Mastering Machine Learning Algorithms

An example of a variational autoencoder with TensorFlow

Let's continue working with the Fashion MNIST dataset to build a variational autoencoder. As explained, the output of the encoder is now split into two components: the mean and covariance vectors (both with dimensions equal to (width · height)) and the decoder input is obtained by sampling from a normal distribution and projecting the code components. The complete Graph is as follows:

import tensorflow as tfgraph = tf.Graph()with graph.as_default():    input_images = tf.placeholder(tf.float32, shape=(batch_size, width, height, 1))        # Encoder    conv_0 = tf.layers.conv2d(inputs=input_images,                              filters=32,                              kernel_size=(3, 3),                              strides=(2, 2),                              activation=tf.nn.relu,                              padding='same')     conv_1 = tf.layers.conv2d(inputs=conv_0, ...