Neural Network Programming with TensorFlow

Autoencoder class

We initialize the autoencoder defined in class AdditiveGaussianNoiseAutoEncoder by passing following parameters:

• num_input: Number of input samples
• num_hidden: Number of neurons in the hidden layer
• transfer_function=tf.nn.sigmoid: Transfer function
• optimizer = tf.train.AdamOptimizer(): Optimizer
• scale=0.1: Scale for corruption of the image

def __init__(self, num_input, num_hidden,                  transfer_function=tf.nn.sigmoid,                  optimizer=tf.train.AdamOptimizer(),                 scale=0.1):

Assign the passed parameters to the instance variables:

        self.num_input = num_input        self.num_hidden = num_hidden        self.transfer = transfer_function        self.scale = tf.placeholder(tf.float32)        self.training_scale = scale        n_weights = self._initialize_weights()        self.weights ...