fashion-mnist-initialize-parameters.py

def initialize_parameters():
    '''
    Initializes parameters to build a neural network with tensorflow. The shapes are:
                        W1 : [n_hidden_1, n_input]
                        b1 : [n_hidden_1, 1]
                        W2 : [n_hidden_2, n_hidden_1]
                        b2 : [n_hidden_2, 1]
                        W3 : [n_classes, n_hidden_2]
                        b3 : [n_classes, 1]
    
    Returns:
    parameters -- a dictionary of tensors containing W1, b1, W2, b2, W3, b3
    '''
    
    # Set random seed for reproducibility
    tf.set_random_seed(42)
    
    # Initialize weights and biases for each layer
    # First hidden layer
    W1 = tf.get_variable("W1", [n_hidden_1, n_input], initializer=tf.contrib.layers.xavier_initializer(seed=42))
    b1 = tf.get_variable("b1", [n_hidden_1, 1], initializer=tf.zeros_initializer())
    
    # Second hidden layer
    W2 = tf.get_variable("W2", [n_hidden_2, n_hidden_1], initializer=tf.contrib.layers.xavier_initializer(seed=42))
    b2 = tf.get_variable("b2", [n_hidden_2, 1], initializer=tf.zeros_initializer())
    
    # Output layer
    W3 = tf.get_variable("W3", [n_classes, n_hidden_2], initializer=tf.contrib.layers.xavier_initializer(seed=42))
    b3 = tf.get_variable("b3", [n_classes, 1], initializer=tf.zeros_initializer())
    
    # Store initializations as a dictionary of parameters
    parameters = {
        "W1": W1,
        "b1": b1,
        "W2": W2,
        "b2": b2,
        "W3": W3,
        "b3": b3
    }
    
    return parameters