Pixel_Quest_CIFAR_10_TF.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/shilpasy/Pixel_Quest_CIFAR-10_TensorFlow/blob/main/Pixel_Quest_CIFAR_10_TF.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-18T18:43:08.021181Z",
          "start_time": "2023-02-18T18:43:02.242395Z"
        },
        "id": "le1xXXyHPACR"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "import matplotlib.pyplot as plt\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HBlh-3jiPACR"
      },
      "source": [
        "# Pixel Quest: CIFAR-10 Image Classification with TensorFlow"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pZeczELLPACS"
      },
      "source": [
        "In the world of image classification here we encounter a fascinating task of distinguishing between various objects and creatures captured in the CIFAR-10 dataset.\n",
        "\n",
        "In the past, scientists have utilized clever filters to unveil the essence of these images. However, modern techniques have evolved, and now the finest algorithms employ the magic of linear algebra to accomplish this feat.\n",
        "\n",
        "Lets build a few of models with increasing complexity to classify these images into one of ten classes. For simplicity and speed, these images are pretty small ($32\\times32\\times3$).  This can make classification a bit tricky. I do not have reference but I have read that human performance is only about 94% on this dataset.\n",
        "\n",
        "In this project, we start with the classic K-Nearest Neighbors (KNN) algorithm and gradually advance to more sophisticated models, including a neural network with a single hidden layer, a Convolutional Neural Network (CNN), and finally, the power of transfer learning using Xception.\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E40fxK7JPACU"
      },
      "source": [
        "## Getting the input data"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Dw6BLb2-PACU"
      },
      "source": [
        "We will be using the `CIFAR-10` data set.  It consists of 60,000 images, each $32\\times32$ color pixels, each belonging to one of ten classes. This is quite popular dataset and we can get it from [various links](https://www.cs.toronto.edu/~kriz/cifar.html) [online](https://www.kaggle.com/datasets/emadtolba/cifar10-comp)."
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "metadata": {
        "id": "RMZ-gv7seQlI",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ed146883-24dc-45a1-eba9-78b3146c1b69"
      },
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /content/drive\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-18T18:43:12.179438Z",
          "start_time": "2023-02-18T18:43:08.025400Z"
        },
        "id": "xHlDJOKFPACV"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "import gzip\n",
        "\n",
        "#train_images = np.load(gzip.open('train_images.npy.gz', 'rb'))\n",
        "#X = np.load(gzip.open('train_labels.npy.gz', 'rb'))\n",
        "#validation_images = np.load(gzip.open('validation_images.npy.gz', 'rb'))\n",
        "\n",
        "train_images = np.load(gzip.open('/content/drive/My Drive/TheDataIncubator/datacourse/tensorflow/miniprojects/train_images.npy.gz', 'rb'))\n",
        "train_labels = np.load(gzip.open('/content/drive/My Drive/TheDataIncubator/datacourse/tensorflow/miniprojects/train_labels.npy.gz', 'rb'))\n",
        "validation_images = np.load(gzip.open('/content/drive/My Drive/TheDataIncubator/datacourse/tensorflow/miniprojects/validation_images.npy.gz', 'rb'))\n"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        ".npy format is the standard binary file format in NumPy for persisting a single arbitrary NumPy array"
      ],
      "metadata": {
        "id": "EHmntVvY3W0J"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LisZNPmLPACW"
      },
      "source": [
        "The images are stored as four-dimensional arrays.  The first index indicates the image number, the second and third the $x$ and $y$ positions, and the fourth index the color channel.  Each pixel color is a floating point number between 0 and 1.  This convention allows us to view the images with matplotlib:"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "train_images[0:1]"
      ],
      "metadata": {
        "id": "qGH7Fq6P0Aij",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "243f6799-ff89-4537-8c62-34ab68ade8c0"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[[[0.23137255, 0.24313726, 0.24705882],\n",
              "         [0.16862746, 0.18039216, 0.1764706 ],\n",
              "         [0.19607843, 0.1882353 , 0.16862746],\n",
              "         ...,\n",
              "         [0.61960787, 0.5176471 , 0.42352942],\n",
              "         [0.59607846, 0.49019608, 0.4       ],\n",
              "         [0.5803922 , 0.4862745 , 0.40392157]],\n",
              "\n",
              "        [[0.0627451 , 0.07843138, 0.07843138],\n",
              "         [0.        , 0.        , 0.        ],\n",
              "         [0.07058824, 0.03137255, 0.        ],\n",
              "         ...,\n",
              "         [0.48235294, 0.34509805, 0.21568628],\n",
              "         [0.46666667, 0.3254902 , 0.19607843],\n",
              "         [0.47843137, 0.34117648, 0.22352941]],\n",
              "\n",
              "        [[0.09803922, 0.09411765, 0.08235294],\n",
              "         [0.0627451 , 0.02745098, 0.        ],\n",
              "         [0.19215687, 0.10588235, 0.03137255],\n",
              "         ...,\n",
              "         [0.4627451 , 0.32941177, 0.19607843],\n",
              "         [0.47058824, 0.32941177, 0.19607843],\n",
              "         [0.42745098, 0.28627452, 0.16470589]],\n",
              "\n",
              "        ...,\n",
              "\n",
              "        [[0.8156863 , 0.6666667 , 0.3764706 ],\n",
              "         [0.7882353 , 0.6       , 0.13333334],\n",
              "         [0.7764706 , 0.6313726 , 0.10196079],\n",
              "         ...,\n",
              "         [0.627451  , 0.52156866, 0.27450982],\n",
              "         [0.21960784, 0.12156863, 0.02745098],\n",
              "         [0.20784314, 0.13333334, 0.07843138]],\n",
              "\n",
              "        [[0.7058824 , 0.54509807, 0.3764706 ],\n",
              "         [0.6784314 , 0.48235294, 0.16470589],\n",
              "         [0.7294118 , 0.5647059 , 0.11764706],\n",
              "         ...,\n",
              "         [0.72156864, 0.5803922 , 0.36862746],\n",
              "         [0.38039216, 0.24313726, 0.13333334],\n",
              "         [0.3254902 , 0.20784314, 0.13333334]],\n",
              "\n",
              "        [[0.69411767, 0.5647059 , 0.45490196],\n",
              "         [0.65882355, 0.5058824 , 0.36862746],\n",
              "         [0.7019608 , 0.5568628 , 0.34117648],\n",
              "         ...,\n",
              "         [0.84705883, 0.72156864, 0.54901963],\n",
              "         [0.5921569 , 0.4627451 , 0.32941177],\n",
              "         [0.48235294, 0.36078432, 0.28235295]]]], dtype=float32)"
            ]
          },
          "metadata": {},
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T21:42:14.478239Z",
          "start_time": "2023-02-14T21:42:14.151997Z"
        },
        "id": "1kcVEfIwPACW",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 430
        },
        "outputId": "c42aa9e7-1fda-46c6-f5f5-f1dcd7adda84"
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ],
      "source": [
        "import matplotlib.pyplot\n",
        "matplotlib.pyplot.rcParams[\"axes.grid\"] = False  #  Remove the grid lines from the image.\n",
        "matplotlib.pyplot.imshow(train_images[12])\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "let's savor the sight of a noble horse in a slightly blurry image, appreciating its mystique"
      ],
      "metadata": {
        "id": "t6pqqaXuyFgF"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T21:42:14.590685Z",
          "start_time": "2023-02-14T21:42:14.480919Z"
        },
        "id": "6Oxy5QUtPACX",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 147
        },
        "outputId": "5ec2edba-ce8f-47b4-e50d-9ce633364536"
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 110x110 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ],
      "source": [
        "#if you want to see those images smaller for better clarity you can do the following\n",
        "fig, axes = plt.subplots(figsize = (1.1,1.1))\n",
        "axes.imshow(train_images[12,1:])\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "The human-readable names associated with this classes are :"
      ],
      "metadata": {
        "id": "-DmkRuMJzqGn"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T21:42:14.760266Z",
          "start_time": "2023-02-14T21:42:14.593258Z"
        },
        "id": "mg2eQ8XaPACX",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 469
        },
        "outputId": "802aa5b6-c89b-465a-ef4c-8214ffd9aa78"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Text(0.5, 1.0, 'Class distribution in training set')"
            ]
          },
          "metadata": {},
          "execution_count": 7
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ],
      "source": [
        "classes_name = ['Airplane', 'Automobile', 'Bird', 'Cat', 'Deer', 'Dog', 'Frog', 'Horse', 'Ship', 'Truck']\n",
        "classes, counts = np.unique(train_labels, return_counts=True)\n",
        "plt.barh(classes_name, counts)\n",
        "plt.title('Class distribution in training set')"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "From 'airplanes' soaring through the skies to 'automobiles' speeding on the ground, from 'birds' singing melodies to 'cats' prowling gracefully, and from 'dogs' loyal as friends to 'ships' sailing the vast seas - we shall encounter a multitude of wonders!"
      ],
      "metadata": {
        "id": "iYa9zpLLyQ1J"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nGQVCZdAPACX"
      },
      "source": [
        "The classes have already been numbered 0-9 for us; those numbers are stored in the vector `train_labels`."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iSbVl8j3PACY"
      },
      "source": [
        "## k-nearest neighbors algorithm"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ykcEl_7aPACY"
      },
      "source": [
        "Our initial model is based on a simple algorithm of KNN. We shall measure the differences between colors of the corresponding pixels between two images and use this information to find the nearest companions among these images. Let's calculate a simple euclidean distance for this purpose.\n",
        "\n",
        "$$dist = \\sqrt{\\Delta R^2 + \\Delta G^2 + \\Delta B^2} $$\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-11T06:01:01.142063Z",
          "start_time": "2023-02-11T06:01:01.136900Z"
        },
        "scrolled": true,
        "id": "6Z1wAKIAPACZ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "35a03012-abd2-4ec9-c67d-c465717b3ce0"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "shape of R1 is (32, 32) and shape of train_images[0] is (32, 32, 3)\n",
            "shape of Rall is (50000, 32, 32)\n",
            "\n"
          ]
        }
      ],
      "source": [
        "R1 = train_images[0, :, :, 0]\n",
        "#this is how you access Red\n",
        "G1 = train_images[0, :, :, 1]\n",
        "#this is how you access Green\n",
        "\n",
        "B1 = train_images[0, :, :, 2]\n",
        "#this is how you access Blue\n",
        "\n",
        "print(f\"shape of R1 is {R1.shape} and shape of train_images[0] is {train_images[0].shape}\")\n",
        "\n",
        "Rall = train_images[:,:,:,0]\n",
        "print(f\"shape of Rall is {Rall.shape}\" )\n",
        "# print(train_images[0])\n",
        "# print(R1)\n",
        "print()\n",
        "#print(train_images)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-18T18:52:41.464426Z",
          "start_time": "2023-02-18T18:52:41.457695Z"
        },
        "id": "3LiTNIWSPACZ"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "\n",
        "@tf.function\n",
        "def pixel_diff(images, base):\n",
        "\n",
        "#   delta_R = tf.math.subtract(images[:,:,:,0], base[:,:,:,0]) #Red\n",
        "#   delta_G = tf.math.subtract(images[:,:,:,1], base[:,:,:,1]) #Green\n",
        "#   delta_B = tf.math.subtract(images[:,:,:,2], base[:,:,:,2]) #Blue\n",
        "#   Alternate way to do this: Split the images and base into separate RGB color channels\n",
        "    images_r, images_g, images_b = tf.split(images, 3, axis=-1)\n",
        "    base_r, base_g, base_b = tf.split(base, 3, axis=-1)\n",
        "\n",
        "#   Calculate the differences between the corresponding pixels in images and base for each color channel\n",
        "    delta_r = images_r - base_r\n",
        "    delta_g = images_g - base_g\n",
        "    delta_b = images_b - base_b\n",
        "\n",
        "    # Calculate R_mean\n",
        "    #R_mean = 0.5 * (images[:,:,:,0] + base[:,:,:,0])\n",
        "    R_mean = 0.5 * (images_r + base_r)\n",
        "\n",
        "    #calculate delC for each pixel\n",
        "    dist = tf.sqrt(delta_r**2 + delta_g**2 + delta_b**2 )\n",
        "    avg_dist = tf.math.reduce_sum(dist, axis=(1, 2, 3))/ 1024 #32*32 is 1024\n",
        "\n",
        "    return avg_dist"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-18T18:52:43.547424Z",
          "start_time": "2023-02-18T18:52:42.129983Z"
        },
        "id": "w821BOZJPACZ"
      },
      "outputs": [],
      "source": [
        "base = validation_images[12] # SELECTING THE BASE IMAGE AS THE HORSE\n",
        "dist_all_training_img = pixel_diff(train_images, base)\n",
        "indices_closest_5 = np.argsort(dist_all_training_img)[:5]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-18T18:52:49.527171Z",
          "start_time": "2023-02-18T18:52:49.523102Z"
        },
        "id": "7kmFsRCPPACZ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "815f24b9-5e29-4410-ff6d-09e410532c09"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "TensorShape([50000])"
            ]
          },
          "metadata": {},
          "execution_count": 11
        }
      ],
      "source": [
        "dist_all_training_img.shape"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-18T18:52:54.506620Z",
          "start_time": "2023-02-18T18:52:54.499141Z"
        },
        "id": "DXXqg35MPACa",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4d01a0c1-b41d-4d88-a8a3-fca761f95434"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([ 7565, 19064,  2977, 27189, 19417])"
            ]
          },
          "metadata": {},
          "execution_count": 11
        }
      ],
      "source": [
        "indices_closest_5"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's have a look at these 5 images which match closest with the above hourse."
      ],
      "metadata": {
        "id": "4GVIr8jJHmQN"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "ax1 = plt.subplot(555)\n",
        "# equivalent but more general\n",
        "ax1 = plt.subplot(551)\n",
        "ax1.imshow(train_images[7565,1:])\n",
        "\n",
        "# add a subplot with no frame\n",
        "ax2 = plt.subplot(552)\n",
        "ax2.imshow(train_images[19064,1:])\n",
        "\n",
        "# add a polar subplot\n",
        "ax3 = plt.subplot(553)\n",
        "ax3.imshow(train_images[2977,1:])\n",
        "\n",
        "# add a red subplot that shares the x-axis with ax1\n",
        "ax4 = plt.subplot(554)\n",
        "ax4.imshow(train_images[27189,1:])\n",
        "\n",
        "ax5 = plt.subplot(555)\n",
        "ax5.imshow(train_images[19417,1:])\n"
      ],
      "metadata": {
        "id": "gyocLnMR-IfC",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 145
        },
        "outputId": "b63c61f4-7176-47ac-c026-c5559ef242f7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f1fb43a0160>"
            ]
          },
          "metadata": {},
          "execution_count": 12
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 5 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Behold! The Magic of Simplicity: Identifying an image of a horse without Deep Neural Networks.... This model does work pretty good but on larger dataset and for higher accuracy, we will move to deep neural networks.\n"
      ],
      "metadata": {
        "id": "VnZzDuf1BGBg"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vW5uPQ8KPACa"
      },
      "source": [
        "## Softmax model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BHzecLarPACj"
      },
      "source": [
        "We'll start with a basic neural network called a softmax classifier. This network has just one dense layer that takes 32×32×3 inputs and produces 10 outputs, each representing a class. It will be using an activation function called softmax and a `CategoricalCrossentropy` loss function."
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#Split the train dataset\n",
        "from sklearn.model_selection import train_test_split\n",
        "X_train,X_test,y_train,y_test = train_test_split(train_images,train_labels,test_size=0.2,random_state=120422)\n",
        "print(X_train.shape,X_test.shape)\n",
        "print(y_train.shape,y_test.shape)"
      ],
      "metadata": {
        "id": "o5amCj8H6FtG",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "eeb9851e-5c03-44ac-9273-fed690fec78b"
      },
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(40000, 32, 32, 3) (10000, 32, 32, 3)\n",
            "(40000,) (10000,)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:31:26.826163Z",
          "start_time": "2023-02-14T04:31:26.787722Z"
        },
        "id": "z8Ux0fJlPACj"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.layers import Dense, Input, Dropout,BatchNormalization,Flatten\n",
        "from tensorflow.keras import backend\n",
        "from tensorflow.keras import optimizers\n",
        "from tensorflow.keras import losses\n",
        "backend.clear_session()\n",
        "\n",
        "model = keras.models.Sequential()\n",
        "\n",
        "model.add(Dense(input_shape = (32,32,3), activation='relu',units = 10))\n",
        "model.add(Flatten())\n",
        "model.add(Dense(10,activation = 'softmax'))\n",
        "\n",
        "adam = optimizers.Adam(learning_rate=1e-3)\n",
        "model.compile(loss=losses.categorical_crossentropy, optimizer=adam, metrics=['accuracy'])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:31:27.490010Z",
          "start_time": "2023-02-14T04:31:27.486342Z"
        },
        "scrolled": true,
        "id": "SyS5haAKPACj",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "dd72262e-a2b4-478b-8c56-45f7f62318c6"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " dense (Dense)               (None, 32, 32, 10)        40        \n",
            "                                                                 \n",
            " flatten (Flatten)           (None, 10240)             0         \n",
            "                                                                 \n",
            " dense_1 (Dense)             (None, 10)                102410    \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 102,450\n",
            "Trainable params: 102,450\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ],
      "source": [
        "model.summary()"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#Visualization of CNN architecture\n",
        "from IPython.display import SVG\n",
        "from keras.utils.vis_utils import model_to_dot\n",
        "SVG(model_to_dot(model,show_shapes=True,show_layer_names=True,rankdir='TB').create(prog='dot',format='svg'))\n"
      ],
      "metadata": {
        "id": "nyckVtXs599B",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 561
        },
        "outputId": "c308ae3a-76ae-4dae-d17f-8eb750e688d5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<IPython.core.display.SVG object>"
            ],
            "image/svg+xml": "<svg xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\" width=\"353pt\" height=\"405pt\" viewBox=\"0.00 0.00 265.00 304.00\">\n<g id=\"graph0\" class=\"graph\" transform=\"scale(0.75 0.75) rotate(0) translate(4 300)\">\n<title>G</title>\n<polygon fill=\"white\" stroke=\"transparent\" points=\"-4,4 -4,-300 261,-300 261,4 -4,4\"/>\n<!-- 139453671606832 -->\n<g id=\"node1\" class=\"node\">\n<title>139453671606832</title>\n<polygon fill=\"none\" stroke=\"black\" points=\"0,-249.5 0,-295.5 257,-295.5 257,-249.5 0,-249.5\"/>\n<text text-anchor=\"middle\" x=\"41\" y=\"-280.3\" font-family=\"Times,serif\" font-size=\"14.00\">dense_input</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"0,-272.5 82,-272.5 \"/>\n<text text-anchor=\"middle\" x=\"41\" y=\"-257.3\" font-family=\"Times,serif\" font-size=\"14.00\">InputLayer</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"82,-249.5 82,-295.5 \"/>\n<text text-anchor=\"middle\" x=\"109.5\" y=\"-280.3\" font-family=\"Times,serif\" font-size=\"14.00\">input:</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"82,-272.5 137,-272.5 \"/>\n<text text-anchor=\"middle\" x=\"109.5\" y=\"-257.3\" font-family=\"Times,serif\" font-size=\"14.00\">output:</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"137,-249.5 137,-295.5 \"/>\n<text text-anchor=\"middle\" x=\"197\" y=\"-280.3\" font-family=\"Times,serif\" font-size=\"14.00\">[(None, 32, 32, 3)]</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"137,-272.5 257,-272.5 \"/>\n<text text-anchor=\"middle\" x=\"197\" y=\"-257.3\" font-family=\"Times,serif\" font-size=\"14.00\">[(None, 32, 32, 3)]</text>\n</g>\n<!-- 139450345868016 -->\n<g id=\"node2\" class=\"node\">\n<title>139450345868016</title>\n<polygon fill=\"none\" stroke=\"black\" points=\"17,-166.5 17,-212.5 240,-212.5 240,-166.5 17,-166.5\"/>\n<text text-anchor=\"middle\" x=\"42\" y=\"-197.3\" font-family=\"Times,serif\" font-size=\"14.00\">dense</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"17,-189.5 67,-189.5 \"/>\n<text text-anchor=\"middle\" x=\"42\" y=\"-174.3\" font-family=\"Times,serif\" font-size=\"14.00\">Dense</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"67,-166.5 67,-212.5 \"/>\n<text text-anchor=\"middle\" x=\"94.5\" y=\"-197.3\" font-family=\"Times,serif\" font-size=\"14.00\">input:</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"67,-189.5 122,-189.5 \"/>\n<text text-anchor=\"middle\" x=\"94.5\" y=\"-174.3\" font-family=\"Times,serif\" font-size=\"14.00\">output:</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"122,-166.5 122,-212.5 \"/>\n<text text-anchor=\"middle\" x=\"181\" y=\"-197.3\" font-family=\"Times,serif\" font-size=\"14.00\">(None, 32, 32, 3)</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"122,-189.5 240,-189.5 \"/>\n<text text-anchor=\"middle\" x=\"181\" y=\"-174.3\" font-family=\"Times,serif\" font-size=\"14.00\">(None, 32, 32, 10)</text>\n</g>\n<!-- 139453671606832&#45;&gt;139450345868016 -->\n<g id=\"edge1\" class=\"edge\">\n<title>139453671606832-&gt;139450345868016</title>\n<path fill=\"none\" stroke=\"black\" d=\"M128.5,-249.37C128.5,-241.15 128.5,-231.66 128.5,-222.73\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"132,-222.61 128.5,-212.61 125,-222.61 132,-222.61\"/>\n</g>\n<!-- 139450345870416 -->\n<g id=\"node3\" class=\"node\">\n<title>139450345870416</title>\n<polygon fill=\"none\" stroke=\"black\" points=\"15,-83.5 15,-129.5 242,-129.5 242,-83.5 15,-83.5\"/>\n<text text-anchor=\"middle\" x=\"42\" y=\"-114.3\" font-family=\"Times,serif\" font-size=\"14.00\">flatten</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"15,-106.5 69,-106.5 \"/>\n<text text-anchor=\"middle\" x=\"42\" y=\"-91.3\" font-family=\"Times,serif\" font-size=\"14.00\">Flatten</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"69,-83.5 69,-129.5 \"/>\n<text text-anchor=\"middle\" x=\"96.5\" y=\"-114.3\" font-family=\"Times,serif\" font-size=\"14.00\">input:</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"69,-106.5 124,-106.5 \"/>\n<text text-anchor=\"middle\" x=\"96.5\" y=\"-91.3\" font-family=\"Times,serif\" font-size=\"14.00\">output:</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"124,-83.5 124,-129.5 \"/>\n<text text-anchor=\"middle\" x=\"183\" y=\"-114.3\" font-family=\"Times,serif\" font-size=\"14.00\">(None, 32, 32, 10)</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"124,-106.5 242,-106.5 \"/>\n<text text-anchor=\"middle\" x=\"183\" y=\"-91.3\" font-family=\"Times,serif\" font-size=\"14.00\">(None, 10240)</text>\n</g>\n<!-- 139450345868016&#45;&gt;139450345870416 -->\n<g id=\"edge2\" class=\"edge\">\n<title>139450345868016-&gt;139450345870416</title>\n<path fill=\"none\" stroke=\"black\" d=\"M128.5,-166.37C128.5,-158.15 128.5,-148.66 128.5,-139.73\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"132,-139.61 128.5,-129.61 125,-139.61 132,-139.61\"/>\n</g>\n<!-- 139450345878864 -->\n<g id=\"node4\" class=\"node\">\n<title>139450345878864</title>\n<polygon fill=\"none\" stroke=\"black\" points=\"22.5,-0.5 22.5,-46.5 234.5,-46.5 234.5,-0.5 22.5,-0.5\"/>\n<text text-anchor=\"middle\" x=\"53\" y=\"-31.3\" font-family=\"Times,serif\" font-size=\"14.00\">dense_1</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"22.5,-23.5 83.5,-23.5 \"/>\n<text text-anchor=\"middle\" x=\"53\" y=\"-8.3\" font-family=\"Times,serif\" font-size=\"14.00\">Dense</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"83.5,-0.5 83.5,-46.5 \"/>\n<text text-anchor=\"middle\" x=\"111\" y=\"-31.3\" font-family=\"Times,serif\" font-size=\"14.00\">input:</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"83.5,-23.5 138.5,-23.5 \"/>\n<text text-anchor=\"middle\" x=\"111\" y=\"-8.3\" font-family=\"Times,serif\" font-size=\"14.00\">output:</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"138.5,-0.5 138.5,-46.5 \"/>\n<text text-anchor=\"middle\" x=\"186.5\" y=\"-31.3\" font-family=\"Times,serif\" font-size=\"14.00\">(None, 10240)</text>\n<polyline fill=\"none\" stroke=\"black\" points=\"138.5,-23.5 234.5,-23.5 \"/>\n<text text-anchor=\"middle\" x=\"186.5\" y=\"-8.3\" font-family=\"Times,serif\" font-size=\"14.00\">(None, 10)</text>\n</g>\n<!-- 139450345870416&#45;&gt;139450345878864 -->\n<g id=\"edge3\" class=\"edge\">\n<title>139450345870416-&gt;139450345878864</title>\n<path fill=\"none\" stroke=\"black\" d=\"M128.5,-83.37C128.5,-75.15 128.5,-65.66 128.5,-56.73\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"132,-56.61 128.5,-46.61 125,-56.61 132,-56.61\"/>\n</g>\n</g>\n</svg>"
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:31:41.057386Z",
          "start_time": "2023-02-14T04:31:41.052760Z"
        },
        "id": "cyWWMQTwPACj"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.utils import to_categorical as one_hot\n",
        "\n",
        "y_train_hot = one_hot(y_train)\n",
        "# y_test_hot  = one_hot(y_test)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:33:04.432039Z",
          "start_time": "2023-02-14T04:31:41.704941Z"
        },
        "id": "287mSJAEPACk",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "b552a1a9-c20a-4bbe-fbc3-04ac706bff6e"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/6\n",
            "3200/3200 [==============================] - 26s 8ms/step - loss: 1.8079 - accuracy: 0.3651 - val_loss: 1.7099 - val_accuracy: 0.4013\n",
            "Epoch 2/6\n",
            "3200/3200 [==============================] - 21s 7ms/step - loss: 1.6746 - accuracy: 0.4161 - val_loss: 1.6619 - val_accuracy: 0.4148\n",
            "Epoch 3/6\n",
            "3200/3200 [==============================] - 27s 8ms/step - loss: 1.6100 - accuracy: 0.4417 - val_loss: 1.6423 - val_accuracy: 0.4281\n",
            "Epoch 4/6\n",
            "3200/3200 [==============================] - 22s 7ms/step - loss: 1.5628 - accuracy: 0.4590 - val_loss: 1.6263 - val_accuracy: 0.4374\n",
            "Epoch 5/6\n",
            "3200/3200 [==============================] - 22s 7ms/step - loss: 1.5328 - accuracy: 0.4716 - val_loss: 1.6078 - val_accuracy: 0.4390\n",
            "Epoch 6/6\n",
            "3200/3200 [==============================] - 22s 7ms/step - loss: 1.5087 - accuracy: 0.4823 - val_loss: 1.5995 - val_accuracy: 0.4446\n"
          ]
        }
      ],
      "source": [
        "history=model.fit(X_train, y_train_hot,\n",
        "          validation_split=0.2,\n",
        "          epochs=10,\n",
        "          batch_size=10)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:35:12.315645Z",
          "start_time": "2023-02-14T04:35:11.687396Z"
        },
        "id": "2gZ-3MUYPACk",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8f4f52d6-7213-4a3c-ce36-2017fc54090a"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "313/313 [==============================] - 1s 3ms/step\n"
          ]
        }
      ],
      "source": [
        "y_pred = model.predict(X_test)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:35:12.323282Z",
          "start_time": "2023-02-14T04:35:12.319528Z"
        },
        "id": "t3xj7p4GPACk"
      },
      "outputs": [],
      "source": [
        "predicted_classes=np.argmax(y_pred,axis=1)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FushYP0pPACk"
      },
      "source": [
        "## Fully-connected model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "H6Vsgg3NPACk"
      },
      "source": [
        "Now, let's introduce a hidden layer in the hope that it will enhance our image classification capabilities."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T21:04:39.616648Z",
          "start_time": "2023-02-14T21:04:39.592754Z"
        },
        "id": "1RikKFTUPACl"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras import backend\n",
        "from tensorflow.keras.layers import Dense, Input, Dropout,BatchNormalization,Flatten\n",
        "from tensorflow.keras import optimizers\n",
        "from tensorflow.keras import losses\n",
        "backend.clear_session()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:42:42.409552Z",
          "start_time": "2023-02-14T04:42:42.329863Z"
        },
        "id": "5oFAzn9MPACl"
      },
      "outputs": [],
      "source": [
        "model = keras.models.Sequential()\n",
        "\n",
        "model.add(Dense(input_shape = (32,32,3), activation='relu',units = 10))\n",
        "model.add(Dense(128,activation = 'relu'))\n",
        "model.add(Dropout(0.3)) #######this time adding dropout\n",
        "model.add(Flatten())\n",
        "model.add(Dense(10,activation = 'softmax'))\n",
        "\n",
        "adam = optimizers.Adam(learning_rate=1e-3)\n",
        "model.compile(loss=losses.categorical_crossentropy, optimizer=adam, metrics=['accuracy'])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:42:43.555705Z",
          "start_time": "2023-02-14T04:42:43.551292Z"
        },
        "id": "utxJ0nSpPACl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c7c000cf-4729-440e-b037-5b8109099884"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " dense (Dense)               (None, 32, 32, 10)        40        \n",
            "                                                                 \n",
            " dense_1 (Dense)             (None, 32, 32, 128)       1408      \n",
            "                                                                 \n",
            " dropout (Dropout)           (None, 32, 32, 128)       0         \n",
            "                                                                 \n",
            " flatten (Flatten)           (None, 131072)            0         \n",
            "                                                                 \n",
            " dense_2 (Dense)             (None, 10)                1310730   \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 1,312,178\n",
            "Trainable params: 1,312,178\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ],
      "source": [
        "model.summary()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:43:25.637894Z",
          "start_time": "2023-02-14T04:43:25.631737Z"
        },
        "id": "Q8XIjkagPACl"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.utils import to_categorical as one_hot\n",
        "\n",
        "y_train_hot = one_hot(y_train)\n",
        "# y_test_hot  = one_hot(y_test)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T04:47:45.099289Z",
          "start_time": "2023-02-14T04:43:26.274980Z"
        },
        "id": "cfFq-1oYPACl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "db0557e9-0e4c-445b-9ab4-84d5cf35caec"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/30\n",
            "247/247 [==============================] - 4s 17ms/step - loss: 1.2652 - accuracy: 0.5641 - val_loss: 1.4823 - val_accuracy: 0.4846\n",
            "Epoch 2/30\n",
            "247/247 [==============================] - 4s 17ms/step - loss: 1.2432 - accuracy: 0.5717 - val_loss: 1.5055 - val_accuracy: 0.4720\n",
            "Epoch 3/30\n",
            "247/247 [==============================] - 4s 15ms/step - loss: 1.2203 - accuracy: 0.5798 - val_loss: 1.4922 - val_accuracy: 0.4900\n",
            "Epoch 4/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 1.1980 - accuracy: 0.5884 - val_loss: 1.5210 - val_accuracy: 0.4774\n",
            "Epoch 5/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 1.1760 - accuracy: 0.5967 - val_loss: 1.5334 - val_accuracy: 0.4800\n",
            "Epoch 6/30\n",
            "247/247 [==============================] - 5s 20ms/step - loss: 1.1476 - accuracy: 0.6057 - val_loss: 1.5365 - val_accuracy: 0.4733\n",
            "Epoch 7/30\n",
            "247/247 [==============================] - 5s 20ms/step - loss: 1.1223 - accuracy: 0.6169 - val_loss: 1.5408 - val_accuracy: 0.4755\n",
            "Epoch 8/30\n",
            "247/247 [==============================] - 5s 22ms/step - loss: 1.1063 - accuracy: 0.6217 - val_loss: 1.5723 - val_accuracy: 0.4686\n",
            "Epoch 9/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 1.0868 - accuracy: 0.6290 - val_loss: 1.5849 - val_accuracy: 0.4776\n",
            "Epoch 10/30\n",
            "247/247 [==============================] - 4s 15ms/step - loss: 1.0628 - accuracy: 0.6361 - val_loss: 1.5925 - val_accuracy: 0.4715\n",
            "Epoch 11/30\n",
            "247/247 [==============================] - 4s 15ms/step - loss: 1.0419 - accuracy: 0.6435 - val_loss: 1.6062 - val_accuracy: 0.4746\n",
            "Epoch 12/30\n",
            "247/247 [==============================] - 4s 17ms/step - loss: 1.0162 - accuracy: 0.6520 - val_loss: 1.6215 - val_accuracy: 0.4726\n",
            "Epoch 13/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 1.0041 - accuracy: 0.6562 - val_loss: 1.6640 - val_accuracy: 0.4683\n",
            "Epoch 14/30\n",
            "247/247 [==============================] - 5s 19ms/step - loss: 0.9767 - accuracy: 0.6660 - val_loss: 1.6559 - val_accuracy: 0.4595\n",
            "Epoch 15/30\n",
            "247/247 [==============================] - 5s 21ms/step - loss: 0.9604 - accuracy: 0.6732 - val_loss: 1.6461 - val_accuracy: 0.4684\n",
            "Epoch 16/30\n",
            "247/247 [==============================] - 5s 19ms/step - loss: 0.9386 - accuracy: 0.6812 - val_loss: 1.6497 - val_accuracy: 0.4708\n",
            "Epoch 17/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.9242 - accuracy: 0.6835 - val_loss: 1.7154 - val_accuracy: 0.4611\n",
            "Epoch 18/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.9075 - accuracy: 0.6926 - val_loss: 1.7344 - val_accuracy: 0.4600\n",
            "Epoch 19/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.8933 - accuracy: 0.6968 - val_loss: 1.7174 - val_accuracy: 0.4658\n",
            "Epoch 20/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.8663 - accuracy: 0.7049 - val_loss: 1.7621 - val_accuracy: 0.4684\n",
            "Epoch 21/30\n",
            "247/247 [==============================] - 4s 15ms/step - loss: 0.8601 - accuracy: 0.7092 - val_loss: 1.7620 - val_accuracy: 0.4610\n",
            "Epoch 22/30\n",
            "247/247 [==============================] - 4s 17ms/step - loss: 0.8378 - accuracy: 0.7157 - val_loss: 1.7862 - val_accuracy: 0.4530\n",
            "Epoch 23/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.8228 - accuracy: 0.7193 - val_loss: 1.8451 - val_accuracy: 0.4530\n",
            "Epoch 24/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.8066 - accuracy: 0.7304 - val_loss: 1.8252 - val_accuracy: 0.4544\n",
            "Epoch 25/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.7976 - accuracy: 0.7310 - val_loss: 1.8713 - val_accuracy: 0.4501\n",
            "Epoch 26/30\n",
            "247/247 [==============================] - 5s 18ms/step - loss: 0.7815 - accuracy: 0.7341 - val_loss: 1.8493 - val_accuracy: 0.4575\n",
            "Epoch 27/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.7627 - accuracy: 0.7411 - val_loss: 1.8504 - val_accuracy: 0.4647\n",
            "Epoch 28/30\n",
            "247/247 [==============================] - 4s 16ms/step - loss: 0.7549 - accuracy: 0.7448 - val_loss: 1.8971 - val_accuracy: 0.4566\n",
            "Epoch 29/30\n",
            "247/247 [==============================] - 4s 17ms/step - loss: 0.7357 - accuracy: 0.7508 - val_loss: 1.9283 - val_accuracy: 0.4466\n",
            "Epoch 30/30\n",
            "247/247 [==============================] - 4s 15ms/step - loss: 0.7176 - accuracy: 0.7576 - val_loss: 1.9458 - val_accuracy: 0.4469\n"
          ]
        }
      ],
      "source": [
        "history=model.fit(X_train, y_train_hot,\n",
        "          validation_split=0.2,\n",
        "          epochs=30,\n",
        "          batch_size=130)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "As you can notice there isn't much difference in the accuracy of this and the previous model!"
      ],
      "metadata": {
        "id": "UnEgeYJwB8zp"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FAR84GDwPACm"
      },
      "source": [
        "## Convolutional neural network"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "x5trZzdPPACm"
      },
      "source": [
        "Now let's build our most anticipated and historically successful CNN\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T05:55:55.638935Z",
          "start_time": "2023-02-14T05:55:55.035618Z"
        },
        "id": "XMKXF-CUPACm"
      },
      "outputs": [],
      "source": [
        "train_images_normalized = X_train.astype('float32')\n",
        "validation_images_normalized = X_test.astype('float32')\n",
        "\n",
        "# normalize to range 0-1\n",
        "train_images_normalized = train_images_normalized / 255.0\n",
        "validation_images_normalized = validation_images_normalized / 255.0"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T05:55:59.164853Z",
          "start_time": "2023-02-14T05:55:59.052276Z"
        },
        "id": "4GQaEkdfPACm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "f47d9926-62ba-4744-9c2e-a2310e3cfa66"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " conv2d (Conv2D)             (None, 32, 32, 32)        896       \n",
            "                                                                 \n",
            " max_pooling2d (MaxPooling2D  (None, 16, 16, 32)       0         \n",
            " )                                                               \n",
            "                                                                 \n",
            " batch_normalization (BatchN  (None, 16, 16, 32)       128       \n",
            " ormalization)                                                   \n",
            "                                                                 \n",
            " conv2d_1 (Conv2D)           (None, 16, 16, 64)        18496     \n",
            "                                                                 \n",
            " max_pooling2d_1 (MaxPooling  (None, 8, 8, 64)         0         \n",
            " 2D)                                                             \n",
            "                                                                 \n",
            " batch_normalization_1 (Batc  (None, 8, 8, 64)         256       \n",
            " hNormalization)                                                 \n",
            "                                                                 \n",
            " flatten (Flatten)           (None, 4096)              0         \n",
            "                                                                 \n",
            " dense (Dense)               (None, 128)               524416    \n",
            "                                                                 \n",
            " dropout (Dropout)           (None, 128)               0         \n",
            "                                                                 \n",
            " dense_1 (Dense)             (None, 64)                8256      \n",
            "                                                                 \n",
            " batch_normalization_2 (Batc  (None, 64)               256       \n",
            " hNormalization)                                                 \n",
            "                                                                 \n",
            " dense_2 (Dense)             (None, 10)                650       \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 553,354\n",
            "Trainable params: 553,034\n",
            "Non-trainable params: 320\n",
            "_________________________________________________________________\n"
          ]
        }
      ],
      "source": [
        "from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dropout\n",
        "\n",
        "callback = tf.keras.callbacks.EarlyStopping(monitor='loss', patience=3)\n",
        "\n",
        "backend.clear_session()\n",
        "\n",
        "model = keras.models.Sequential()\n",
        "model.add(Conv2D(32, (3, 3), activation='relu', padding=\"same\", input_shape=(32, 32, 3)))\n",
        "# Adding max pooling to reduce the size of output of first conv layer\n",
        "model.add(MaxPooling2D((2, 2), padding = 'same'))\n",
        "model.add(BatchNormalization())\n",
        "\n",
        "model.add(Conv2D(64, (3, 3), activation='relu', padding=\"same\"))\n",
        "model.add(MaxPooling2D((2, 2), padding = 'same'))\n",
        "model.add(BatchNormalization())\n",
        "\n",
        "# flattening the output of the conv layer after max pooling to make it ready for creating dense connections\n",
        "model.add(Flatten())\n",
        "\n",
        "# Adding a fully connected dense layer\n",
        "model.add(Dense(128, activation='relu'))\n",
        "model.add(Dropout(0.3))\n",
        "model.add(Dense(64, activation='relu'))\n",
        "model.add(BatchNormalization())\n",
        "# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem\n",
        "model.add(Dense(10, activation='softmax'))\n",
        "\n",
        "opt=optimizers.Adam(learning_rate=0.001)\n",
        "# Compile model\n",
        "model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])\n",
        "\n",
        "# Generating the summary of the model\n",
        "model.summary()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T05:56:02.801462Z",
          "start_time": "2023-02-14T05:56:02.796806Z"
        },
        "id": "KfgIOuwWPACn"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.utils import to_categorical as one_hot\n",
        "y_train_hot = one_hot(y_train)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T05:58:47.754349Z",
          "start_time": "2023-02-14T05:56:05.374296Z"
        },
        "id": "ClL55GUCPACn",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "aad6f468-d2f7-4551-e4fa-a8aefe54ca5e"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/40\n",
            "250/250 [==============================] - 13s 10ms/step - loss: 1.5270 - accuracy: 0.4551 - val_loss: 4.8429 - val_accuracy: 0.1029\n",
            "Epoch 2/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 1.0958 - accuracy: 0.6103 - val_loss: 4.2353 - val_accuracy: 0.1566\n",
            "Epoch 3/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.8992 - accuracy: 0.6847 - val_loss: 1.0082 - val_accuracy: 0.6424\n",
            "Epoch 4/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.7765 - accuracy: 0.7274 - val_loss: 1.3069 - val_accuracy: 0.5715\n",
            "Epoch 5/40\n",
            "250/250 [==============================] - 2s 10ms/step - loss: 0.6563 - accuracy: 0.7706 - val_loss: 1.1037 - val_accuracy: 0.6349\n",
            "Epoch 6/40\n",
            "250/250 [==============================] - 2s 9ms/step - loss: 0.5609 - accuracy: 0.8039 - val_loss: 1.3450 - val_accuracy: 0.6084\n",
            "Epoch 7/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.4730 - accuracy: 0.8372 - val_loss: 1.0020 - val_accuracy: 0.6775\n",
            "Epoch 8/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.3879 - accuracy: 0.8660 - val_loss: 1.1817 - val_accuracy: 0.6593\n",
            "Epoch 9/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.3365 - accuracy: 0.8822 - val_loss: 1.1573 - val_accuracy: 0.6635\n",
            "Epoch 10/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.2989 - accuracy: 0.8968 - val_loss: 1.0650 - val_accuracy: 0.6891\n",
            "Epoch 11/40\n",
            "250/250 [==============================] - 2s 7ms/step - loss: 0.2596 - accuracy: 0.9109 - val_loss: 1.2138 - val_accuracy: 0.6745\n",
            "Epoch 12/40\n",
            "250/250 [==============================] - 2s 9ms/step - loss: 0.2230 - accuracy: 0.9216 - val_loss: 1.2538 - val_accuracy: 0.6696\n",
            "Epoch 13/40\n",
            "250/250 [==============================] - 2s 9ms/step - loss: 0.2044 - accuracy: 0.9289 - val_loss: 1.2644 - val_accuracy: 0.6864\n",
            "Epoch 14/40\n",
            "250/250 [==============================] - 2s 7ms/step - loss: 0.1969 - accuracy: 0.9323 - val_loss: 1.2589 - val_accuracy: 0.6806\n",
            "Epoch 15/40\n",
            "250/250 [==============================] - 2s 7ms/step - loss: 0.1716 - accuracy: 0.9419 - val_loss: 1.3724 - val_accuracy: 0.6827\n",
            "Epoch 16/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1582 - accuracy: 0.9450 - val_loss: 1.3460 - val_accuracy: 0.6950\n",
            "Epoch 17/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1541 - accuracy: 0.9456 - val_loss: 1.3243 - val_accuracy: 0.6862\n",
            "Epoch 18/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1433 - accuracy: 0.9505 - val_loss: 1.4578 - val_accuracy: 0.6819\n",
            "Epoch 19/40\n",
            "250/250 [==============================] - 2s 10ms/step - loss: 0.1350 - accuracy: 0.9536 - val_loss: 1.4081 - val_accuracy: 0.6925\n",
            "Epoch 20/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1287 - accuracy: 0.9559 - val_loss: 1.7109 - val_accuracy: 0.6571\n",
            "Epoch 21/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1189 - accuracy: 0.9591 - val_loss: 1.5494 - val_accuracy: 0.6877\n",
            "Epoch 22/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1162 - accuracy: 0.9599 - val_loss: 1.8826 - val_accuracy: 0.6327\n",
            "Epoch 23/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1271 - accuracy: 0.9563 - val_loss: 1.6766 - val_accuracy: 0.6768\n",
            "Epoch 24/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1087 - accuracy: 0.9621 - val_loss: 1.7817 - val_accuracy: 0.6593\n",
            "Epoch 25/40\n",
            "250/250 [==============================] - 2s 9ms/step - loss: 0.1099 - accuracy: 0.9618 - val_loss: 1.5679 - val_accuracy: 0.7011\n",
            "Epoch 26/40\n",
            "250/250 [==============================] - 2s 9ms/step - loss: 0.1008 - accuracy: 0.9652 - val_loss: 1.5028 - val_accuracy: 0.6656\n",
            "Epoch 27/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1022 - accuracy: 0.9655 - val_loss: 1.6813 - val_accuracy: 0.6885\n",
            "Epoch 28/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.1021 - accuracy: 0.9658 - val_loss: 1.6412 - val_accuracy: 0.6814\n",
            "Epoch 29/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.0968 - accuracy: 0.9657 - val_loss: 1.6271 - val_accuracy: 0.6845\n",
            "Epoch 30/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.0909 - accuracy: 0.9687 - val_loss: 1.8086 - val_accuracy: 0.6751\n",
            "Epoch 31/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.0930 - accuracy: 0.9669 - val_loss: 1.9218 - val_accuracy: 0.6672\n",
            "Epoch 32/40\n",
            "250/250 [==============================] - 2s 9ms/step - loss: 0.0934 - accuracy: 0.9683 - val_loss: 1.8628 - val_accuracy: 0.6758\n",
            "Epoch 33/40\n",
            "250/250 [==============================] - 3s 10ms/step - loss: 0.0817 - accuracy: 0.9721 - val_loss: 1.7162 - val_accuracy: 0.6805\n",
            "Epoch 34/40\n",
            "250/250 [==============================] - 3s 11ms/step - loss: 0.0852 - accuracy: 0.9703 - val_loss: 1.6800 - val_accuracy: 0.6837\n",
            "Epoch 35/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.0828 - accuracy: 0.9703 - val_loss: 1.7217 - val_accuracy: 0.6812\n",
            "Epoch 36/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.0781 - accuracy: 0.9731 - val_loss: 1.8476 - val_accuracy: 0.6664\n",
            "Epoch 37/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.0810 - accuracy: 0.9723 - val_loss: 1.8368 - val_accuracy: 0.6672\n",
            "Epoch 38/40\n",
            "250/250 [==============================] - 2s 9ms/step - loss: 0.0773 - accuracy: 0.9728 - val_loss: 1.7121 - val_accuracy: 0.6779\n",
            "Epoch 39/40\n",
            "250/250 [==============================] - 2s 9ms/step - loss: 0.0763 - accuracy: 0.9737 - val_loss: 1.6949 - val_accuracy: 0.6837\n",
            "Epoch 40/40\n",
            "250/250 [==============================] - 2s 8ms/step - loss: 0.0708 - accuracy: 0.9763 - val_loss: 1.6934 - val_accuracy: 0.6956\n"
          ]
        }
      ],
      "source": [
        "history=model.fit(X_train, y_train_hot,\n",
        "          validation_split=0.2,\n",
        "          epochs=40,\n",
        "          batch_size=128,\n",
        "          callbacks=[callback],\n",
        "          shuffle=True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "ExecuteTime": {
          "end_time": "2023-02-14T05:58:54.146941Z",
          "start_time": "2023-02-14T05:58:52.332928Z"
        },
        "id": "MM1fGdN8PACn",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "421860fd-65a4-4d3b-eb1d-595831a036ac"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "313/313 [==============================] - 1s 2ms/step\n"
          ]
        }
      ],
      "source": [
        "y_pred = model.predict(validation_images_normalized)\n",
        "predicted_classes=np.argmax(y_pred,axis=1)"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "y_test_hot  = one_hot(y_test)\n",
        "accuracy = model.evaluate(X_test, y_test_hot, verbose=2)"
      ],
      "metadata": {
        "id": "TTbeCxR7Mscv",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "26336aee-4fd0-462c-ed44-3ab39ef9ea2c"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "313/313 - 1s - loss: 1.6122 - accuracy: 0.7039 - 779ms/epoch - 2ms/step\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Plot training & validation accuracy values\n",
        "plt.plot(history.history['accuracy'])\n",
        "plt.plot(history.history['val_accuracy'])\n",
        "\n",
        "plt.title('Model accuracy')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Test'], loc='upper left')\n",
        "plt.show()"
      ],
      "metadata": {
        "id": "JpM2B7fEDyAJ",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 472
        },
        "outputId": "8542c4e5-790e-4217-836f-a82b09c141e4"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "There is significant increase in accuracy with CNN :-) 🥲\n",
        "\n",
        "---\n",
        "\n"
      ],
      "metadata": {
        "id": "4Ska0jk7Ui7I"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "m0RSsoe9PACn"
      },
      "source": [
        "## Transfer learning with Xception"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xkT2QKi5PACo"
      },
      "source": [
        "Xception is the extreme version of the Inception model with depthwise separable convolutions and has been created by Google ( [Xception reference](https://https://towardsdatascience.com/review-xception-with-depthwise-separable-convolution-better-than-inception-v3-image-dc967dd42568) ) and a combination of depthwise and pointwise convolutions. This model has outperformed VGG, Inception and Resnet in training the Imagenet dataset."
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Clearing backend\n",
        "from tensorflow.keras import backend\n",
        "backend.clear_session()\n",
        "\n",
        "# Fixing the seed for random number generators\n",
        "import random\n",
        "np.random.seed(42)\n",
        "random.seed(42)\n",
        "tf.random.set_seed(42)\n",
        "\n",
        "from tensorflow.keras.models import Model\n",
        "from keras.applications.xception import Xception\n",
        "\n",
        "#xception_model = Xception(weights='imagenet', include_top = False, input_shape = (128,128,3))\n",
        "xception_model = Xception(weights='imagenet', include_top = False, input_shape = (128,128,3))\n",
        "xception_model.summary()\n"
      ],
      "metadata": {
        "id": "1JwhGoq0N2Fp",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "32a9d63c-ebb2-4c7c-b432-19443bec78fa"
      },
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/xception/xception_weights_tf_dim_ordering_tf_kernels_notop.h5\n",
            "83683744/83683744 [==============================] - 5s 0us/step\n",
            "Model: \"xception\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_1 (InputLayer)           [(None, 128, 128, 3  0           []                               \n",
            "                                )]                                                                \n",
            "                                                                                                  \n",
            " block1_conv1 (Conv2D)          (None, 63, 63, 32)   864         ['input_1[0][0]']                \n",
            "                                                                                                  \n",
            " block1_conv1_bn (BatchNormaliz  (None, 63, 63, 32)  128         ['block1_conv1[0][0]']           \n",
            " ation)                                                                                           \n",
            "                                                                                                  \n",
            " block1_conv1_act (Activation)  (None, 63, 63, 32)   0           ['block1_conv1_bn[0][0]']        \n",
            "                                                                                                  \n",
            " block1_conv2 (Conv2D)          (None, 61, 61, 64)   18432       ['block1_conv1_act[0][0]']       \n",
            "                                                                                                  \n",
            " block1_conv2_bn (BatchNormaliz  (None, 61, 61, 64)  256         ['block1_conv2[0][0]']           \n",
            " ation)                                                                                           \n",
            "                                                                                                  \n",
            " block1_conv2_act (Activation)  (None, 61, 61, 64)   0           ['block1_conv2_bn[0][0]']        \n",
            "                                                                                                  \n",
            " block2_sepconv1 (SeparableConv  (None, 61, 61, 128)  8768       ['block1_conv2_act[0][0]']       \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block2_sepconv1_bn (BatchNorma  (None, 61, 61, 128)  512        ['block2_sepconv1[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block2_sepconv2_act (Activatio  (None, 61, 61, 128)  0          ['block2_sepconv1_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block2_sepconv2 (SeparableConv  (None, 61, 61, 128)  17536      ['block2_sepconv2_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block2_sepconv2_bn (BatchNorma  (None, 61, 61, 128)  512        ['block2_sepconv2[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " conv2d (Conv2D)                (None, 31, 31, 128)  8192        ['block1_conv2_act[0][0]']       \n",
            "                                                                                                  \n",
            " block2_pool (MaxPooling2D)     (None, 31, 31, 128)  0           ['block2_sepconv2_bn[0][0]']     \n",
            "                                                                                                  \n",
            " batch_normalization (BatchNorm  (None, 31, 31, 128)  512        ['conv2d[0][0]']                 \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " add (Add)                      (None, 31, 31, 128)  0           ['block2_pool[0][0]',            \n",
            "                                                                  'batch_normalization[0][0]']    \n",
            "                                                                                                  \n",
            " block3_sepconv1_act (Activatio  (None, 31, 31, 128)  0          ['add[0][0]']                    \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block3_sepconv1 (SeparableConv  (None, 31, 31, 256)  33920      ['block3_sepconv1_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block3_sepconv1_bn (BatchNorma  (None, 31, 31, 256)  1024       ['block3_sepconv1[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block3_sepconv2_act (Activatio  (None, 31, 31, 256)  0          ['block3_sepconv1_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block3_sepconv2 (SeparableConv  (None, 31, 31, 256)  67840      ['block3_sepconv2_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block3_sepconv2_bn (BatchNorma  (None, 31, 31, 256)  1024       ['block3_sepconv2[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " conv2d_1 (Conv2D)              (None, 16, 16, 256)  32768       ['add[0][0]']                    \n",
            "                                                                                                  \n",
            " block3_pool (MaxPooling2D)     (None, 16, 16, 256)  0           ['block3_sepconv2_bn[0][0]']     \n",
            "                                                                                                  \n",
            " batch_normalization_1 (BatchNo  (None, 16, 16, 256)  1024       ['conv2d_1[0][0]']               \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " add_1 (Add)                    (None, 16, 16, 256)  0           ['block3_pool[0][0]',            \n",
            "                                                                  'batch_normalization_1[0][0]']  \n",
            "                                                                                                  \n",
            " block4_sepconv1_act (Activatio  (None, 16, 16, 256)  0          ['add_1[0][0]']                  \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block4_sepconv1 (SeparableConv  (None, 16, 16, 728)  188672     ['block4_sepconv1_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block4_sepconv1_bn (BatchNorma  (None, 16, 16, 728)  2912       ['block4_sepconv1[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block4_sepconv2_act (Activatio  (None, 16, 16, 728)  0          ['block4_sepconv1_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block4_sepconv2 (SeparableConv  (None, 16, 16, 728)  536536     ['block4_sepconv2_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block4_sepconv2_bn (BatchNorma  (None, 16, 16, 728)  2912       ['block4_sepconv2[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " conv2d_2 (Conv2D)              (None, 8, 8, 728)    186368      ['add_1[0][0]']                  \n",
            "                                                                                                  \n",
            " block4_pool (MaxPooling2D)     (None, 8, 8, 728)    0           ['block4_sepconv2_bn[0][0]']     \n",
            "                                                                                                  \n",
            " batch_normalization_2 (BatchNo  (None, 8, 8, 728)   2912        ['conv2d_2[0][0]']               \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " add_2 (Add)                    (None, 8, 8, 728)    0           ['block4_pool[0][0]',            \n",
            "                                                                  'batch_normalization_2[0][0]']  \n",
            "                                                                                                  \n",
            " block5_sepconv1_act (Activatio  (None, 8, 8, 728)   0           ['add_2[0][0]']                  \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block5_sepconv1 (SeparableConv  (None, 8, 8, 728)   536536      ['block5_sepconv1_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block5_sepconv1_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block5_sepconv1[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block5_sepconv2_act (Activatio  (None, 8, 8, 728)   0           ['block5_sepconv1_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block5_sepconv2 (SeparableConv  (None, 8, 8, 728)   536536      ['block5_sepconv2_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block5_sepconv2_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block5_sepconv2[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block5_sepconv3_act (Activatio  (None, 8, 8, 728)   0           ['block5_sepconv2_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block5_sepconv3 (SeparableConv  (None, 8, 8, 728)   536536      ['block5_sepconv3_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block5_sepconv3_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block5_sepconv3[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " add_3 (Add)                    (None, 8, 8, 728)    0           ['block5_sepconv3_bn[0][0]',     \n",
            "                                                                  'add_2[0][0]']                  \n",
            "                                                                                                  \n",
            " block6_sepconv1_act (Activatio  (None, 8, 8, 728)   0           ['add_3[0][0]']                  \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block6_sepconv1 (SeparableConv  (None, 8, 8, 728)   536536      ['block6_sepconv1_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block6_sepconv1_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block6_sepconv1[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block6_sepconv2_act (Activatio  (None, 8, 8, 728)   0           ['block6_sepconv1_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block6_sepconv2 (SeparableConv  (None, 8, 8, 728)   536536      ['block6_sepconv2_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block6_sepconv2_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block6_sepconv2[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block6_sepconv3_act (Activatio  (None, 8, 8, 728)   0           ['block6_sepconv2_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block6_sepconv3 (SeparableConv  (None, 8, 8, 728)   536536      ['block6_sepconv3_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block6_sepconv3_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block6_sepconv3[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " add_4 (Add)                    (None, 8, 8, 728)    0           ['block6_sepconv3_bn[0][0]',     \n",
            "                                                                  'add_3[0][0]']                  \n",
            "                                                                                                  \n",
            " block7_sepconv1_act (Activatio  (None, 8, 8, 728)   0           ['add_4[0][0]']                  \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block7_sepconv1 (SeparableConv  (None, 8, 8, 728)   536536      ['block7_sepconv1_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block7_sepconv1_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block7_sepconv1[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block7_sepconv2_act (Activatio  (None, 8, 8, 728)   0           ['block7_sepconv1_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block7_sepconv2 (SeparableConv  (None, 8, 8, 728)   536536      ['block7_sepconv2_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block7_sepconv2_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block7_sepconv2[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block7_sepconv3_act (Activatio  (None, 8, 8, 728)   0           ['block7_sepconv2_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block7_sepconv3 (SeparableConv  (None, 8, 8, 728)   536536      ['block7_sepconv3_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block7_sepconv3_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block7_sepconv3[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " add_5 (Add)                    (None, 8, 8, 728)    0           ['block7_sepconv3_bn[0][0]',     \n",
            "                                                                  'add_4[0][0]']                  \n",
            "                                                                                                  \n",
            " block8_sepconv1_act (Activatio  (None, 8, 8, 728)   0           ['add_5[0][0]']                  \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block8_sepconv1 (SeparableConv  (None, 8, 8, 728)   536536      ['block8_sepconv1_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block8_sepconv1_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block8_sepconv1[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block8_sepconv2_act (Activatio  (None, 8, 8, 728)   0           ['block8_sepconv1_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block8_sepconv2 (SeparableConv  (None, 8, 8, 728)   536536      ['block8_sepconv2_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block8_sepconv2_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block8_sepconv2[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block8_sepconv3_act (Activatio  (None, 8, 8, 728)   0           ['block8_sepconv2_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block8_sepconv3 (SeparableConv  (None, 8, 8, 728)   536536      ['block8_sepconv3_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block8_sepconv3_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block8_sepconv3[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " add_6 (Add)                    (None, 8, 8, 728)    0           ['block8_sepconv3_bn[0][0]',     \n",
            "                                                                  'add_5[0][0]']                  \n",
            "                                                                                                  \n",
            " block9_sepconv1_act (Activatio  (None, 8, 8, 728)   0           ['add_6[0][0]']                  \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block9_sepconv1 (SeparableConv  (None, 8, 8, 728)   536536      ['block9_sepconv1_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block9_sepconv1_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block9_sepconv1[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block9_sepconv2_act (Activatio  (None, 8, 8, 728)   0           ['block9_sepconv1_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block9_sepconv2 (SeparableConv  (None, 8, 8, 728)   536536      ['block9_sepconv2_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block9_sepconv2_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block9_sepconv2[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " block9_sepconv3_act (Activatio  (None, 8, 8, 728)   0           ['block9_sepconv2_bn[0][0]']     \n",
            " n)                                                                                               \n",
            "                                                                                                  \n",
            " block9_sepconv3 (SeparableConv  (None, 8, 8, 728)   536536      ['block9_sepconv3_act[0][0]']    \n",
            " 2D)                                                                                              \n",
            "                                                                                                  \n",
            " block9_sepconv3_bn (BatchNorma  (None, 8, 8, 728)   2912        ['block9_sepconv3[0][0]']        \n",
            " lization)                                                                                        \n",
            "                                                                                                  \n",
            " add_7 (Add)                    (None, 8, 8, 728)    0           ['block9_sepconv3_bn[0][0]',     \n",
            "                                                                  'add_6[0][0]']                  \n",
            "                                                                                                  \n",
            " block10_sepconv1_act (Activati  (None, 8, 8, 728)   0           ['add_7[0][0]']                  \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block10_sepconv1 (SeparableCon  (None, 8, 8, 728)   536536      ['block10_sepconv1_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block10_sepconv1_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block10_sepconv1[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block10_sepconv2_act (Activati  (None, 8, 8, 728)   0           ['block10_sepconv1_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block10_sepconv2 (SeparableCon  (None, 8, 8, 728)   536536      ['block10_sepconv2_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block10_sepconv2_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block10_sepconv2[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block10_sepconv3_act (Activati  (None, 8, 8, 728)   0           ['block10_sepconv2_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block10_sepconv3 (SeparableCon  (None, 8, 8, 728)   536536      ['block10_sepconv3_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block10_sepconv3_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block10_sepconv3[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " add_8 (Add)                    (None, 8, 8, 728)    0           ['block10_sepconv3_bn[0][0]',    \n",
            "                                                                  'add_7[0][0]']                  \n",
            "                                                                                                  \n",
            " block11_sepconv1_act (Activati  (None, 8, 8, 728)   0           ['add_8[0][0]']                  \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block11_sepconv1 (SeparableCon  (None, 8, 8, 728)   536536      ['block11_sepconv1_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block11_sepconv1_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block11_sepconv1[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block11_sepconv2_act (Activati  (None, 8, 8, 728)   0           ['block11_sepconv1_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block11_sepconv2 (SeparableCon  (None, 8, 8, 728)   536536      ['block11_sepconv2_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block11_sepconv2_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block11_sepconv2[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block11_sepconv3_act (Activati  (None, 8, 8, 728)   0           ['block11_sepconv2_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block11_sepconv3 (SeparableCon  (None, 8, 8, 728)   536536      ['block11_sepconv3_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block11_sepconv3_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block11_sepconv3[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " add_9 (Add)                    (None, 8, 8, 728)    0           ['block11_sepconv3_bn[0][0]',    \n",
            "                                                                  'add_8[0][0]']                  \n",
            "                                                                                                  \n",
            " block12_sepconv1_act (Activati  (None, 8, 8, 728)   0           ['add_9[0][0]']                  \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block12_sepconv1 (SeparableCon  (None, 8, 8, 728)   536536      ['block12_sepconv1_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block12_sepconv1_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block12_sepconv1[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block12_sepconv2_act (Activati  (None, 8, 8, 728)   0           ['block12_sepconv1_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block12_sepconv2 (SeparableCon  (None, 8, 8, 728)   536536      ['block12_sepconv2_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block12_sepconv2_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block12_sepconv2[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block12_sepconv3_act (Activati  (None, 8, 8, 728)   0           ['block12_sepconv2_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block12_sepconv3 (SeparableCon  (None, 8, 8, 728)   536536      ['block12_sepconv3_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block12_sepconv3_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block12_sepconv3[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " add_10 (Add)                   (None, 8, 8, 728)    0           ['block12_sepconv3_bn[0][0]',    \n",
            "                                                                  'add_9[0][0]']                  \n",
            "                                                                                                  \n",
            " block13_sepconv1_act (Activati  (None, 8, 8, 728)   0           ['add_10[0][0]']                 \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block13_sepconv1 (SeparableCon  (None, 8, 8, 728)   536536      ['block13_sepconv1_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block13_sepconv1_bn (BatchNorm  (None, 8, 8, 728)   2912        ['block13_sepconv1[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block13_sepconv2_act (Activati  (None, 8, 8, 728)   0           ['block13_sepconv1_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block13_sepconv2 (SeparableCon  (None, 8, 8, 1024)  752024      ['block13_sepconv2_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block13_sepconv2_bn (BatchNorm  (None, 8, 8, 1024)  4096        ['block13_sepconv2[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " conv2d_3 (Conv2D)              (None, 4, 4, 1024)   745472      ['add_10[0][0]']                 \n",
            "                                                                                                  \n",
            " block13_pool (MaxPooling2D)    (None, 4, 4, 1024)   0           ['block13_sepconv2_bn[0][0]']    \n",
            "                                                                                                  \n",
            " batch_normalization_3 (BatchNo  (None, 4, 4, 1024)  4096        ['conv2d_3[0][0]']               \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " add_11 (Add)                   (None, 4, 4, 1024)   0           ['block13_pool[0][0]',           \n",
            "                                                                  'batch_normalization_3[0][0]']  \n",
            "                                                                                                  \n",
            " block14_sepconv1 (SeparableCon  (None, 4, 4, 1536)  1582080     ['add_11[0][0]']                 \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block14_sepconv1_bn (BatchNorm  (None, 4, 4, 1536)  6144        ['block14_sepconv1[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block14_sepconv1_act (Activati  (None, 4, 4, 1536)  0           ['block14_sepconv1_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            " block14_sepconv2 (SeparableCon  (None, 4, 4, 2048)  3159552     ['block14_sepconv1_act[0][0]']   \n",
            " v2D)                                                                                             \n",
            "                                                                                                  \n",
            " block14_sepconv2_bn (BatchNorm  (None, 4, 4, 2048)  8192        ['block14_sepconv2[0][0]']       \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " block14_sepconv2_act (Activati  (None, 4, 4, 2048)  0           ['block14_sepconv2_bn[0][0]']    \n",
            " on)                                                                                              \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 20,861,480\n",
            "Trainable params: 20,806,952\n",
            "Non-trainable params: 54,528\n",
            "__________________________________________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Making all the layers of the VGG model non-trainable. i.e. freezing them\n",
        "for layer in xception_model.layers:\n",
        "    layer.trainable = False"
      ],
      "metadata": {
        "id": "a9vudIvnJG8f"
      },
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from tensorflow.keras.layers import Conv2D, MaxPooling2D,Dropout,Resizing\n",
        "from tensorflow.keras import backend\n",
        "from tensorflow.keras.layers import Dense, Input, Dropout,BatchNormalization,Flatten\n",
        "from tensorflow.keras import optimizers\n",
        "from tensorflow.keras import losses\n",
        "backend.clear_session()\n",
        "import gc\n",
        "\n",
        "callback = tf.keras.callbacks.EarlyStopping(monitor='loss', patience=3)"
      ],
      "metadata": {
        "id": "Kpz8DAemSKRy"
      },
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from tensorflow.keras.layers import Conv2D, MaxPooling2D,Dropout,Resizing\n",
        "\n",
        "backend.clear_session()\n",
        "transfer_learning_model = keras.models.Sequential()\n",
        "\n",
        "transfer_learning_model.add(Resizing(128, 128, interpolation=\"bilinear\", crop_to_aspect_ratio=False))\n",
        "\n",
        "# Adding the convolutional part of the VGG16 model from above\n",
        "transfer_learning_model.add(xception_model)\n",
        "# Flattening the output of the VGG16 model because it is from a convolutional layer\n",
        "transfer_learning_model.add(Flatten())\n",
        "\n",
        "# Adding a dense output layer\n",
        "transfer_learning_model.add(Dense(128, activation='relu'))\n",
        "transfer_learning_model.add(BatchNormalization())\n",
        "transfer_learning_model.add(Dense(32, activation='relu'))\n",
        "# transfer_learning_model.add(Dropout(0.2))\n",
        "transfer_learning_model.add(Dense(10, activation='softmax'))\n",
        "\n",
        "# Compile model\n",
        "transfer_learning_model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])\n"
      ],
      "metadata": {
        "id": "P-WbImOUOH20"
      },
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "transfer_learning_model.build((None,32,32,3))\n",
        "transfer_learning_model.summary()"
      ],
      "metadata": {
        "id": "wfTIEIO_RNgq",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d1da2cb0-139e-4d74-a3b2-2819faf725cc"
      },
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " resizing (Resizing)         (None, 128, 128, 3)       0         \n",
            "                                                                 \n",
            " xception (Functional)       (None, 4, 4, 2048)        20861480  \n",
            "                                                                 \n",
            " flatten (Flatten)           (None, 32768)             0         \n",
            "                                                                 \n",
            " dense (Dense)               (None, 128)               4194432   \n",
            "                                                                 \n",
            " batch_normalization (BatchN  (None, 128)              512       \n",
            " ormalization)                                                   \n",
            "                                                                 \n",
            " dense_1 (Dense)             (None, 32)                4128      \n",
            "                                                                 \n",
            " dense_2 (Dense)             (None, 10)                330       \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 25,060,882\n",
            "Trainable params: 4,199,146\n",
            "Non-trainable params: 20,861,736\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from tensorflow.keras.utils import to_categorical as one_hot\n",
        "\n",
        "train_images_normalized = X_train.astype('float32')\n",
        "validation_images_normalized = X_test.astype('float32')\n",
        "\n",
        "# normalize to range 0-1\n",
        "train_images_normalized = train_images_normalized / 255.0\n",
        "validation_images_normalized = validation_images_normalized / 255.0\n",
        "\n",
        "y_train_hot = one_hot(y_train)\n",
        "# y_test_hot  = one_hot(y_test)"
      ],
      "metadata": {
        "id": "jcV3lv9PQiX9"
      },
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "history = transfer_learning_model.fit(X_train, y_train_hot,\n",
        "          validation_split = 0.2,\n",
        "          epochs = 20,\n",
        "          batch_size = 128,\n",
        "          callbacks = [callback])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Hwa-AXnEQzgg",
        "outputId": "7d1c87f3-52f6-48c6-fc51-efe8af492b9c"
      },
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/20\n",
            "250/250 [==============================] - 64s 190ms/step - loss: 0.7393 - accuracy: 0.7570 - val_loss: 0.6095 - val_accuracy: 0.7910\n",
            "Epoch 2/20\n",
            "250/250 [==============================] - 48s 192ms/step - loss: 0.5010 - accuracy: 0.8292 - val_loss: 0.5976 - val_accuracy: 0.7946\n",
            "Epoch 3/20\n",
            "250/250 [==============================] - 48s 190ms/step - loss: 0.3927 - accuracy: 0.8656 - val_loss: 0.6238 - val_accuracy: 0.7966\n",
            "Epoch 4/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.3038 - accuracy: 0.8954 - val_loss: 0.6510 - val_accuracy: 0.7900\n",
            "Epoch 5/20\n",
            "250/250 [==============================] - 47s 187ms/step - loss: 0.2275 - accuracy: 0.9227 - val_loss: 0.7000 - val_accuracy: 0.7890\n",
            "Epoch 6/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.1778 - accuracy: 0.9397 - val_loss: 0.7856 - val_accuracy: 0.7818\n",
            "Epoch 7/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.1377 - accuracy: 0.9545 - val_loss: 0.8517 - val_accuracy: 0.7797\n",
            "Epoch 8/20\n",
            "250/250 [==============================] - 47s 187ms/step - loss: 0.1123 - accuracy: 0.9617 - val_loss: 0.9081 - val_accuracy: 0.7789\n",
            "Epoch 9/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0835 - accuracy: 0.9726 - val_loss: 0.8994 - val_accuracy: 0.7843\n",
            "Epoch 10/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0774 - accuracy: 0.9747 - val_loss: 0.9918 - val_accuracy: 0.7766\n",
            "Epoch 11/20\n",
            "250/250 [==============================] - 47s 187ms/step - loss: 0.0703 - accuracy: 0.9770 - val_loss: 1.0749 - val_accuracy: 0.7805\n",
            "Epoch 12/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0505 - accuracy: 0.9834 - val_loss: 1.0755 - val_accuracy: 0.7822\n",
            "Epoch 13/20\n",
            "250/250 [==============================] - 47s 187ms/step - loss: 0.0460 - accuracy: 0.9854 - val_loss: 1.1292 - val_accuracy: 0.7939\n",
            "Epoch 14/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0581 - accuracy: 0.9804 - val_loss: 1.0960 - val_accuracy: 0.7845\n",
            "Epoch 15/20\n",
            "250/250 [==============================] - 47s 187ms/step - loss: 0.0451 - accuracy: 0.9850 - val_loss: 1.1816 - val_accuracy: 0.7872\n",
            "Epoch 16/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0428 - accuracy: 0.9859 - val_loss: 1.1517 - val_accuracy: 0.7893\n",
            "Epoch 17/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0364 - accuracy: 0.9880 - val_loss: 1.2450 - val_accuracy: 0.7803\n",
            "Epoch 18/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0388 - accuracy: 0.9871 - val_loss: 1.2170 - val_accuracy: 0.7878\n",
            "Epoch 19/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0348 - accuracy: 0.9876 - val_loss: 1.2058 - val_accuracy: 0.7885\n",
            "Epoch 20/20\n",
            "250/250 [==============================] - 48s 191ms/step - loss: 0.0338 - accuracy: 0.9887 - val_loss: 1.3047 - val_accuracy: 0.7857\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "y_test_hot  = one_hot(y_test)\n",
        "accuracy = transfer_learning_model.evaluate(X_test, y_test_hot, verbose=2)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xBYpgg41E4ki",
        "outputId": "d9fd7130-368d-4a56-e78f-480aa37f87e1"
      },
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "313/313 - 13s - loss: 1.2628 - accuracy: 0.7892 - 13s/epoch - 42ms/step\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Plot training & validation accuracy values\n",
        "plt.plot(history.history['accuracy'])\n",
        "plt.plot(history.history['val_accuracy'])\n",
        "\n",
        "plt.title('Model accuracy')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Test'], loc='upper left')\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 472
        },
        "id": "gYX0fC4e1vZ7",
        "outputId": "c8c425da-94df-456b-83d7-02c16752895a"
      },
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Xception emerged as the clear winner with exceptional accuracy—forgive the pun, we couldn't resist! 😄\n",
        "\n",
        "The journey allowed us to gain a comprehensive understanding of the strengths and weaknesses of each model. While KNN provided a straightforward and interpretable approach, its limitations became evident as the complexity of the task increased. The single-hidden-layer neural network showed promise but lacked the ability to capture the intricate patterns present in the images. The CNN, with its convolutional layers and feature extraction prowess, demonstrated considerable improvements over the previous models.\n",
        "\n",
        "However, it was the Xception model, leveraging transfer learning and pretrained weights from a vast dataset, that truly set a new benchmark. Its ability to extract complex hierarchical features and generalize well to our specific classification task was the key to its exceptional accuracy."
      ],
      "metadata": {
        "id": "pshnD_2L2_m6"
      }
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.9.7"
    },
    "nbclean": true,
    "colab": {
      "provenance": [],
      "gpuType": "T4",
      "toc_visible": true,
      "include_colab_link": true
    },
    "accelerator": "GPU"
  },
  "nbformat": 4,
  "nbformat_minor": 0
}