transfer_learning_using_vgg16_base_model_v02.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "colab": {
      "name": "transfer-learning-using-vgg16-base-model-v02.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1Zv6fg6glzWA"
      },
      "source": [
        "## Pneumonia classificaton using Transfer Learning - Base Model\n",
        "\n",
        "This will be the **base model** using VGG16 pretrained model.\n",
        "\n",
        "The base model will not have any data augmentation.\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rPpXAldslzWC"
      },
      "source": [
        "## Import Packages"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:20.717989Z",
          "iopub.status.busy": "2021-08-14T21:47:20.717552Z",
          "iopub.status.idle": "2021-08-14T21:47:26.285946Z",
          "shell.execute_reply": "2021-08-14T21:47:26.284286Z",
          "shell.execute_reply.started": "2021-08-14T21:47:20.717907Z"
        },
        "id": "pPme4T49lzWD",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "64a27631-8c91-4002-b1df-6eb1009bd0db"
      },
      "source": [
        "# import necessary packages\n",
        "import os\n",
        "import warnings\n",
        "import math\n",
        "import pandas as pd\n",
        "import numpy as np\n",
        "from numpy.random import seed\n",
        "from glob import glob\n",
        "\n",
        "# for plotting\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "from datetime import datetime\n",
        "\n",
        "# for deep learning\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.models import (\n",
        "    save_model, \n",
        "    load_model\n",
        ")\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "from tensorflow.keras.utils import plot_model\n",
        "from tensorflow.keras.applications.vgg16 import VGG16\n",
        "from tensorflow.keras.optimizers import Adam\n",
        "from tensorflow.keras.metrics import Accuracy\n",
        "from tensorflow.keras.losses import binary_crossentropy\n",
        "from tensorflow.keras import Sequential\n",
        "from tensorflow.keras.layers import(\n",
        "    Dense,\n",
        "    GlobalAveragePooling2D,\n",
        ")\n",
        "# for metrics\n",
        "from sklearn.utils.class_weight import compute_class_weight\n",
        "from sklearn.metrics import (\n",
        "    classification_report, \n",
        "    confusion_matrix, \n",
        "    plot_confusion_matrix, \n",
        "    auc,\n",
        "    roc_auc_score,\n",
        "    roc_curve,\n",
        ")\n",
        "# ignore warnings\n",
        "warnings.filterwarnings(\"ignore\")\n",
        "print(\"[INFO] All required packages are ready...\")\n"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] All required packages are ready...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wBH0jWlNlzWF"
      },
      "source": [
        "### Check for GPU access"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_KVLkIqIlzWF",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c9fb6bde-407e-4078-984a-5315c7310216"
      },
      "source": [
        "# check TF vesion\n",
        "print(\"Tensorflow version: \", tf.__version__)\n",
        "\n",
        "# check if GPU is available\n",
        "physical_devices = tf.config.list_physical_devices('GPU')\n",
        "print(\"Num GPUs:\", len(physical_devices))\n",
        "\n",
        "if len(physical_devices) == 0:\n",
        "    print(\"GPU device not found. Running on CPU.\")\n",
        "else:\n",
        "    print('Found GPU at: {}'.format(physical_devices))\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensorflow version:  2.6.0\n",
            "Num GPUs: 1\n",
            "Found GPU at: [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AUkIALhnlzWF"
      },
      "source": [
        "### Set reproducibility seed"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:26.288074Z",
          "iopub.status.busy": "2021-08-14T21:47:26.287743Z",
          "iopub.status.idle": "2021-08-14T21:47:26.298942Z",
          "shell.execute_reply": "2021-08-14T21:47:26.295943Z",
          "shell.execute_reply.started": "2021-08-14T21:47:26.288045Z"
        },
        "id": "9M0IqFPolzWG",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3d889bba-9066-4be0-d0b4-355bccf31d84"
      },
      "source": [
        "# Reproducibility\n",
        "def set_seed(seed=34):\n",
        "    np.random.seed(seed)\n",
        "    tf.random.set_seed(seed)\n",
        "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
        "    os.environ['TF_DETERMINISTIC_OPS'] = '1'\n",
        "    \n",
        "\n",
        "print(\"[INFO] Reproducubilities are ready...\")\n",
        "set_seed(34)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Reproducubilities are ready...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "St-vqBjjyUhv"
      },
      "source": [
        "### Mount google drive"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fqWsHFk_yN9a",
        "outputId": "5916d76f-fcbc-465f-d203-65fb402fcb56"
      },
      "source": [
        "# mount google drive\n",
        "from google.colab import drive\n",
        "\n",
        "drive.mount('/content/drive')\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "izxTZA9QlzWG"
      },
      "source": [
        "### Set working directory"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:26.533969Z",
          "iopub.status.busy": "2021-08-14T21:47:26.533222Z",
          "iopub.status.idle": "2021-08-14T21:47:26.542146Z",
          "shell.execute_reply": "2021-08-14T21:47:26.540807Z",
          "shell.execute_reply.started": "2021-08-14T21:47:26.533927Z"
        },
        "id": "-SmxdTvylzWG",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1ef63489-475c-4f73-8e6b-7184e8b7ae62"
      },
      "source": [
        "import sys\n",
        "\n",
        "# set working directory for google colab\n",
        "path = '/content/drive/MyDrive/Colab Notebooks'\n",
        "sys.path.insert(0, path)\n",
        "print(\"[INFO]  Working directory path is set up...\")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO]  Working directory path is set up...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3cbLQrOAlzWH"
      },
      "source": [
        "### Download dataset and unzipped it"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W3tdajkalzWH"
      },
      "source": [
        "# download data from kaggle api\n",
        "# !pip install -q kaggle\n",
        "# !mkdir ~/.kaggle\n",
        "# !cp kaggle.json ~/.kaggle/\n",
        "# !chmod 600 ~/.kaggle/kaggle.json\n",
        "\n",
        "# !kaggle datasets download -d \"tolgadincer/labeled-chest-xray-images\" -p \"~/workspace/ml-apps/input/\"\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "54Blyz_tlzWH"
      },
      "source": [
        "from utils.unzip_file import unzip_file\n",
        "\n",
        "# path where our data resides\n",
        "file_path = '/content/drive/MyDrive/input/chest_xray.zip'\n",
        "\n",
        "unzip_file(file_path)\n",
        "print(\"[INFO]  Data has been unzipped...\")\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fntpT_U6lzWI"
      },
      "source": [
        "# Load Data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:26.544869Z",
          "iopub.status.busy": "2021-08-14T21:47:26.544030Z",
          "iopub.status.idle": "2021-08-14T21:47:26.566793Z",
          "shell.execute_reply": "2021-08-14T21:47:26.565471Z",
          "shell.execute_reply.started": "2021-08-14T21:47:26.544822Z"
        },
        "id": "jdrKMkoglzWI",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d0a160f7-7739-4e09-bb7e-6cb294d1790c"
      },
      "source": [
        "# set image data directories\n",
        "MAIN_DIR = \"/content/\"\n",
        "TRAIN_DIR = os.path.join(MAIN_DIR, 'train')\n",
        "TEST_DIR = os.path.join(MAIN_DIR, 'test')\n",
        "\n",
        "# look into train folder for .jpg files for normal and pneumonia images\n",
        "NORMAL_TRAIN_IMAGES = os.path.join(MAIN_DIR, \"train/NORMAL/\")\n",
        "PNEUMONIA_TRAIN_IMAGES = os.path.join(MAIN_DIR, \"train/PNEUMONIA/\")\n",
        "\n",
        "# look into train folder for .jpg files for normal and pneumonia images\n",
        "NORMAL_TEST_IMAGES = os.path.join(MAIN_DIR, \"NORMAL\")\n",
        "PNEUMONIA_TES_IMAGES = os.path.join(MAIN_DIR, \"PNEUMONIA\")\n",
        "\n",
        "print(\"[INFO] Data directories are ready...\")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Data directories are ready...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VewLUNSylzWI",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c68b4d9c-aa43-40b8-f152-9ddfe96dcdad"
      },
      "source": [
        "# print some image file names\n",
        "train_normal_names = os.listdir(NORMAL_TRAIN_IMAGES)\n",
        "print(train_normal_names[:3])\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['NORMAL-8311712-0001.jpeg', 'NORMAL-5015362-0001.jpeg', 'NORMAL-5740995-0001.jpeg']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:26.761980Z",
          "iopub.status.busy": "2021-08-14T21:47:26.761496Z",
          "iopub.status.idle": "2021-08-14T21:47:26.772952Z",
          "shell.execute_reply": "2021-08-14T21:47:26.771774Z",
          "shell.execute_reply.started": "2021-08-14T21:47:26.761939Z"
        },
        "id": "_2rkVSIklzWJ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9110bdab-63b3-4f0c-a59d-340e9c55bde9"
      },
      "source": [
        "# let's verify we have two classes\n",
        "classes = os.listdir(TRAIN_DIR)\n",
        "num_classes = len(classes)\n",
        "print('Number of classes: ', num_classes)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of classes:  2\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bxKOA8fzlzWJ"
      },
      "source": [
        "# Data Preprocessing"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "m0Qd14SZlzWJ"
      },
      "source": [
        "### Set somve variables"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:26.781001Z",
          "iopub.status.busy": "2021-08-14T21:47:26.779072Z",
          "iopub.status.idle": "2021-08-14T21:47:26.809433Z",
          "shell.execute_reply": "2021-08-14T21:47:26.808283Z",
          "shell.execute_reply.started": "2021-08-14T21:47:26.780959Z"
        },
        "id": "Kk8puHRIlzWJ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "be04621c-fda7-4306-a5a9-e1339200ea3d"
      },
      "source": [
        "# set our image size\n",
        "image_size = (224, 224)\n",
        "\n",
        "# Initiate an input layer to pass our image size\n",
        "input_shape = image_size + (3, )\n",
        "\n",
        "# ste batch size\n",
        "batch_size = 32\n",
        "\n",
        "# initial number of epochs\n",
        "epochs = 10\n",
        "\n",
        "# set class labels\n",
        "class_names = ['NORMAL', 'PNEUMONIA']\n",
        "\n",
        "# set a seed for data generators\n",
        "data_seed = 1\n",
        "\n",
        "# Set variables for initial learning rate\n",
        "learning_rate = 1e-3\n",
        "\n",
        "# set initial optimizer\n",
        "opt = Adam(\n",
        "    learning_rate=learning_rate,\n",
        "    amsgrad=False,\n",
        "    name='Adam',\n",
        ")\n",
        "\n",
        "# set metrics and loss for model training\n",
        "metrics = ['accuracy']\n",
        "loss = ['binary_crossentropy']\n",
        "\n",
        "print(\"[INFO] Some useful variables have been declared...\")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Some useful variables have been declared...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7Yk--DwHlzWK"
      },
      "source": [
        "## Generate data batches for train/test images"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:26.817953Z",
          "iopub.status.busy": "2021-08-14T21:47:26.815231Z",
          "iopub.status.idle": "2021-08-14T21:47:26.834520Z",
          "shell.execute_reply": "2021-08-14T21:47:26.832397Z",
          "shell.execute_reply.started": "2021-08-14T21:47:26.817910Z"
        },
        "id": "HqlXPozclzWK",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4c7cdcca-3622-442c-8544-1602f30cf606"
      },
      "source": [
        "# Generate batches of tensor image data with real-time data augmentation\n",
        "# Takes the path to a directory & generates batches of augmented data\n",
        "# Training/Validation Split with ImageDataGenerator\n",
        "datagen = ImageDataGenerator(\n",
        "    rescale=1./255,\n",
        "    validation_split=0.1,\n",
        ")\n",
        "\n",
        "\n",
        "def get_train_generator(\n",
        "    dir_path, image_size, batch_size, class_names, seed, shuffle=True\n",
        "):\n",
        "    \"\"\"\n",
        "    Initiating data augment generator that rescale and resizes.\n",
        "    Scale all pixel intensities to the range [0, 1].\n",
        "    \"\"\"\n",
        "    # initialize the training generator\n",
        "    train_generator_ = datagen.flow_from_directory(\n",
        "        directory=dir_path,\n",
        "        target_size=image_size,\n",
        "        batch_size=batch_size,\n",
        "        class_mode=\"binary\",\n",
        "        subset=\"training\",\n",
        "        classes=class_names,\n",
        "        seed=seed,\n",
        "        shuffle=shuffle,\n",
        "    )\n",
        "\n",
        "    return train_generator_\n",
        "\n",
        "\n",
        "def get_val_generator(\n",
        "    dir_path, image_size, batch_size, class_names, seed, shuffle=False\n",
        "):\n",
        "    \"\"\"\n",
        "    initiating data augment generator that rescale and resizes.\n",
        "    Scale all pixel intensities to the range [0, 1].\n",
        "    \"\"\"\n",
        "    # initialize the training generator\n",
        "    val_generator_ = datagen.flow_from_directory(\n",
        "        directory=dir_path,\n",
        "        target_size=image_size,\n",
        "        batch_size=batch_size,\n",
        "        class_mode=\"binary\",\n",
        "        subset=\"validation\",\n",
        "        classes=class_names,\n",
        "        seed=seed,\n",
        "        shuffle=shuffle,\n",
        "    )\n",
        "\n",
        "    return val_generator_\n",
        "\n",
        "def get_test_generator(\n",
        "    dir_path, image_size, batch_size, class_names, seed, shuffle=False\n",
        "):\n",
        "    \"\"\"\n",
        "    initiating data augment generator that rescale and resizes.\n",
        "    Scale all pixel intensities to the range [0, 1].\n",
        "    \"\"\"\n",
        "    # initialize the training data augmentation object\n",
        "    datagen_test = ImageDataGenerator(\n",
        "        rescale=1./255,\n",
        "    )\n",
        "\n",
        "    # initialize the training generator\n",
        "    test_generator_ = datagen_test.flow_from_directory(\n",
        "        directory=dir_path,\n",
        "        target_size=image_size,\n",
        "        batch_size=batch_size,\n",
        "        class_mode=\"binary\",\n",
        "        classes=class_names,\n",
        "        seed=seed,\n",
        "        shuffle=shuffle,\n",
        "    )\n",
        "\n",
        "    return test_generator_\n",
        "\n",
        "print(\"[INFO] Pipeline for data batches has been generated...\")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Pipeline for data batches has been generated...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:26.837519Z",
          "iopub.status.busy": "2021-08-14T21:47:26.836324Z",
          "iopub.status.idle": "2021-08-14T21:47:32.348060Z",
          "shell.execute_reply": "2021-08-14T21:47:32.346865Z",
          "shell.execute_reply.started": "2021-08-14T21:47:26.837440Z"
        },
        "id": "IJ7XmPR6lzWL",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "f5d953e9-c4ba-4979-929e-a05bb690ed41"
      },
      "source": [
        "print(\"[INFO] Getting information for data generators...\")\n",
        "\n",
        "# get training images\n",
        "train_data = get_train_generator(\n",
        "    TRAIN_DIR,\n",
        "    image_size,\n",
        "    batch_size,\n",
        "    class_names,\n",
        "    data_seed,\n",
        ")\n",
        "\n",
        "# get val images\n",
        "val_data = get_val_generator(\n",
        "    TRAIN_DIR,\n",
        "    image_size,\n",
        "    batch_size,\n",
        "    class_names,\n",
        "    data_seed,\n",
        ")\n",
        "\n",
        "# get test images\n",
        "test_data = get_test_generator(\n",
        "    TEST_DIR,\n",
        "    image_size,\n",
        "    batch_size,\n",
        "    class_names,\n",
        "    data_seed,\n",
        ")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Getting information for data generators...\n",
            "Found 4710 images belonging to 2 classes.\n",
            "Found 522 images belonging to 2 classes.\n",
            "Found 624 images belonging to 2 classes.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vsePIhMelzWL"
      },
      "source": [
        "### Check shape of image batches"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:32.350209Z",
          "iopub.status.busy": "2021-08-14T21:47:32.349736Z",
          "iopub.status.idle": "2021-08-14T21:47:33.021306Z",
          "shell.execute_reply": "2021-08-14T21:47:33.020099Z",
          "shell.execute_reply.started": "2021-08-14T21:47:32.350167Z"
        },
        "id": "9QFCKFLalzWL",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7ca10b3b-4b34-44e8-a29e-489ccd9c623b"
      },
      "source": [
        "# see the shape of the image batches\n",
        "for image_batch, labels_batch in train_data:\n",
        "    print('\\nImage Shape: ')\n",
        "    print(image_batch.shape)\n",
        "    print(labels_batch.shape)\n",
        "    break\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Image Shape: \n",
            "(32, 224, 224, 3)\n",
            "(32,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "icpV0rMalzWL"
      },
      "source": [
        "The image_batch is a tensor of the shape (32, 224, 224, 3).\n",
        "\n",
        "This is a batch of 32 images of shape 224x224x3 (the last dimension refers to color channels RGB). The label_batch is a tensor of the shape (32,), these are corresponding labels to the 32 images."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PzAt2yEJlzWM"
      },
      "source": [
        "# Modeling and Training"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6hR4aUbdlzWM"
      },
      "source": [
        "## Create the base model from the pre-trained convnets"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:33.023693Z",
          "iopub.status.busy": "2021-08-14T21:47:33.023005Z",
          "iopub.status.idle": "2021-08-14T21:47:36.420368Z",
          "shell.execute_reply": "2021-08-14T21:47:36.419273Z",
          "shell.execute_reply.started": "2021-08-14T21:47:33.023650Z"
        },
        "id": "SXmuhFAElzWM",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a45cc4bf-094d-45ac-fb38-5e85dfd54345"
      },
      "source": [
        "# Create the base model from the pre-trained model\n",
        "\n",
        "\n",
        "def get_model():\n",
        "    '''load the base model from the pre-trained model VGG16.\n",
        "       Load weights pre-trained on ImageNet.\n",
        "       Do not include the ImageNet classifier at the top.\n",
        "    '''\n",
        "    # load vgg16 pretrained model\n",
        "    pretrained_model = VGG16(\n",
        "        include_top=False,\n",
        "        weights='imagenet',\n",
        "        input_shape=input_shape,\n",
        "        classes=1000,\n",
        "        classifier_activation='softmax'\n",
        "    )\n",
        "\n",
        "    return pretrained_model\n",
        "\n",
        "# get pretraind model\n",
        "base_model = get_model()\n",
        "\n",
        "# base model architecture\n",
        "print(\"[INFO] Summary of the pretrained base model architecture.\\n\")\n",
        "base_model.summary()\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Summary of the pretrained base model architecture.\n",
            "\n",
            "Model: \"vgg16\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "input_1 (InputLayer)         [(None, 224, 224, 3)]     0         \n",
            "_________________________________________________________________\n",
            "block1_conv1 (Conv2D)        (None, 224, 224, 64)      1792      \n",
            "_________________________________________________________________\n",
            "block1_conv2 (Conv2D)        (None, 224, 224, 64)      36928     \n",
            "_________________________________________________________________\n",
            "block1_pool (MaxPooling2D)   (None, 112, 112, 64)      0         \n",
            "_________________________________________________________________\n",
            "block2_conv1 (Conv2D)        (None, 112, 112, 128)     73856     \n",
            "_________________________________________________________________\n",
            "block2_conv2 (Conv2D)        (None, 112, 112, 128)     147584    \n",
            "_________________________________________________________________\n",
            "block2_pool (MaxPooling2D)   (None, 56, 56, 128)       0         \n",
            "_________________________________________________________________\n",
            "block3_conv1 (Conv2D)        (None, 56, 56, 256)       295168    \n",
            "_________________________________________________________________\n",
            "block3_conv2 (Conv2D)        (None, 56, 56, 256)       590080    \n",
            "_________________________________________________________________\n",
            "block3_conv3 (Conv2D)        (None, 56, 56, 256)       590080    \n",
            "_________________________________________________________________\n",
            "block3_pool (MaxPooling2D)   (None, 28, 28, 256)       0         \n",
            "_________________________________________________________________\n",
            "block4_conv1 (Conv2D)        (None, 28, 28, 512)       1180160   \n",
            "_________________________________________________________________\n",
            "block4_conv2 (Conv2D)        (None, 28, 28, 512)       2359808   \n",
            "_________________________________________________________________\n",
            "block4_conv3 (Conv2D)        (None, 28, 28, 512)       2359808   \n",
            "_________________________________________________________________\n",
            "block4_pool (MaxPooling2D)   (None, 14, 14, 512)       0         \n",
            "_________________________________________________________________\n",
            "block5_conv1 (Conv2D)        (None, 14, 14, 512)       2359808   \n",
            "_________________________________________________________________\n",
            "block5_conv2 (Conv2D)        (None, 14, 14, 512)       2359808   \n",
            "_________________________________________________________________\n",
            "block5_conv3 (Conv2D)        (None, 14, 14, 512)       2359808   \n",
            "_________________________________________________________________\n",
            "block5_pool (MaxPooling2D)   (None, 7, 7, 512)         0         \n",
            "=================================================================\n",
            "Total params: 14,714,688\n",
            "Trainable params: 14,714,688\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:36.422330Z",
          "iopub.status.busy": "2021-08-14T21:47:36.421914Z",
          "iopub.status.idle": "2021-08-14T21:47:36.432399Z",
          "shell.execute_reply": "2021-08-14T21:47:36.430620Z",
          "shell.execute_reply.started": "2021-08-14T21:47:36.422286Z"
        },
        "id": "zOdl1ZoOlzWM",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4331e945-a7f7-49ba-b65c-94f8b95e3caf"
      },
      "source": [
        "# check available layers\n",
        "base_layers = len(base_model.layers)\n",
        "print(\"[INFO] There are a total of\", base_layers, \"available layers.\")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] There are a total of 19 available layers.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sJXaNByFlzWM"
      },
      "source": [
        "## Feature extraction\n",
        "\n",
        "### Freeze the convolutional base"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:36.434245Z",
          "iopub.status.busy": "2021-08-14T21:47:36.433734Z",
          "iopub.status.idle": "2021-08-14T21:47:36.448871Z",
          "shell.execute_reply": "2021-08-14T21:47:36.447654Z",
          "shell.execute_reply.started": "2021-08-14T21:47:36.434191Z"
        },
        "id": "BCfCqmRUlzWM",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e0e3a66f-9d7f-415f-db6e-8961ac575a1d"
      },
      "source": [
        "# loop over all layers in the base model and freeze them so they will\n",
        "# *not* be updated during the first training process\n",
        "print(\"[INFO] Base model layers have been freeze...\")\n",
        "for layer in base_model.layers:\n",
        "    layer.trainable = False\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Base model layers have been freeze...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:36.451150Z",
          "iopub.status.busy": "2021-08-14T21:47:36.450679Z",
          "iopub.status.idle": "2021-08-14T21:47:36.462522Z",
          "shell.execute_reply": "2021-08-14T21:47:36.461185Z",
          "shell.execute_reply.started": "2021-08-14T21:47:36.451106Z"
        },
        "id": "GgNRxBnTlzWN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4a86118d-b147-4901-e53e-48e9bad490af"
      },
      "source": [
        "# how many trainable varibles we have\n",
        "train_var = len(base_model.trainable_variables)\n",
        "print(\"[INFO] There are\", train_var , \"trainable variables.\")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] There are 0 trainable variables.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:36.465190Z",
          "iopub.status.busy": "2021-08-14T21:47:36.464708Z",
          "iopub.status.idle": "2021-08-14T21:47:36.485989Z",
          "shell.execute_reply": "2021-08-14T21:47:36.484949Z",
          "shell.execute_reply.started": "2021-08-14T21:47:36.465146Z"
        },
        "id": "levMCch5lzWN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "cc4214ec-4e1c-4c0d-e4ad-39851140e40d"
      },
      "source": [
        "# Let's confirm our layers are not trainable\n",
        "# Should be noted as False = Not trainable\n",
        "for (i, layer) in enumerate(base_model.layers):\n",
        "    print(str(i) + \" \" + layer.__class__.__name__, layer.trainable)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0 InputLayer False\n",
            "1 Conv2D False\n",
            "2 Conv2D False\n",
            "3 MaxPooling2D False\n",
            "4 Conv2D False\n",
            "5 Conv2D False\n",
            "6 MaxPooling2D False\n",
            "7 Conv2D False\n",
            "8 Conv2D False\n",
            "9 Conv2D False\n",
            "10 MaxPooling2D False\n",
            "11 Conv2D False\n",
            "12 Conv2D False\n",
            "13 Conv2D False\n",
            "14 MaxPooling2D False\n",
            "15 Conv2D False\n",
            "16 Conv2D False\n",
            "17 Conv2D False\n",
            "18 MaxPooling2D False\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5xuuIgfllzWN"
      },
      "source": [
        "## Add a classification head to the base model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:36.487950Z",
          "iopub.status.busy": "2021-08-14T21:47:36.487511Z",
          "iopub.status.idle": "2021-08-14T21:47:44.873772Z",
          "shell.execute_reply": "2021-08-14T21:47:44.872468Z",
          "shell.execute_reply.started": "2021-08-14T21:47:36.487910Z"
        },
        "id": "f9g1L92TlzWN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "023f8e25-16af-4055-f942-4035aa7a596c"
      },
      "source": [
        "image_batch, label_batch = next(iter(train_data))\n",
        "feature_batch = base_model(image_batch)\n",
        "print(feature_batch.shape)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(32, 7, 7, 512)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:44.876077Z",
          "iopub.status.busy": "2021-08-14T21:47:44.875609Z",
          "iopub.status.idle": "2021-08-14T21:47:44.896398Z",
          "shell.execute_reply": "2021-08-14T21:47:44.894630Z",
          "shell.execute_reply.started": "2021-08-14T21:47:44.875988Z"
        },
        "id": "8kj2HZVHlzWO",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6f3f4724-36e8-422f-9c7b-e4b5d6dd921b"
      },
      "source": [
        "global_average_layer = tf.keras.layers.GlobalAveragePooling2D()\n",
        "feature_batch_average = global_average_layer(feature_batch)\n",
        "print(feature_batch_average.shape)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(32, 512)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tyJzLgAElzWO"
      },
      "source": [
        "## Construct a top model and compile the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:44.899365Z",
          "iopub.status.busy": "2021-08-14T21:47:44.898611Z",
          "iopub.status.idle": "2021-08-14T21:47:45.008403Z",
          "shell.execute_reply": "2021-08-14T21:47:45.007373Z",
          "shell.execute_reply.started": "2021-08-14T21:47:44.899307Z"
        },
        "id": "kkULaHNclzWO",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "24a261ea-20a8-4f28-92e1-208ae543f2de"
      },
      "source": [
        "# Construct a new model on top of base model\n",
        "\n",
        "\n",
        "def head_model():\n",
        "    '''\n",
        "    Instantiate a Sequential model without an input shape\n",
        "    construct the head model that will be placed on top of the\n",
        "    the pretrained base model\n",
        "    '''\n",
        "    model = Sequential(\n",
        "        [\n",
        "            base_model,\n",
        "            GlobalAveragePooling2D(),\n",
        "            Dense(1, activation='sigmoid', name='prediction_layer'),\n",
        "        ],\n",
        "        name='head_model'\n",
        "    )\n",
        "\n",
        "    model.compile(\n",
        "        optimizer=opt,\n",
        "        loss=loss,\n",
        "        metrics=metrics,\n",
        "    )\n",
        "\n",
        "    return model\n",
        "\n",
        "\n",
        "# place the head FC model on top of the base model (this will become\n",
        "# the actual model we will train)\n",
        "model = head_model()\n",
        "print(\"[INFO] Getting Head model...\")\n",
        "print(\"[INFO] Classification head model has been compiled ...\")\n",
        "model.summary()\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Getting Head model...\n",
            "[INFO] Classification head model has been compiled ...\n",
            "Model: \"head_model\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "vgg16 (Functional)           (None, 7, 7, 512)         14714688  \n",
            "_________________________________________________________________\n",
            "global_average_pooling2d_1 ( (None, 512)               0         \n",
            "_________________________________________________________________\n",
            "prediction_layer (Dense)     (None, 1)                 513       \n",
            "=================================================================\n",
            "Total params: 14,715,201\n",
            "Trainable params: 513\n",
            "Non-trainable params: 14,714,688\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:45.012332Z",
          "iopub.status.busy": "2021-08-14T21:47:45.012048Z",
          "iopub.status.idle": "2021-08-14T21:47:45.018665Z",
          "shell.execute_reply": "2021-08-14T21:47:45.017229Z",
          "shell.execute_reply.started": "2021-08-14T21:47:45.012303Z"
        },
        "id": "R8HUrn2LlzWO",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "0d18b287-f77c-403f-faa0-86cf5c09aa4c"
      },
      "source": [
        "# how many trainable varibles we have\n",
        "train_var = len(model.trainable_variables)\n",
        "print(\"[INFO] There are\", train_var , \"trainable variables.\")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] There are 2 trainable variables.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:45.020949Z",
          "iopub.status.busy": "2021-08-14T21:47:45.020372Z",
          "iopub.status.idle": "2021-08-14T21:47:45.554350Z",
          "shell.execute_reply": "2021-08-14T21:47:45.552966Z",
          "shell.execute_reply.started": "2021-08-14T21:47:45.020872Z"
        },
        "id": "5jVdF_KalzWO"
      },
      "source": [
        "# show and save our base model architecture\n",
        "!mkdir -p output/model_architecture\n",
        "\n",
        "head_model_arch = plot_model(\n",
        "    model,\n",
        "    to_file=\"output/model_architecture/base_model_architecture-01.png\",\n",
        "    show_shapes=True,\n",
        "    expand_nested=True,\n",
        ")\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WVzFW088lzWO"
      },
      "source": [
        "# Train the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T21:47:45.572098Z",
          "iopub.status.busy": "2021-08-14T21:47:45.571609Z",
          "iopub.status.idle": "2021-08-14T22:01:24.105335Z",
          "shell.execute_reply": "2021-08-14T22:01:24.102954Z",
          "shell.execute_reply.started": "2021-08-14T21:47:45.572059Z"
        },
        "id": "1GFqXeHilzWP",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "13645bac-8b07-4590-d278-229e9967ff78"
      },
      "source": [
        "# build model and train it\n",
        "print(\"[INFO] Training head model...\")\n",
        "start = datetime.now()\n",
        "\n",
        "# train model\n",
        "head_hist = model.fit(\n",
        "    train_data,\n",
        "    validation_data=val_data,\n",
        "    epochs=epochs,\n",
        "    verbose=1,\n",
        ")\n",
        "\n",
        "# get the training time\n",
        "training_time = datetime.now() - start\n",
        "print(\"Training completed in time: \", training_time)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Training head model...\n",
            "Epoch 1/10\n",
            "148/148 [==============================] - 61s 403ms/step - loss: 0.5034 - accuracy: 0.7425 - val_loss: 0.4361 - val_accuracy: 0.7586\n",
            "Epoch 2/10\n",
            "148/148 [==============================] - 57s 383ms/step - loss: 0.3966 - accuracy: 0.7934 - val_loss: 0.3609 - val_accuracy: 0.8103\n",
            "Epoch 3/10\n",
            "148/148 [==============================] - 56s 381ms/step - loss: 0.3371 - accuracy: 0.8558 - val_loss: 0.3156 - val_accuracy: 0.8621\n",
            "Epoch 4/10\n",
            "148/148 [==============================] - 56s 380ms/step - loss: 0.2993 - accuracy: 0.8854 - val_loss: 0.2876 - val_accuracy: 0.8927\n",
            "Epoch 5/10\n",
            "148/148 [==============================] - 56s 381ms/step - loss: 0.2744 - accuracy: 0.9004 - val_loss: 0.2670 - val_accuracy: 0.8927\n",
            "Epoch 6/10\n",
            "148/148 [==============================] - 56s 379ms/step - loss: 0.2555 - accuracy: 0.9110 - val_loss: 0.2528 - val_accuracy: 0.8985\n",
            "Epoch 7/10\n",
            "148/148 [==============================] - 56s 377ms/step - loss: 0.2406 - accuracy: 0.9151 - val_loss: 0.2409 - val_accuracy: 0.9004\n",
            "Epoch 8/10\n",
            "148/148 [==============================] - 56s 378ms/step - loss: 0.2291 - accuracy: 0.9191 - val_loss: 0.2296 - val_accuracy: 0.9138\n",
            "Epoch 9/10\n",
            "148/148 [==============================] - 56s 377ms/step - loss: 0.2192 - accuracy: 0.9236 - val_loss: 0.2218 - val_accuracy: 0.9138\n",
            "Epoch 10/10\n",
            "148/148 [==============================] - 56s 377ms/step - loss: 0.2107 - accuracy: 0.9280 - val_loss: 0.2172 - val_accuracy: 0.9138\n",
            "Training completed in time:  0:09:26.865023\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "scdeK3s9lzWP"
      },
      "source": [
        "### Save the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T22:01:24.107721Z",
          "iopub.status.busy": "2021-08-14T22:01:24.107295Z",
          "iopub.status.idle": "2021-08-14T22:01:25.017168Z",
          "shell.execute_reply": "2021-08-14T22:01:25.015779Z",
          "shell.execute_reply.started": "2021-08-14T22:01:24.107665Z"
        },
        "id": "8JQNAtcSlzWP"
      },
      "source": [
        "# Save the entire model as a SavedModel.\n",
        "!mkdir -p output/saved_model\n",
        "\n",
        "# Save the entire model to a HDF5 file.\n",
        "# The '.h5' extension indicates that the model should be saved to HDF5.\n",
        "model.save('output/saved_model/base_model_full-01.h5')\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dH3iQhZ_lzWP"
      },
      "source": [
        "### Check our training history values"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T22:01:25.021511Z",
          "iopub.status.busy": "2021-08-14T22:01:25.021061Z",
          "iopub.status.idle": "2021-08-14T22:01:25.059278Z",
          "shell.execute_reply": "2021-08-14T22:01:25.058283Z",
          "shell.execute_reply.started": "2021-08-14T22:01:25.021450Z"
        },
        "id": "Dyke2_QClzWP",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 359
        },
        "outputId": "0d010dc1-7deb-4298-92f5-ce4cce215a3b"
      },
      "source": [
        "# put on a dataframe our model history values\n",
        "hist_df = pd.DataFrame(data=head_hist.history)\n",
        "hist_df\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>val_loss</th>\n",
              "      <th>val_accuracy</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0.503358</td>\n",
              "      <td>0.742463</td>\n",
              "      <td>0.436134</td>\n",
              "      <td>0.758621</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>0.396615</td>\n",
              "      <td>0.793418</td>\n",
              "      <td>0.360944</td>\n",
              "      <td>0.810345</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>0.337086</td>\n",
              "      <td>0.855839</td>\n",
              "      <td>0.315649</td>\n",
              "      <td>0.862069</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>0.299289</td>\n",
              "      <td>0.885350</td>\n",
              "      <td>0.287624</td>\n",
              "      <td>0.892720</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>0.274405</td>\n",
              "      <td>0.900425</td>\n",
              "      <td>0.267024</td>\n",
              "      <td>0.892720</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5</th>\n",
              "      <td>0.255498</td>\n",
              "      <td>0.911040</td>\n",
              "      <td>0.252824</td>\n",
              "      <td>0.898467</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6</th>\n",
              "      <td>0.240552</td>\n",
              "      <td>0.915074</td>\n",
              "      <td>0.240921</td>\n",
              "      <td>0.900383</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7</th>\n",
              "      <td>0.229117</td>\n",
              "      <td>0.919108</td>\n",
              "      <td>0.229562</td>\n",
              "      <td>0.913793</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8</th>\n",
              "      <td>0.219208</td>\n",
              "      <td>0.923567</td>\n",
              "      <td>0.221775</td>\n",
              "      <td>0.913793</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>9</th>\n",
              "      <td>0.210710</td>\n",
              "      <td>0.928025</td>\n",
              "      <td>0.217236</td>\n",
              "      <td>0.913793</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "       loss  accuracy  val_loss  val_accuracy\n",
              "0  0.503358  0.742463  0.436134      0.758621\n",
              "1  0.396615  0.793418  0.360944      0.810345\n",
              "2  0.337086  0.855839  0.315649      0.862069\n",
              "3  0.299289  0.885350  0.287624      0.892720\n",
              "4  0.274405  0.900425  0.267024      0.892720\n",
              "5  0.255498  0.911040  0.252824      0.898467\n",
              "6  0.240552  0.915074  0.240921      0.900383\n",
              "7  0.229117  0.919108  0.229562      0.913793\n",
              "8  0.219208  0.923567  0.221775      0.913793\n",
              "9  0.210710  0.928025  0.217236      0.913793"
            ]
          },
          "metadata": {},
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zf7IZ8LHlzWP"
      },
      "source": [
        "# Plot Learning curves\n",
        "\n",
        "Plot learning curves to check our training/validation performance\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T22:01:25.061222Z",
          "iopub.status.busy": "2021-08-14T22:01:25.060794Z",
          "iopub.status.idle": "2021-08-14T22:01:26.027356Z",
          "shell.execute_reply": "2021-08-14T22:01:26.026250Z",
          "shell.execute_reply.started": "2021-08-14T22:01:25.061180Z"
        },
        "id": "_VfhXmXylzWP",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 584
        },
        "outputId": "cb225c1c-473a-4e24-8658-18289ca6b846"
      },
      "source": [
        "# let's take a look of our base model performance for 12 epochs\n",
        "!mkdir -p output/plots\n",
        "\n",
        "def plot_model_performance(head_hist):\n",
        "    \"\"\"\n",
        "    plot the performance history of our first model\n",
        "    \"\"\"\n",
        "    # get loss scores\n",
        "    loss = head_hist.history[\"loss\"]\n",
        "    val_loss = head_hist.history[\"val_loss\"]\n",
        "    # get accurcay scores\n",
        "    acc = head_hist.history[\"accuracy\"]\n",
        "    val_acc = head_hist.history[\"val_accuracy\"]\n",
        "    # plot all plots\n",
        "    plt.style.use(\"fivethirtyeight\")\n",
        "    plt.figure(figsize=(10, 10))\n",
        "    # plot loss\n",
        "    plt.subplot(2, 1, 1)\n",
        "    plt.plot(loss, label=\"Training Loss\")\n",
        "    plt.plot(val_loss, label=\"Validation Loss\")\n",
        "    plt.legend(loc=\"best\")\n",
        "    plt.ylabel(\"Cross Entropy\")\n",
        "    plt.title(\"Loss\")\n",
        "    # plot accuracy\n",
        "    plt.subplot(2, 1, 2)\n",
        "    plt.plot(acc, label=\"Training Accuracy\")\n",
        "    plt.plot(val_acc, label=\"Validation Accuracy\")\n",
        "    plt.legend(loc=\"best\")\n",
        "    plt.ylabel(\"Accuracy\")\n",
        "    plt.title(\"Accuracy\")\n",
        "    # save plot\n",
        "    plt.savefig(\"output/plots/learning_curves_for_base_model-01\", dpi=300)\n",
        "    plt.show()\n",
        "\n",
        "\n",
        "plot_model_performance(head_hist)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x720 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-5jXHhPylzWQ"
      },
      "source": [
        "### Evaluate model on test data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.execute_input": "2021-08-14T22:01:26.029651Z",
          "iopub.status.busy": "2021-08-14T22:01:26.029213Z",
          "iopub.status.idle": "2021-08-14T22:01:34.665780Z",
          "shell.execute_reply": "2021-08-14T22:01:34.664575Z",
          "shell.execute_reply.started": "2021-08-14T22:01:26.029608Z"
        },
        "id": "-NcFfkKvlzWQ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "eebc906a-3849-462c-cbba-7dc4841bad93"
      },
      "source": [
        "# Evaluate the model on the test data using 'evaluate'\n",
        "print(\"[INFO] Evaluate on test data...\")\n",
        "test_loss, test_accuracy = model.evaluate(\n",
        "    test_data,\n",
        "    batch_size=batch_size,\n",
        ")\n",
        "\n",
        "# round test_loss to 3 decimals and test accuracy to 2\n",
        "test_loss = round(test_loss, 3)\n",
        "test_accuracy = round(test_accuracy*100, 2)\n",
        "\n",
        "# print model evaluation results\n",
        "print('test loss: ', test_loss)\n",
        "print('test acc:', test_accuracy)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[INFO] Evaluate on test data...\n",
            "20/20 [==============================] - 8s 406ms/step - loss: 0.3880 - accuracy: 0.8077\n",
            "test loss:  0.388\n",
            "test acc: 80.77\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ixzTDc_VlzWQ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e8d0b186-79a3-431a-911a-123ca85986cc"
      },
      "source": [
        "# set threshold of 0.5 if value > 0.5 == class 1 [PNEUMONIA]\n",
        "# if not belongs to class 0 [NORMAL]\n",
        "threshold = 0.5\n",
        "test_data.reset()\n",
        "\n",
        "# real predictions not between 0 and 1. \n",
        "Y_pred = model.predict(\n",
        "    test_data, \n",
        "    batch_size=batch_size,\n",
        "    verbose=1)\n",
        "\n",
        "# getting values for our threshold between 0 and 1\n",
        "y_pred = np.where(Y_pred > threshold, 1, 0)\n",
        "# print(y_pred)\n",
        "# print(y_pred_thresh)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "20/20 [==============================] - 6s 286ms/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ce3x9djglzWQ"
      },
      "source": [
        "### Classification Report"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "F7D_RAVWlzWQ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ec6c59f0-362d-4d35-d9e1-121b1e7b2a0d"
      },
      "source": [
        "classification_report = classification_report(\n",
        "    test_data.classes,\n",
        "    y_pred,\n",
        "    target_names=class_names,\n",
        ")\n",
        "\n",
        "print(classification_report)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "      NORMAL       0.95      0.51      0.67       234\n",
            "   PNEUMONIA       0.77      0.98      0.86       390\n",
            "\n",
            "    accuracy                           0.81       624\n",
            "   macro avg       0.86      0.75      0.77       624\n",
            "weighted avg       0.84      0.81      0.79       624\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ri5aCzSOlzWQ"
      },
      "source": [
        "### Confusion Matrix"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UwYYBQ7UlzWQ",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 619
        },
        "outputId": "5b463b37-205d-4965-9f4b-279981567766"
      },
      "source": [
        "group_names = ['TN', 'FP', 'FN', 'TP']\n",
        "cm = confusion_matrix(test_data.classes, y_pred)\n",
        "\n",
        "\n",
        "def plot_matrix(cm):\n",
        "    plt.figure(figsize=(8, 6))\n",
        "    plt.title('Confusion matrix @0.50\\n')\n",
        "    heatmap = sns.heatmap(\n",
        "            cm,\n",
        "            cmap='OrRd',\n",
        "            linecolor='black',\n",
        "            linewidths=1,\n",
        "            annot=True,\n",
        "            cbar=False,\n",
        "            fmt='',\n",
        "            xticklabels=class_names,\n",
        "            yticklabels=class_names\n",
        "    )\n",
        "    plt.xlabel('Predicted Label')\n",
        "    plt.ylabel('Actual (True) Label')\n",
        "    plt.savefig('output/plots/confusion_matrix_base_model-01', dpi=300)\n",
        "\n",
        "    return heatmap\n",
        "\n",
        "\n",
        "plot_matrix(cm)\n",
        "print('Normal cases correctly Detected - (True Negatives): ', cm[0][0])\n",
        "print('Normal cases incorrectly Detected when is Pneumonia - (False Positives): ', cm[0][1])\n",
        "print('Pneumonia cases correctly Detected - (True Positives): ', cm[1][1])\n",
        "print('Pneumonia cases incorrectly Detected when is Normal -  (False Negatives): ', cm[1][0])\n",
        "print('Total Pneumonia cases: ', np.sum(cm[1]))\n",
        "print('Total Normal cases: ', np.sum(cm[0]))\n",
        "print('\\n')\n",
        "plt.show()\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Normal cases correctly Detected - (True Negatives):  120\n",
            "Normal cases incorrectly Detected when is Pneumonia - (False Positives):  114\n",
            "Pneumonia cases correctly Detected - (True Positives):  384\n",
            "Pneumonia cases incorrectly Detected when is Normal -  (False Negatives):  6\n",
            "Total Pneumonia cases:  390\n",
            "Total Normal cases:  234\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 576x432 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "J2ZW2KvVlzWR"
      },
      "source": [
        "# AUC, Sensitivity and Specificity scores\n",
        "\n",
        "Since we know that our data is imbalanced.  Accuracy might not be the best model evaluation metric every time.  It can convey the health of a model well only when all the classes have similar prevalence in the data.  Thus, when the classes’ distribution is imbalanced, accuracy is not a good model evaluation metric.\n",
        "For this reason we will rely on Sensitivity and Specificity as well as AUC ROC Scores.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pTYh44Q7lzWR",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5701cf2f-3610-457b-f791-ba37b21aa713"
      },
      "source": [
        "TN, FP, FN, TP = cm.ravel()\n",
        "\n",
        "accuracy = (TP + TN)/(TP + TN + FN + FP)\n",
        "sensitivity = (TP)/(TP + FN)\n",
        "specificity = (TN)/(TN + FP)\n",
        "prevalance = np.sum(test_data.classes)/len(test_data.classes)\n",
        "\n",
        "print(\"Metrics:\\n\")\n",
        "print(\"accuracy: {:.2%}\".format(accuracy))\n",
        "# positive cases (pneumonia [1]) predicted correctly\n",
        "print(\"sensitivity: {:.2%}\".format(sensitivity))\n",
        "# negative casse (normal [0]) predicted correctly\n",
        "print(\"specificity: {:.2%}\".format(specificity))\n",
        "# \n",
        "print(\"prevalance: {:.2%}\".format(prevalance))\n",
        "\n",
        "# \n",
        "AUC = roc_auc_score(test_data.classes, y_pred)\n",
        "print(\"AUC ROC SCORE: {:.2%}\".format(AUC))\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Metrics:\n",
            "\n",
            "accuracy: 80.77%\n",
            "sensitivity: 98.46%\n",
            "specificity: 51.28%\n",
            "prevalance: 62.50%\n",
            "AUC ROC SCORE: 74.87%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "x82YghkMDrUy"
      },
      "source": [
        "As we already know from the EDA section, this dataset is highly unbalanced.  There is an imbalance in the accuracy for pneumonia and normal predictions.  This is because the prevalence of pneumonia in this dataset is high.\n",
        "\n",
        "We can determine that in 81% of the cases this model is predicting right for both cases.\n",
        "\n",
        "That is why we won't rely on accuracy as an overall metric for this model.  We could measure both the sensitivity and specificity.\n",
        "\n",
        "We have a low prevalance of 63% because the proportion of normal cases is low.  \n",
        "\n",
        "The sensitivity indicates that out of 100 detections of pneumonia cases our model will predict the class 1 just about right in 98% of the time.  That is, out of 100 pneumonia cases 98 will be predicted right and 2 pneumonia cases will be wrongly predicted as normal cases.\n",
        "\n",
        "However, the specificity out of 100 predictions regarding the normal cases, it will predict just about right those cases 51% of the time.  Yet, it will predict wrong about 49.\n",
        "\n",
        "Meaning, out of 100 normal cases 51 will be predicted right but it will missed 49 cases and will end up being predicted as pneumonia cases.\n",
        "\n",
        "**Note**: These initial base model is good but not great. We know the dataset is imbalanced.  Therefore, let'see if we can tuned this model a bit so that we can get better metrics."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Vtr1MeHJlzWR"
      },
      "source": [
        "# References\n",
        "\n",
        "[Transfer learning and fine-tuning - tensorflow tutorials](https://www.tensorflow.org/tutorials/images/transfer_learning)\n",
        "\n",
        "[Fine tuning with Keras and Deep Learning - tutorial](https://www.pyimagesearch.com/2019/06/03/fine-tuning-with-keras-and-deep-learning/)\n",
        "\n",
        "[writing your own callbacks - tensorflow tutorials](https://www.tensorflow.org/guide/keras/custom_callback#setup)\n",
        "\n",
        "[Data Augmentation in Medical Images](https://towardsdatascience.com/data-augmentation-in-medical-images-95c774e6eaae)"
      ]
    }
  ]
}