VGG16.py

class VGG16Test(tf.keras.models.Sequential):
  """Construct a Sequential model using transfer learning with 
     VGG16 as the base.

    This class uses some transfer learning and follows the work of Dr. Sivarama Krishnan Rajaraman, et al in 
    using 'pre-trained convolutional neural networks' to detect malaria infections in thin blood smear samples; 
    specifically, the pretrained VGG16 model.

    As is the case with SequentialTest, we're building a Sequential model in which the first layer is a part of 
    the VGG16 model.  Therefore, we can follow the same approach in which this class inherits from the 
    Sequential class and the model is completely built a model in the __init__() method.  

    https://lhncbc.nlm.nih.gov/publication/pub9932
    https://github.com/sivaramakrishnan-rajaraman/CNN-for-malaria-parasite-detection 
    https://androidkt.com/how-to-use-vgg-model-in-tensorflow-keras/
    https://keras.io/api/applications/vgg/#vgg16-function
    https://arxiv.org/abs/1409.1556
    
  """ 
  
  # Model name
  model_name=''
  # Loss function
  loss_object=None
  # optimizer
  optimizer=None

  def __init__(self, 
               model_name: str, 
               input_shape: tuple,
               num_classes: int):
    """
      Args:
        model_name: str
          A name for this model
        input_shape: tuple
          The input_layer's input shape, (image_height, image_width, channels)
        num_classes: int
          The number of classes in this classification task
    """

    # Use the super function to access methods in the superclass, Keras's Model
    super(VGG16Test, self).__init__()  

    # Following Rajaraman, et al build a model starting with DenseNet
    base_model = applications.VGG16(weights='imagenet', 
                                    include_top=False, 
                                    input_shape=input_shape)

    # And  a particular layer
    base_output=base_model.get_layer('block5_conv2').output
    base_model=Model(inputs=base_model.input, outputs=base_output)

    # Freeze the base_model so these layers are not trained
    base_model.trainable = False

    # Add base_model to the class
    self.add(base_model)

    # Add a global average pooling layer
    self.add(tf.keras.layers.GlobalAveragePooling2D())

    # Add a dense layer
    self.add(tf.keras.layers.Dense(units=1024, activation='relu'))

    # Add a dropout layer
    self.add(tf.keras.layers.Dropout(rate=0.5))

    # Add the predictions (dense) layer
    self.add(tf.keras.layers.Dense(units=num_classes, activation='softmax'))  
       
    # Set the loss object
    self.loss_object=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)

    # Set the optimizer
    self.optimizer=tf.keras.optimizers.SGD(lr=0.00001, decay=1e-6, 
                                           momentum=0.9, nesterov=True)