autocoder definition
def autoencoder(dimensions=[784, 512, 256, 64]):
    """Build a deep autoencoder w/ tied weights.
    
    Parameters
    ----------
    dimensions : list, optional
        The number of neurons for each layer of the autoencoder
        
    Returns
    -------
    x : Tensor
        Input placeholder to the network
    z : Tensor
        Inner-most latent representation
    y : Tensor
        Output reconstruction of the input
    cost : Tensor
        Overall cost to use for training
    """
    # %% input ot the network
    x = tf.placeholder(tf.float32, [None, dimensions[0]], name='x')
    current_input = x
    
    # %% Build the encoder
    encoder = []
    for layer_i, n_output in enumerate(dimensions[1:]):
        n_input = int(current_input.get_shape()[1])
        W = tf.Variable(
            tf.random_normal([n_input, n_output],
                             -1.0 / math.sqrt(n_input),
                             1.0 / math.sqrt(n_input)))
        b = tf.Variable(tf.zeros([n_output]))
        encoder.append(W)
        output = tf.nn.tanh(tf.matmul(current_input, W) + b)
        current_input = output
        
    # %% latent representation
    z = current_input
    encoder.reverse()
    
    # %% Build the decoder using the same weights
    for layer_i, n_output in enumerate(dimensions[:-1][::-1]):
        W = tf.transpose(encoder[layer_i])
        b = tf.Variable(tf.zeros([n_output]))
        output = tf.nn.tanh(tf.matmul(current_input, W) + b)
        current_input = output
        
    # %% now have the reconstruction through the network
    y = current_input
    
    # %% cost function measures pixel-wise difference
    cost = tf.reduce_sum(tf.square(y - x))
    return {'x': x, 'z': z, 'y': y, 'cost': cost}
In [4]:
# %% Basic test
def test_mnist():
    """Test the autoencoder using MNIST."""
    import tensorflow as tf
    from tensorflow.examples.tutorials.mnist import input_data
    import matplotlib.pyplot as plt
    
    # %%
    # load MNIST as before
    mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
    mean_img = np.mean(mnist.train.images, axis=0)
    ae = autoencoder(dimensions=[784, 256, 64])
    
    # %%
    learning_rate = 0.001
    optimizer = tf.train.AdamOptimizer(learning_rate).minimize(ae['cost'])
    
    # %%
    # We create a session to use the graph
    sess = tf.Session()
    sess.run(tf.initialize_all_variables())
    
    # %%
    # Fit all training data
    batch_size = 50
    n_epochs = 10
    for epoch_i in range(n_epochs):
        for batch_i in range(mnist.train.num_examples // batch_size):
            batch_xs, _ = mnist.train.next_batch(batch_size)
            train = np.array([img - mean_img for img in batch_xs])
            sess.run(optimizer, feed_dict={ae['x']: train})
        print(epoch_i, sess.run(ae['cost'], feed_dict={ae['x']: train}))
    
    # %%
    # Plot example reconstructions
    n_examples = 15
    test_xs, _ = mnist.test.next_batch(n_examples)
    test_xs_norm = np.array([img - mean_img for img in test_xs])
    recon = sess.run(ae['y'], feed_dict={ae['x']: test_xs_norm})
    fig, axs = plt.subplots(2, n_examples, figsize=(10, 2))
    for example_i in range(n_examples):
        axs[0][example_i].imshow(
            np.reshape(test_xs[example_i, :], (28, 28)))
        axs[1][example_i].imshow(
            np.reshape([recon[example_i, :] + mean_img], (28, 28)))
    plt.draw()
In [5]:
# %%
if __name__ == '__main__':
    test_mnist()
Extracting MNIST_data/train-images-idx3-ubyte.gz
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz
0 18031.4
1 12303.5
2 1337.15
3 683.05
4 641.797
5 528.142
6 463.697
7 459.439
8 472.742
9 399.869