model.py

import tensorflow as tf


class MyModel(tf.keras.Model):
    """
    Our own custon MLP model, which inherits from the keras.Model class
      Functions:
        init: constructor of our model
        get_layer: returns list with our layers
        call: performs forward pass of our model
    """

    def __init__(self):
        """
        Constructs our model.
        """

        super(MyModel, self).__init__()
        
        # feature learning
        self.l1 = tf.keras.layers.Conv2D(filters = 12, kernel_size = 5, strides=1,padding="same",activation='relu')
        self.l2 = tf.keras.layers.Conv2D(filters = 12, kernel_size = 3, strides=1,padding="same",activation='relu')
        self.l3 = tf.keras.layers.Dropout(0.2)
        self.l4 = tf.keras.layers.MaxPool2D(pool_size = 2, strides = 2, padding="same")
        self.l5 = tf.keras.layers.Conv2D(filters = 32, kernel_size = 9, strides=1,padding="same", kernel_regularizer="l1_l2",activation='relu')
        self.l6 = tf.keras.layers.GlobalAvgPool2D()
        # classification
        self.l7 = tf.keras.layers.Dense(10, kernel_regularizer="l1_l2", activation='softmax')

    def get_layers(self):
        """
        Returns list with all layers.

          Returns:
            - <list<tensorflow.keras.layer>> list with the models layers
        """
        return [self.l1,self.l2,self.l3,self.l4,self.l5,self.l6,self.l7]


    def call(self, inputs,is_training):
        """
        Performs a forward step in our MLP
          Args:
            inputs: <tensorflow.tensor> our preprocessed input data, we send through our model
            is_training: <bool> variable which determines if dropout is applied
          Results:
            output: <tensorflow.tensor> the predicted output of our input data
        """

        x = self.l1(inputs,training = is_training)
        x = self.l2(x,training = is_training)
        x = self.l3(x,training = is_training)
        x = self.l4(x,training = is_training)
        x = self.l5(x,training = is_training)
        x = self.l6(x,training = is_training)
        output = self.l7(x,training = is_training)

        return output