SIT744_prac_9.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "SIT744_prac_9.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "machine_shape": "hm"
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "J0PWKlbt6_dm",
        "colab_type": "text"
      },
      "source": [
        "# SIT744 Practical 9: ResNet\n",
        " \n",
        "\n",
        "*Dr Wei Luo*"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0JEtYNgr7JuP",
        "colab_type": "text"
      },
      "source": [
        "<div class=\"alert alert-info\">\n",
        "We suggest that you run this notebook using Google Colab.\n",
        "</div>\n",
        "\n",
        "\n",
        "## Pre-practical readings\n",
        "\n",
        "- [Writing custom layers and models with Keras](https://www.tensorflow.org/guide/keras/custom_layers_and_models)\n",
        "\n",
        "By now you should be familiar with the classical ConvNet architectures. In this practical, you will practice building deep models with more layers using residual units.\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4wqb8r86pD1W",
        "colab_type": "text"
      },
      "source": [
        "\n",
        "## Task 1 Keras ImageDataGenerator\n",
        "\n",
        "We will continue using the *cats vs dogs* dataset in this practical. This time, we will be more serious about data augmentation by using Keras's `ImageDataGenerator`. This allows us to use a variety of predefined types of data augmentation. For this to work, we have to organise training and test images into folders.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7FXupORbRg5C",
        "colab_type": "code",
        "outputId": "712bf0fc-2abc-4a1f-edf7-065df922234f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        }
      },
      "source": [
        "!wget --no-check-certificate \\\n",
        "    https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip \\\n",
        "    -O /tmp/cats_and_dogs_filtered.zip\n",
        "\n",
        "import os\n",
        "import zipfile\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "\n",
        "local_zip = '/tmp/cats_and_dogs_filtered.zip'\n",
        "zip_ref = zipfile.ZipFile(local_zip, 'r')\n",
        "zip_ref.extractall('/tmp')\n",
        "zip_ref.close()\n",
        "\n",
        "\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-05-08 02:34:18--  https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip\n",
            "Resolving storage.googleapis.com (storage.googleapis.com)... 172.217.204.128, 2607:f8b0:400c:c07::80\n",
            "Connecting to storage.googleapis.com (storage.googleapis.com)|172.217.204.128|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 68606236 (65M) [application/zip]\n",
            "Saving to: ‘/tmp/cats_and_dogs_filtered.zip’\n",
            "\n",
            "/tmp/cats_and_dogs_ 100%[===================>]  65.43M   108MB/s    in 0.6s    \n",
            "\n",
            "2020-05-08 02:34:19 (108 MB/s) - ‘/tmp/cats_and_dogs_filtered.zip’ saved [68606236/68606236]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w1bzLlk-afVj",
        "colab_type": "text"
      },
      "source": [
        "Take a look into the folder `/tmp/cats_and_dogs_filtered` and see how the images are organised."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KyMjnV8Taenb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "base_dir = '/tmp/cats_and_dogs_filtered'\n",
        "train_dir = os.path.join(base_dir, 'train')\n",
        "validation_dir = os.path.join(base_dir, 'validation')\n",
        "\n",
        "# Directory with our training cat pictures\n",
        "train_cats_dir = os.path.join(train_dir, 'cats')\n",
        "\n",
        "# Directory with our training dog pictures\n",
        "train_dogs_dir = os.path.join(train_dir, 'dogs')\n",
        "\n",
        "# Directory with our validation cat pictures\n",
        "validation_cats_dir = os.path.join(validation_dir, 'cats')\n",
        "\n",
        "# Directory with our validation dog pictures\n",
        "validation_dogs_dir = os.path.join(validation_dir, 'dogs')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "e6J5pJG8aJV4",
        "colab_type": "text"
      },
      "source": [
        "`ImageDataGenerator` contains a collection of data augmentation operations. It works with the function `flow_from_directory` to create an input pipeline."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fnZDg0voR3IZ",
        "colab_type": "code",
        "outputId": "3f41547d-f689-46fc-9a17-d318b0fd9111",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "\n",
        "# All images will be rescaled by 1./255\n",
        "train_datagen = ImageDataGenerator(\n",
        "    rescale=1./255,\n",
        "    rotation_range=40,\n",
        "    width_shift_range=0.2,\n",
        "    height_shift_range=0.2,\n",
        "    shear_range=0.2,\n",
        "    zoom_range=0.2,\n",
        "    horizontal_flip=True,\n",
        "    fill_mode='nearest')\n",
        "\n",
        "test_datagen = ImageDataGenerator(rescale=1./255)\n",
        "\n",
        "IMAGE_SIZE = 100\n",
        "BATCH_SIZE = 32\n",
        "\n",
        "# Flow training images in batches of 20 using train_datagen generator\n",
        "train_generator = train_datagen.flow_from_directory(\n",
        "        train_dir,  # This is the source directory for training images\n",
        "        target_size=(IMAGE_SIZE, IMAGE_SIZE),  # All images will be resized to 150x150\n",
        "        batch_size=BATCH_SIZE,\n",
        "        # Since we use binary_crossentropy loss, we need binary labels\n",
        "        class_mode='binary')\n",
        "\n",
        "# Flow validation images in batches of 20 using test_datagen generator\n",
        "validation_generator = test_datagen.flow_from_directory(\n",
        "        validation_dir,\n",
        "        target_size=(IMAGE_SIZE, IMAGE_SIZE),\n",
        "        batch_size=BATCH_SIZE,\n",
        "        class_mode='binary')\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Found 2000 images belonging to 2 classes.\n",
            "Found 1000 images belonging to 2 classes.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Zkjlk8pmpK_k",
        "colab_type": "text"
      },
      "source": [
        "**exercise** Read the documentation for `ImageDataGenerator` and understand what data-augmentation operations are defined. Experiment with different configurations and see what each data augmentation does.\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "d47yNCI0O2k1",
        "colab_type": "text"
      },
      "source": [
        "## Task 2 ResNet\n",
        "\n",
        "We have seen VGG in the last two weeks. VGG has a very classical ConvNet architecture. Compared with VGG, ResNets can go much deeper, achieve a much higher classification accuracy, and is easier to train. Moreover, ResNets have a surprisingly simple architecture. At the centre is a repetition of multiple **residual units**.\n",
        "\n",
        "A ResNet contains two types of residual units, depending on whether subsampling happens (stride>1).\n",
        "![Residual units (src: d2l)](https://d2l.ai/_images/resnet-block.svg)\n",
        "\n",
        "Subsampling happens at the first of the two convolutional layers. When the first convolutional layer has stride 1, the skip connection is just a pass-through. Otherwise, the skip connection is replaced with a 1x1 convolution layer of the matching stride setting (Note that this is no longer an identity mapping).\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9VsOm32DdDIj",
        "colab_type": "text"
      },
      "source": [
        "### Write custom layers\n",
        "\n",
        "Here we define a ResidualUnit layer subclassing `keras.layers.Layer`. A layer defines a transformation from the input to the output, parametrised by the layer weights."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2EeyhUODXKZ3",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from functools import partial\n",
        "\n",
        "DefaultConv2D = partial(keras.layers.Conv2D, kernel_size=3, strides=1,\n",
        "                        padding=\"SAME\", \n",
        "                        use_bias=False,\n",
        "                        kernel_regularizer=keras.regularizers.l2(0.01)\n",
        "                        )\n",
        "\n",
        "class ResidualUnit(keras.layers.Layer):\n",
        "    def __init__(self, filters, strides=1, activation=\"relu\", **kwargs):\n",
        "        super().__init__(**kwargs)\n",
        "        self.activation = keras.activations.get(activation)\n",
        "        self.main_layers = [\n",
        "            DefaultConv2D(filters, strides=strides),\n",
        "            keras.layers.BatchNormalization(),\n",
        "            self.activation,\n",
        "            DefaultConv2D(filters),\n",
        "            keras.layers.BatchNormalization()]\n",
        "        self.skip_layers = []\n",
        "        if strides > 1:\n",
        "            self.skip_layers = [\n",
        "                DefaultConv2D(filters, kernel_size=1, strides=strides),\n",
        "                keras.layers.BatchNormalization()]\n",
        "\n",
        "    def call(self, inputs):\n",
        "        Z = inputs\n",
        "        for layer in self.main_layers:\n",
        "            Z = layer(Z)\n",
        "        skip_Z = inputs\n",
        "        for layer in self.skip_layers:\n",
        "            skip_Z = layer(skip_Z)\n",
        "        return self.activation(Z + skip_Z)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aBEpF5H9dJRn",
        "colab_type": "text"
      },
      "source": [
        "### Write the custom model\n",
        "\n",
        "Now we have a custom layer for the residual unit. We will build a relatively shallow ResNet."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k9suim5fXM83",
        "colab_type": "code",
        "outputId": "21acd735-2591-4d84-f720-cdb5995a38dd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 425
        }
      },
      "source": [
        "model = keras.models.Sequential()\n",
        "\n",
        "prev_filters = 64\n",
        "c = [prev_filters, 2*prev_filters, 4*prev_filters, 4*prev_filters]\n",
        "model.add(DefaultConv2D(c[0], \n",
        "                        input_shape=[IMAGE_SIZE, IMAGE_SIZE, 3]))\n",
        "model.add(ResidualUnit(c[0], strides=1))\n",
        "model.add(ResidualUnit(c[1], strides=2))\n",
        "model.add(ResidualUnit(c[2], strides=2))\n",
        "model.add(ResidualUnit(c[3], strides=2))\n",
        "model.add(keras.layers.MaxPool2D(pool_size=(4, 4)))\n",
        "model.add(keras.layers.Flatten())\n",
        "model.add(keras.layers.Dense(1, activation=\"sigmoid\"))\n",
        "\n",
        "\n",
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d (Conv2D)              (None, 100, 100, 64)      1728      \n",
            "_________________________________________________________________\n",
            "residual_unit (ResidualUnit) (None, 100, 100, 64)      74240     \n",
            "_________________________________________________________________\n",
            "residual_unit_1 (ResidualUni (None, 50, 50, 128)       230912    \n",
            "_________________________________________________________________\n",
            "residual_unit_2 (ResidualUni (None, 25, 25, 256)       920576    \n",
            "_________________________________________________________________\n",
            "residual_unit_3 (ResidualUni (None, 13, 13, 256)       1248256   \n",
            "_________________________________________________________________\n",
            "max_pooling2d (MaxPooling2D) (None, 3, 3, 256)         0         \n",
            "_________________________________________________________________\n",
            "flatten (Flatten)            (None, 2304)              0         \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 1)                 2305      \n",
            "=================================================================\n",
            "Total params: 2,478,017\n",
            "Trainable params: 2,473,921\n",
            "Non-trainable params: 4,096\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "i_Xew-RYdnCn",
        "colab_type": "text"
      },
      "source": [
        "### Train the model\n",
        "\n",
        "Training will be just like for any Keras model. Because our input data is from a Keras input pipeline (generator), we have to specify how many steps per epoch.  "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M5vihjXRMOus",
        "colab_type": "code",
        "outputId": "916b1963-b2a1-4bac-b094-312ba979a2f8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "%%time \n",
        "\n",
        "from tensorflow.keras import optimizers\n",
        "\n",
        "model.compile(loss='binary_crossentropy',\n",
        "              optimizer=optimizers.RMSprop(lr=1e-4),\n",
        "              metrics=['acc'])\n",
        "\n",
        "\n",
        "history = model.fit(\n",
        "    train_generator,\n",
        "    steps_per_epoch=100,\n",
        "    epochs=30,\n",
        "    validation_data=validation_generator,\n",
        "    validation_steps=50)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/30\n",
            "100/100 [==============================] - 33s 331ms/step - loss: 18.4898 - acc: 0.5685 - val_loss: 18.0816 - val_acc: 0.5025\n",
            "Epoch 2/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 17.7326 - acc: 0.6285 - val_loss: 17.6064 - val_acc: 0.4929\n",
            "Epoch 3/30\n",
            "100/100 [==============================] - 32s 317ms/step - loss: 16.9715 - acc: 0.6454 - val_loss: 16.5989 - val_acc: 0.5495\n",
            "Epoch 4/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 16.1760 - acc: 0.6515 - val_loss: 16.1210 - val_acc: 0.5161\n",
            "Epoch 5/30\n",
            "100/100 [==============================] - 32s 316ms/step - loss: 15.2930 - acc: 0.6680 - val_loss: 14.8053 - val_acc: 0.6707\n",
            "Epoch 6/30\n",
            "100/100 [==============================] - 31s 313ms/step - loss: 14.3371 - acc: 0.6834 - val_loss: 13.8564 - val_acc: 0.6888\n",
            "Epoch 7/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 13.3648 - acc: 0.6948 - val_loss: 12.8956 - val_acc: 0.6965\n",
            "Epoch 8/30\n",
            "100/100 [==============================] - 32s 316ms/step - loss: 12.3963 - acc: 0.6976 - val_loss: 11.8987 - val_acc: 0.7081\n",
            "Epoch 9/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 11.4369 - acc: 0.7008 - val_loss: 11.0315 - val_acc: 0.6759\n",
            "Epoch 10/30\n",
            "100/100 [==============================] - 31s 315ms/step - loss: 10.5346 - acc: 0.7126 - val_loss: 10.0586 - val_acc: 0.7500\n",
            "Epoch 11/30\n",
            "100/100 [==============================] - 31s 315ms/step - loss: 9.6636 - acc: 0.7377 - val_loss: 9.2446 - val_acc: 0.7506\n",
            "Epoch 12/30\n",
            "100/100 [==============================] - 31s 315ms/step - loss: 8.8999 - acc: 0.7244 - val_loss: 9.0928 - val_acc: 0.5258\n",
            "Epoch 13/30\n",
            "100/100 [==============================] - 31s 315ms/step - loss: 8.1961 - acc: 0.7239 - val_loss: 7.8472 - val_acc: 0.7335\n",
            "Epoch 14/30\n",
            "100/100 [==============================] - 31s 313ms/step - loss: 7.5415 - acc: 0.7323 - val_loss: 7.3051 - val_acc: 0.6952\n",
            "Epoch 15/30\n",
            "100/100 [==============================] - 32s 316ms/step - loss: 6.9642 - acc: 0.7396 - val_loss: 6.7008 - val_acc: 0.7297\n",
            "Epoch 16/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 6.4354 - acc: 0.7431 - val_loss: 6.1989 - val_acc: 0.7390\n",
            "Epoch 17/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 5.9671 - acc: 0.7469 - val_loss: 6.1822 - val_acc: 0.5767\n",
            "Epoch 18/30\n",
            "100/100 [==============================] - 32s 316ms/step - loss: 5.5485 - acc: 0.7468 - val_loss: 5.3797 - val_acc: 0.7221\n",
            "Epoch 19/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 5.1759 - acc: 0.7449 - val_loss: 4.9947 - val_acc: 0.7384\n",
            "Epoch 20/30\n",
            "100/100 [==============================] - 32s 318ms/step - loss: 4.7944 - acc: 0.7692 - val_loss: 4.6520 - val_acc: 0.7462\n",
            "Epoch 21/30\n",
            "100/100 [==============================] - 31s 315ms/step - loss: 4.5142 - acc: 0.7487 - val_loss: 4.3968 - val_acc: 0.7307\n",
            "Epoch 22/30\n",
            "100/100 [==============================] - 32s 316ms/step - loss: 4.2278 - acc: 0.7491 - val_loss: 4.1517 - val_acc: 0.7184\n",
            "Epoch 23/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 3.9501 - acc: 0.7652 - val_loss: 3.8521 - val_acc: 0.7411\n",
            "Epoch 24/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 3.7398 - acc: 0.7579 - val_loss: 3.6442 - val_acc: 0.7461\n",
            "Epoch 25/30\n",
            "100/100 [==============================] - 32s 315ms/step - loss: 3.5226 - acc: 0.7682 - val_loss: 3.4489 - val_acc: 0.7703\n",
            "Epoch 26/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 3.3098 - acc: 0.7775 - val_loss: 3.3650 - val_acc: 0.7165\n",
            "Epoch 27/30\n",
            "100/100 [==============================] - 31s 315ms/step - loss: 3.1566 - acc: 0.7736 - val_loss: 3.1600 - val_acc: 0.7300\n",
            "Epoch 28/30\n",
            "100/100 [==============================] - 31s 315ms/step - loss: 3.0021 - acc: 0.7768 - val_loss: 2.9451 - val_acc: 0.7589\n",
            "Epoch 29/30\n",
            "100/100 [==============================] - 31s 314ms/step - loss: 2.8331 - acc: 0.7882 - val_loss: 2.8532 - val_acc: 0.7397\n",
            "Epoch 30/30\n",
            "100/100 [==============================] - 32s 315ms/step - loss: 2.7157 - acc: 0.7710 - val_loss: 2.7677 - val_acc: 0.7049\n",
            "CPU times: user 18min 59s, sys: 3min 9s, total: 22min 8s\n",
            "Wall time: 15min 58s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-hvlD8HEp1lo",
        "colab_type": "code",
        "outputId": "043cb0d7-5e71-44d1-a706-a1c6afe1dda9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 573
        }
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "# Plot training & validation accuracy values\n",
        "plt.plot(history.history['acc'])\n",
        "plt.plot(history.history['val_acc'])\n",
        "plt.title('Model accuracy')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Test'], loc='upper left')\n",
        "plt.show()\n",
        "\n",
        "# Plot training & validation loss values\n",
        "plt.plot(history.history['loss'])\n",
        "plt.plot(history.history['val_loss'])\n",
        "plt.title('Model loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Test'], loc='upper left')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lP_UX2Xad1AJ",
        "colab_type": "text"
      },
      "source": [
        "**exercise** Try a different batch size. See if it affects the model performance and the training speed."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Er9FnWFmSHx3",
        "colab_type": "text"
      },
      "source": [
        "## Task 3 Depth Separable Convolution\n",
        "\n",
        "In the lecture, we see that Depth Separable Convolution can be used to reduce the number of training parameters. We will explore the potential benefit of using depth separable convolution layers in a residual unit."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PDfHpuj-SMFU",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from functools import partial\n",
        "\n",
        "DefaultConv2D = partial(keras.layers.Conv2D, kernel_size=3, strides=1,\n",
        "                        padding=\"SAME\", \n",
        "                        use_bias=False,\n",
        "                        kernel_regularizer=keras.regularizers.l2(0.01)\n",
        "                        )\n",
        "\n",
        "DefaultSeparableConv2D = partial(keras.layers.SeparableConv2D, kernel_size=3, strides=1,\n",
        "                                 padding=\"SAME\", \n",
        "                                 use_bias=False,\n",
        "                                 kernel_regularizer=keras.regularizers.l2(0.01))\n",
        "\n",
        "class DepthwiseResidualUnit(keras.layers.Layer):\n",
        "    def __init__(self, filters, strides=1, activation=\"relu\", **kwargs):\n",
        "        super().__init__(**kwargs)\n",
        "        self.activation = keras.activations.get(activation)\n",
        "        self.main_layers = [\n",
        "            DefaultSeparableConv2D(filters, strides=strides),\n",
        "            keras.layers.BatchNormalization(),\n",
        "            self.activation,\n",
        "            DefaultSeparableConv2D(filters),\n",
        "            keras.layers.BatchNormalization()]\n",
        "        self.skip_layers = []\n",
        "        if strides > 1:\n",
        "            self.skip_layers = [\n",
        "                DefaultConv2D(filters, kernel_size=1, strides=strides),\n",
        "                keras.layers.BatchNormalization()]\n",
        "\n",
        "    def call(self, inputs):\n",
        "        Z = inputs\n",
        "        for layer in self.main_layers:\n",
        "            Z = layer(Z)\n",
        "        skip_Z = inputs\n",
        "        for layer in self.skip_layers:\n",
        "            skip_Z = layer(skip_Z)\n",
        "        return self.activation(Z + skip_Z)\n",
        "\n",
        "model = keras.models.Sequential()\n",
        "\n",
        "prev_filters = 64\n",
        "c = [prev_filters, 2*prev_filters, 4*prev_filters, 4*prev_filters]\n",
        "model.add(DefaultConv2D(c[0], \n",
        "                        input_shape=[IMAGE_SIZE, IMAGE_SIZE, 3]))\n",
        "model.add(DepthwiseResidualUnit(c[0], strides=1))\n",
        "model.add(DepthwiseResidualUnit(c[1], strides=2))\n",
        "model.add(DepthwiseResidualUnit(c[2], strides=2))\n",
        "model.add(DepthwiseResidualUnit(c[3], strides=2))\n",
        "model.add(keras.layers.MaxPool2D(pool_size=(4, 4)))\n",
        "model.add(keras.layers.Flatten())\n",
        "model.add(keras.layers.Dense(1, activation=\"sigmoid\"))\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z3BGKdcpYazF",
        "colab_type": "code",
        "outputId": "18d11331-03de-4772-bf1f-0cf1d01ec6c6",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 425
        }
      },
      "source": [
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_15\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d_147 (Conv2D)          (None, 100, 100, 64)      1728      \n",
            "_________________________________________________________________\n",
            "depthwise_residual_unit_3 (D (None, 100, 100, 64)      9856      \n",
            "_________________________________________________________________\n",
            "depthwise_residual_unit_4 (D (None, 50, 50, 128)       36032     \n",
            "_________________________________________________________________\n",
            "depthwise_residual_unit_5 (D (None, 25, 25, 256)       137600    \n",
            "_________________________________________________________________\n",
            "depthwise_residual_unit_6 (D (None, 13, 13, 256)       204288    \n",
            "_________________________________________________________________\n",
            "max_pooling2d_12 (MaxPooling (None, 3, 3, 256)         0         \n",
            "_________________________________________________________________\n",
            "flatten_12 (Flatten)         (None, 2304)              0         \n",
            "_________________________________________________________________\n",
            "dense_12 (Dense)             (None, 1)                 2305      \n",
            "=================================================================\n",
            "Total params: 391,809\n",
            "Trainable params: 387,713\n",
            "Non-trainable params: 4,096\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CqHcbm0mY9lg",
        "colab_type": "text"
      },
      "source": [
        "**question** How does the number of parameters change with the introduction of separable convolution layers?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oEs4y243WTk8",
        "colab_type": "code",
        "outputId": "beb03bf7-31a6-4e5a-e256-8f16b81c29fc",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "%%time \n",
        "\n",
        "from tensorflow.keras import optimizers\n",
        "\n",
        "model.compile(loss='binary_crossentropy',\n",
        "              optimizer=optimizers.RMSprop(lr=1e-4),\n",
        "              metrics=['acc'])\n",
        "\n",
        "\n",
        "history = model.fit(\n",
        "    train_generator,\n",
        "    steps_per_epoch=100,\n",
        "    epochs=30,\n",
        "    validation_data=validation_generator,\n",
        "    validation_steps=50)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/30\n",
            "100/100 [==============================] - 35s 352ms/step - loss: 5.8376 - acc: 0.5792 - val_loss: 5.6987 - val_acc: 0.4961\n",
            "Epoch 2/30\n",
            "100/100 [==============================] - 34s 340ms/step - loss: 5.5966 - acc: 0.6212 - val_loss: 5.5073 - val_acc: 0.5019\n",
            "Epoch 3/30\n",
            "100/100 [==============================] - 34s 344ms/step - loss: 5.3827 - acc: 0.6367 - val_loss: 5.3182 - val_acc: 0.4949\n",
            "Epoch 4/30\n",
            "100/100 [==============================] - 34s 343ms/step - loss: 5.1690 - acc: 0.6633 - val_loss: 5.1385 - val_acc: 0.4942\n",
            "Epoch 5/30\n",
            "100/100 [==============================] - 34s 345ms/step - loss: 4.9772 - acc: 0.6755 - val_loss: 5.0312 - val_acc: 0.5057\n",
            "Epoch 6/30\n",
            "100/100 [==============================] - 34s 344ms/step - loss: 4.8043 - acc: 0.6742 - val_loss: 4.8503 - val_acc: 0.5251\n",
            "Epoch 7/30\n",
            "100/100 [==============================] - 34s 344ms/step - loss: 4.6375 - acc: 0.6693 - val_loss: 4.5603 - val_acc: 0.6662\n",
            "Epoch 8/30\n",
            "100/100 [==============================] - 34s 341ms/step - loss: 4.4949 - acc: 0.6749 - val_loss: 4.4242 - val_acc: 0.6501\n",
            "Epoch 9/30\n",
            "100/100 [==============================] - 34s 340ms/step - loss: 4.3112 - acc: 0.6928 - val_loss: 4.2616 - val_acc: 0.6675\n",
            "Epoch 10/30\n",
            "100/100 [==============================] - 35s 346ms/step - loss: 4.1529 - acc: 0.7172 - val_loss: 4.1295 - val_acc: 0.6732\n",
            "Epoch 11/30\n",
            "100/100 [==============================] - 34s 343ms/step - loss: 4.0376 - acc: 0.7011 - val_loss: 4.0277 - val_acc: 0.6675\n",
            "Epoch 12/30\n",
            "100/100 [==============================] - 35s 348ms/step - loss: 3.8851 - acc: 0.7133 - val_loss: 3.8678 - val_acc: 0.6789\n",
            "Epoch 13/30\n",
            "100/100 [==============================] - 34s 337ms/step - loss: 3.7754 - acc: 0.7083 - val_loss: 3.7775 - val_acc: 0.6695\n",
            "Epoch 14/30\n",
            "100/100 [==============================] - 34s 343ms/step - loss: 3.6502 - acc: 0.7136 - val_loss: 3.6203 - val_acc: 0.6849\n",
            "Epoch 15/30\n",
            "100/100 [==============================] - 34s 344ms/step - loss: 3.5066 - acc: 0.7270 - val_loss: 3.4959 - val_acc: 0.7107\n",
            "Epoch 16/30\n",
            "100/100 [==============================] - 34s 343ms/step - loss: 3.4129 - acc: 0.7077 - val_loss: 3.4357 - val_acc: 0.6849\n",
            "Epoch 17/30\n",
            "100/100 [==============================] - 35s 348ms/step - loss: 3.3005 - acc: 0.7252 - val_loss: 3.2904 - val_acc: 0.7024\n",
            "Epoch 18/30\n",
            "100/100 [==============================] - 34s 340ms/step - loss: 3.1942 - acc: 0.7270 - val_loss: 3.1627 - val_acc: 0.7120\n",
            "Epoch 19/30\n",
            "100/100 [==============================] - 34s 339ms/step - loss: 3.0952 - acc: 0.7428 - val_loss: 3.1015 - val_acc: 0.7113\n",
            "Epoch 20/30\n",
            "100/100 [==============================] - 34s 338ms/step - loss: 3.0024 - acc: 0.7314 - val_loss: 3.0144 - val_acc: 0.7037\n",
            "Epoch 21/30\n",
            "100/100 [==============================] - 34s 344ms/step - loss: 2.9137 - acc: 0.7374 - val_loss: 2.9409 - val_acc: 0.6894\n",
            "Epoch 22/30\n",
            "100/100 [==============================] - 35s 345ms/step - loss: 2.8144 - acc: 0.7487 - val_loss: 2.8084 - val_acc: 0.7151\n",
            "Epoch 23/30\n",
            "100/100 [==============================] - 34s 341ms/step - loss: 2.7496 - acc: 0.7367 - val_loss: 2.7916 - val_acc: 0.7088\n",
            "Epoch 24/30\n",
            "100/100 [==============================] - 34s 341ms/step - loss: 2.6700 - acc: 0.7412 - val_loss: 2.7536 - val_acc: 0.6753\n",
            "Epoch 25/30\n",
            "100/100 [==============================] - 34s 342ms/step - loss: 2.5734 - acc: 0.7680 - val_loss: 2.6046 - val_acc: 0.7183\n",
            "Epoch 26/30\n",
            "100/100 [==============================] - 35s 350ms/step - loss: 2.5013 - acc: 0.7582 - val_loss: 2.5050 - val_acc: 0.7313\n",
            "Epoch 27/30\n",
            "100/100 [==============================] - 35s 347ms/step - loss: 2.4342 - acc: 0.7532 - val_loss: 2.4400 - val_acc: 0.7341\n",
            "Epoch 28/30\n",
            "100/100 [==============================] - 34s 341ms/step - loss: 2.3826 - acc: 0.7582 - val_loss: 2.4019 - val_acc: 0.7210\n",
            "Epoch 29/30\n",
            "100/100 [==============================] - 35s 347ms/step - loss: 2.3107 - acc: 0.7588 - val_loss: 2.3022 - val_acc: 0.7436\n",
            "Epoch 30/30\n",
            "100/100 [==============================] - 34s 338ms/step - loss: 2.2521 - acc: 0.7563 - val_loss: 2.2552 - val_acc: 0.7348\n",
            "CPU times: user 19min 34s, sys: 4min 57s, total: 24min 31s\n",
            "Wall time: 17min 22s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b38JApMgWUp5",
        "colab_type": "code",
        "outputId": "8945e3af-3e0c-4e71-9bb0-a4d3283994e1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 573
        }
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "# Plot training & validation accuracy values\n",
        "plt.plot(history.history['acc'])\n",
        "plt.plot(history.history['val_acc'])\n",
        "plt.title('Model accuracy')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Test'], loc='upper left')\n",
        "plt.show()\n",
        "\n",
        "# Plot training & validation loss values\n",
        "plt.plot(history.history['loss'])\n",
        "plt.plot(history.history['val_loss'])\n",
        "plt.title('Model loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Test'], loc='upper left')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kYnsCnfwYdrO",
        "colab_type": "text"
      },
      "source": [
        "**question** Do you see any performance improvement by using separable convolution layers? How about the difference in the training time?\n",
        "\n",
        "**exercise** Continue the training of the above two models until they converge. What validation accuracy can you achieve?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mjrOdskfuGYx",
        "colab_type": "text"
      },
      "source": [
        "## Additional resources\n",
        "\n",
        "- David Page has authored a wonderful series of articles on [practical tips for training ResNets.](https://myrtle.ai/how-to-train-your-resnet/)"
      ]
    }
  ]
}