In [ ]:

    

HIDDEN2_UNITS = 64

weights1 = weight_variable(NUM_PIXELS, HIDDEN1_UNITS, "weights1")
biases1 = bias_variable(HIDDEN1_UNITS, "biases1")
hidden1 = tf.nn.relu(tf.matmul(x, weights1) + biases1, name="hidden1")

weights2 = weight_variable(HIDDEN1_UNITS, HIDDEN2_UNITS, "weights2")
biases2 = bias_variable(HIDDEN2_UNITS, "biases2")
hidden2 = tf.nn.relu(tf.matmul(hidden1, weights2) + biases2, name="hidden2")

weights3 = weight_variable(HIDDEN2_UNITS, NUM_CLASSES, "weights3")
biases3 = bias_variable(NUM_CLASSES, "biases3")

y = tf.matmul(hidden2, weights3) + biases3

with tf.name_scope("hidden_layer_1"):

    # weights and baises for the first layer
    weights1 = weight_variable(NUM_PIXELS, HIDDEN1_UNITS, "weights1")
    biases1 = bias_variable(HIDDEN1_UNITS, "biases1")

    # activations for the first hidden layer
    hidden1 = tf.nn.relu(tf.matmul(x, weights1) + biases1, name="hidden1")
    
with tf.name_scope("hidden_layer_2"):

    # weights and baises for the second layer
    weights2 = weight_variable(HIDDEN1_UNITS, HIDDEN2_UNITS, "weights2")
    biases2 = bias_variable(HIDDEN2_UNITS, "biases2")

    # activations for the second hidden layer
    hidden2 = tf.nn.relu(tf.matmul(hidden1, weights2) + biases2, name="hidden2")

with tf.name_scope("output_layer"):
    
    # weights and biases for the second layer
    weights3 = weight_variable(HIDDEN2_UNITS, NUM_CLASSES, "weights3")
    biases3 = bias_variable(NUM_CLASSES, "biases3")

    # logits - you can think of these (roughly)
    # as unnormalized probabilities, or the amount of
    # evidence we have that the input image corresponds to
    # each digit
    y = tf.matmul(hidden2, weights3) + biases3