220079842_assignment2_solution.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "220079842_assignment2_solution.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "twFQbltnm8da",
        "zdHwmgwOpEfx",
        "cfTL_5kPsl4j",
        "prctXU4BswKK",
        "UaHLKDLas_dF",
        "IaD5oqj3lhuI",
        "adN9Tq-6lyG-",
        "9NTjVIUkmv7S",
        "AH6ZBiECzS75",
        "C8cKfAOjpn7c"
      ],
      "machine_shape": "hm"
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wjif_5FsdOFz"
      },
      "source": [
        "# SIT744 Assignment 2: Transfer learning and Efficient Training of Deep Learning Models \n",
        "\n",
        "<div class=\"alert-info\">\n",
        "    <p>Due: <strong>8:00pm 17 May 2021</strong>  (Monday)</p>\n",
        "\n",
        "This is an <strong>individual</strong> assignment. It contributes <strong>45%</strong> to your final mark. Read the assignment instruction carefully.\n",
        "\n",
        "<h2> What to submit </h2>\n",
        "\n",
        "<p>\n",
        "This assignment is to be completed individually and submitted to CloudDeakin. <strong>By the due date, you are required to submit the following files to the corresponding Assignment (Dropbox) in CloudDeakin</strong>:\n",
        "\n",
        "<ol>\n",
        "<li>\t<strong>[YourID]_assignment2_solution.ipynp</strong>:  This is your Python notebook solution source file. </li>\n",
        "<li>\t<strong>[YourID]_assingment2_output.html</strong>: This is the output of your Python notebook solution <emph>exported</emph> in HTML format.</li>\n",
        "<li>\tExtra files needed to complete your assignment, if any (e.g., images used in your answers).</li>\n",
        "</ol>\n",
        "</p>\n",
        "\n",
        "<p>\n",
        "For example, if your student ID is: 123456, you will then need to submit the following files:\n",
        "<ul>\n",
        "<li> 123456_assignment2_solution.ipynp </li>\n",
        "<li> 123456_assignment2_output.html</li>\n",
        "</ul>\n",
        "</p>\n",
        "\n",
        "<h2> Warning </h2>\n",
        "\n",
        "Some components of this assignment may involve heavy computation that runs for a long duration. Please start early to avoid missing the assignment due date.\n",
        "\n",
        "<h2> Marking criteria </h2>\n",
        "\n",
        "<p>\n",
        "Your submission will be marked using the following criteria.\n",
        "\n",
        "<ul>\n",
        "<li> Showing good effort through completed tasks.</li>\n",
        "<li> Applying deep learning theory to design suitable deep learning solutions for the tasks.</li>\n",
        "<li> Critically evaluating and reflecting on the pros and cons of various design decisions.</li>\n",
        "<li> Demonstrating creativity and resourcefulness in providing unique individual solutions.</li>\n",
        "<li> Showing attention to details through a good quality assignment report.</li>\n",
        "</ul>\n",
        "</p>\n",
        "\n",
        "<p>\n",
        "Indicative weights of various tasks are provided, but the assignment will be marked by the overall quality per the above criteria.\n",
        "</p>\n",
        "</div>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "twFQbltnm8da"
      },
      "source": [
        "## Assignment objective\n",
        "\n",
        "This assignment is to feedback on your learning in deep learning theory and its application to  data analytics or artificial intelligence problems.  \n",
        "\n",
        "It builds on Assignment 1 but requires a higher level of mastery of deep learning theory and programming/engineering skills. In particular, you will experience training a much deeper network on a large-scale dataset. You will encounter  practical issues that help you consolidate textbook learning. "
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3ITc1hw_o7qV"
      },
      "source": [
        "## Task 1 Solving Fashion-MNIST with Convolutional Neural Networks\n",
        "\n",
        "*(weight ~15%)*\n",
        "\n",
        "In Assignment 1, you tackled the image classification problem in Fashion-MNIST. There, you used a Densely Connected Neural Network. You should now know that is not an optimal model architecture for the problem. In Assignment 2, you will apply the best practices of deep-learning computer vision to achieve better image classification performance."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zdHwmgwOpEfx"
      },
      "source": [
        "### Task 1.1 Revisit Fashion-MNIST classification with DNN\n",
        "\n",
        "*(weight ~1%)*\n",
        "\n",
        "Review your Assignment 1 solution, and reproduce the experiment here. Try to improve the model without changing the model architecture.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Dlxzmo4mOZCv",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "fd4f6ab5-6adb-40c9-fd43-a7aff4e440b3"
      },
      "source": [
        "# Producing the Assignment-1, DNN \n",
        "\n",
        "import tensorflow as tf\n",
        "import tensorflow_datasets\n",
        "import math\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from keras.layers.normalization import BatchNormalization\n",
        "from keras.regularizers import l2\n",
        "from keras.layers import Dropout\n",
        "from keras.layers import Activation\n",
        "from keras.optimizers import SGD\n",
        "from keras.optimizers import RMSprop\n",
        "from keras.callbacks import LearningRateScheduler\n",
        "from keras.callbacks import History\n",
        "\n",
        "\n",
        "# Loading the data using tensorflow\n",
        "fashion_mnist = tf.keras.datasets.fashion_mnist\n",
        "\n",
        "(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()\n",
        "\n",
        "# class names\n",
        "class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',\n",
        "               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz\n",
            "32768/29515 [=================================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz\n",
            "26427392/26421880 [==============================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz\n",
            "8192/5148 [===============================================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz\n",
            "4423680/4422102 [==============================] - 0s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RLSzup0tO0kD"
      },
      "source": [
        "# train, test and validation splits\n",
        "\n",
        "train_images = train_images / 255.0\n",
        "test_images = test_images / 255.0\n",
        "\n",
        "# defining batch size\n",
        "BATCH_SIZE = 512\n",
        "\n",
        "# training dataset\n",
        "fmnist_train_ds = tf.data.Dataset.from_tensor_slices((train_images, train_labels))\n",
        "\n",
        "\n",
        "# test dataset\n",
        "fmnist_test_ds = tf.data.Dataset.from_tensor_slices((test_images, test_labels))\n",
        "fmnist_test_ds = fmnist_test_ds.shuffle(5000).batch(BATCH_SIZE)\n",
        "\n",
        "# picking out the validation set from the training dataset\n",
        "fmnist_train = fmnist_train_ds.take(round(len(train_labels)*0.7))\n",
        "fmnist_val = fmnist_train_ds.skip(round(len(train_labels)*0.7))\n",
        "\n",
        "from sklearn.model_selection import train_test_split\n",
        "\n",
        "\n",
        "X_train, X_val, y_train, y_val  = train_test_split(train_images, train_labels, test_size=0.25, random_state=1) \n",
        "\n",
        "#fmnist_train_ds = fmnist_train_ds.values.reshape(-1,28,28,1)\n",
        "#X_val = X_val.values.reshape(-1,28,28,1)\n",
        "\n",
        "\n",
        "# batch size should be in powers of 2 and the larger the better, provided that the batch fits into the (GPU) memory;\n",
        "# small batches can offer a regularizing effect, perhaps due to the noise they add to the learning process.\n",
        "# larger batches provide a more accurate estimate of the gradient.\n",
        "# the batch size here is 512.\n",
        "# train-data shuffling ensures that the model gets to see all permutauions of input in every batch."
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G7Wcy86mPZ4Q"
      },
      "source": [
        "# adding dropout, batch normalisation, changing the learning rate\n",
        "\n",
        "def optimiser_sgd(learning_rate):\n",
        "  learning_rate = learning_rate\n",
        "  decay_rate = learning_rate / 50\n",
        "  momentum = 0.8\n",
        "\n",
        "  sgd = SGD(lr=learning_rate, momentum=momentum, decay=decay_rate, nesterov=False)\n",
        "  return sgd"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tOCwK2geP3Nb"
      },
      "source": [
        "\n",
        "decay_rate = 0.001\n",
        "learning_rate = 0.001\n",
        "\n",
        "def exp_decay(epoch):\n",
        "    lrate = learning_rate * np.exp(-decay_rate*epoch)\n",
        "    return lrate\n",
        "    \n",
        "# learning schedule callback\n",
        "loss_history = History()\n",
        "lr_rate = LearningRateScheduler(exp_decay)\n",
        "callbacks_list = [loss_history, lr_rate]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7zFBg2u1O0rz"
      },
      "source": [
        "# Different layers are appropriate for different tensor formats and different types of data processing. \n",
        "# Simple vector data, stored in 2D tensors of shape (samples, features), is often processed by densely connected layers.\n",
        "\n",
        "# single-label, multiclass classification problem\n",
        "\n",
        "model_d = tf.keras.Sequential([\n",
        "    tf.keras.layers.Flatten(input_shape=(28, 28, 1)),\n",
        "    tf.keras.layers.Dense(128, kernel_regularizer=l2(0.01), bias_regularizer=l2(0.01), activation=tf.nn.relu),\n",
        "    tf.keras.layers.BatchNormalization(),\n",
        "    tf.keras.layers.Dropout(0.1),\n",
        "    tf.keras.layers.Dense(256, kernel_regularizer=l2(0.001), bias_regularizer=l2(0.001), activation=tf.nn.relu),\n",
        "    tf.keras.layers.Dense(10, activation=tf.nn.softmax)\n",
        "])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "stSMER3BPn65"
      },
      "source": [
        "model_d.compile(optimizer=optimiser_sgd(0.001),\n",
        "              loss=tf.keras.losses.SparseCategoricalCrossentropy(),\n",
        "              metrics=['accuracy'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 703
        },
        "id": "2jepHqS8Pv3a",
        "outputId": "cea9fc40-51d2-4b42-81b2-440b500f4dda"
      },
      "source": [
        "regularised_layers_model_history = model_d.fit(train_images, \n",
        "                                     train_labels,\n",
        "                                     validation_split=0.3,\n",
        "                                     batch_size=512,\n",
        "                                     epochs=12,\n",
        "                                    callbacks = callbacks_list)\n",
        "\n",
        "acc = regularised_layers_model_history.history['accuracy']\n",
        "val_acc = regularised_layers_model_history.history['val_accuracy']\n",
        "loss = regularised_layers_model_history.history['loss']\n",
        "val_loss = regularised_layers_model_history.history['val_loss']\n",
        "epochs = range(1, len(acc) + 1)\n",
        "\n",
        "plt.figure(figsize=(15,4))\n",
        "\n",
        "ax1 = plt.subplot(1, 2, 1)\n",
        "plt.plot(epochs, loss, label='Training loss')\n",
        "plt.plot(epochs, val_loss, label='Validation loss')\n",
        "plt.fill_between(epochs, loss,val_loss,color='g',alpha=.1)\n",
        "\n",
        "plt.title('Training and validation loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "\n",
        "ax2 = plt.subplot(1, 2, 2)\n",
        "plt.plot(epochs, acc, label='Training accuracy')\n",
        "plt.plot(epochs, val_acc, label='Validation accuracy')\n",
        "plt.fill_between(epochs, acc,val_acc,color='g',alpha=.1)\n",
        "plt.title('Training and validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/12\n",
            "83/83 [==============================] - 4s 14ms/step - loss: 4.4890 - accuracy: 0.2610 - val_loss: 4.0670 - val_accuracy: 0.5698\n",
            "Epoch 2/12\n",
            "83/83 [==============================] - 1s 6ms/step - loss: 3.4570 - accuracy: 0.6588 - val_loss: 3.5789 - val_accuracy: 0.6957\n",
            "Epoch 3/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 3.2030 - accuracy: 0.7196 - val_loss: 3.2590 - val_accuracy: 0.7336\n",
            "Epoch 4/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 3.0703 - accuracy: 0.7427 - val_loss: 3.0617 - val_accuracy: 0.7582\n",
            "Epoch 5/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 2.9613 - accuracy: 0.7660 - val_loss: 2.9313 - val_accuracy: 0.7744\n",
            "Epoch 6/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 2.8784 - accuracy: 0.7783 - val_loss: 2.8379 - val_accuracy: 0.7850\n",
            "Epoch 7/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 2.8132 - accuracy: 0.7859 - val_loss: 2.7666 - val_accuracy: 0.7955\n",
            "Epoch 8/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 2.7478 - accuracy: 0.7932 - val_loss: 2.7058 - val_accuracy: 0.8017\n",
            "Epoch 9/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 2.6981 - accuracy: 0.8025 - val_loss: 2.6527 - val_accuracy: 0.8059\n",
            "Epoch 10/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 2.6544 - accuracy: 0.8035 - val_loss: 2.6038 - val_accuracy: 0.8104\n",
            "Epoch 11/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 2.6030 - accuracy: 0.8067 - val_loss: 2.5585 - val_accuracy: 0.8142\n",
            "Epoch 12/12\n",
            "83/83 [==============================] - 1s 7ms/step - loss: 2.5484 - accuracy: 0.8164 - val_loss: 2.5160 - val_accuracy: 0.8168\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wky7gVtbQJVv",
        "outputId": "fadc7fe6-35f1-4eb6-cc63-5c8957da5a65"
      },
      "source": [
        "test_loss, test_accuracy = model_d.evaluate(test_images, test_labels, steps=math.ceil(len(test_labels)/32))\n",
        "print('Accuracy on test dataset using densly connected model:', test_accuracy)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "313/313 [==============================] - 1s 3ms/step - loss: 2.5341 - accuracy: 0.8118\n",
            "Accuracy on test dataset after layer specific weight regularisation: 0.8118000030517578\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "69dzV_eTRDEd"
      },
      "source": [
        "Without changine the model architecture, the accuracy on the test set does not show any improvement, although it improves only by a few decimal poitns on the validation set, depending on the learning rate, the regularisers and the Dropouts. The highest reported accuracy in Assignment-1 was 0.8915, any changes in the hyper-parameters result in a decreased accuracy on the test set, as shown above where the values of the kernel_regulariers and the optimiser were changed."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cfTL_5kPsl4j"
      },
      "source": [
        "### Task 1.2 Train a ConvNet from scratch\n",
        "\n",
        "*(weight ~5%)*\n",
        "\n",
        "Build a ConvNet to replace the densely connected network in Task 1.1. Report the classification accuracy on the test set. Aim to achieve higher accuracy. \n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9FGomTWkVz0b",
        "outputId": "1b95bcca-1207-401d-cd9d-9e3de21c36e4"
      },
      "source": [
        "# reshaping the data for ConvNet\n",
        "\n",
        "train_images = np.reshape(train_images, (train_images.shape[0], train_images.shape[1], train_images.shape[2], 1))\n",
        "test_images = np.reshape(test_images, (test_images.shape[0], test_images.shape[1], test_images.shape[2], 1))\n",
        "print(train_images.shape)\n",
        "X_train = np.reshape(X_train, (X_train.shape[0], X_train.shape[1], X_train.shape[2], 1))\n",
        "X_val = np.reshape(X_val, (X_val.shape[0], X_val.shape[1], X_val.shape[2], 1))\n",
        "#train_images = train_images.values.reshape(-1,28,28,1)"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(60000, 28, 28, 1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3KBNjapzRZ2I",
        "outputId": "3a1d781a-665e-41c7-c859-5ef161d027fe"
      },
      "source": [
        "import tensorflow.keras.backend as K\n",
        "from tensorflow.keras.layers import (Input, Conv2D, MaxPooling2D, Flatten, \n",
        "                                     Dense, Dropout, BatchNormalization)\n",
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.optimizers import Adam\n",
        "from tensorflow.keras import regularizers\n",
        "from tensorflow.keras.callbacks import EarlyStopping,ReduceLROnPlateau\n",
        "\n",
        "\n",
        "# defining a Convolutional Network\n",
        "model = Sequential()\n",
        "model.add(Conv2D(64, kernel_size=(3, 3), activation='relu', kernel_initializer='he_normal', padding='same', input_shape=(28, 28, 1)))\n",
        "model.add(Conv2D(64, kernel_size=(3, 3), activation='relu',padding='same'))\n",
        "model.add(tf.keras.layers.BatchNormalization())\n",
        "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "model.add(Dropout(0.3))\n",
        "model.add(Conv2D(128, kernel_size=(3, 3), activation='relu', kernel_initializer='he_normal', padding='same', input_shape=(28, 28, 1)))\n",
        "model.add(Conv2D(128, kernel_size=(3, 3), padding='same', activation='relu'))\n",
        "model.add(tf.keras.layers.BatchNormalization())\n",
        "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "model.add(Dropout(0.3))\n",
        "model.add(Conv2D(256, kernel_size=(3, 3), activation='relu', kernel_initializer='he_normal', padding='same', input_shape=(28, 28, 1)))\n",
        "model.add(Conv2D(256, kernel_size=(3, 3), padding='same', activation='relu'))\n",
        "model.add(Conv2D(256, kernel_size=(3, 3), padding='same', activation='relu'))\n",
        "model.add(tf.keras.layers.BatchNormalization())\n",
        "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "model.add(Dropout(0.3))\n",
        "model.add(Flatten())\n",
        "model.add(Dense(2048, activation='relu'))\n",
        "model.add(Dropout(0.3))\n",
        "model.add(Dense(512, activation='relu'))\n",
        "model.add(Dense(10, activation='softmax'))\n",
        "adam = Adam(lr=0.0001, decay=1e-6)\n",
        "\n",
        "\n",
        "model.compile(optimizer=adam,\n",
        "                  loss='sparse_categorical_crossentropy',\n",
        "                  metrics=['accuracy'])\n",
        "print(model.summary())\n",
        "init_weights = model.get_weights()\n"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d (Conv2D)              (None, 28, 28, 64)        640       \n",
            "_________________________________________________________________\n",
            "conv2d_1 (Conv2D)            (None, 28, 28, 64)        36928     \n",
            "_________________________________________________________________\n",
            "batch_normalization (BatchNo (None, 28, 28, 64)        256       \n",
            "_________________________________________________________________\n",
            "max_pooling2d (MaxPooling2D) (None, 14, 14, 64)        0         \n",
            "_________________________________________________________________\n",
            "dropout (Dropout)            (None, 14, 14, 64)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_2 (Conv2D)            (None, 14, 14, 128)       73856     \n",
            "_________________________________________________________________\n",
            "conv2d_3 (Conv2D)            (None, 14, 14, 128)       147584    \n",
            "_________________________________________________________________\n",
            "batch_normalization_1 (Batch (None, 14, 14, 128)       512       \n",
            "_________________________________________________________________\n",
            "max_pooling2d_1 (MaxPooling2 (None, 7, 7, 128)         0         \n",
            "_________________________________________________________________\n",
            "dropout_1 (Dropout)          (None, 7, 7, 128)         0         \n",
            "_________________________________________________________________\n",
            "conv2d_4 (Conv2D)            (None, 7, 7, 256)         295168    \n",
            "_________________________________________________________________\n",
            "conv2d_5 (Conv2D)            (None, 7, 7, 256)         590080    \n",
            "_________________________________________________________________\n",
            "conv2d_6 (Conv2D)            (None, 7, 7, 256)         590080    \n",
            "_________________________________________________________________\n",
            "batch_normalization_2 (Batch (None, 7, 7, 256)         1024      \n",
            "_________________________________________________________________\n",
            "max_pooling2d_2 (MaxPooling2 (None, 3, 3, 256)         0         \n",
            "_________________________________________________________________\n",
            "dropout_2 (Dropout)          (None, 3, 3, 256)         0         \n",
            "_________________________________________________________________\n",
            "flatten (Flatten)            (None, 2304)              0         \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 2048)              4720640   \n",
            "_________________________________________________________________\n",
            "dropout_3 (Dropout)          (None, 2048)              0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 512)               1049088   \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 10)                5130      \n",
            "=================================================================\n",
            "Total params: 7,510,986\n",
            "Trainable params: 7,510,090\n",
            "Non-trainable params: 896\n",
            "_________________________________________________________________\n",
            "None\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mXKNuZVpYZhQ"
      },
      "source": [
        "# implementing Early stopping to stop training once the model performance stops improving on a hold out validation dataset\n",
        "# implementing ReduceLROnPlateau, a callback that monitors the validation loss (in this case) and if no improvement is seen for a 'patience' number of epochs, the learning rate is reduced\n",
        "reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.2,\n",
        "                              patience=3, min_lr=0.0001)\n",
        "earlystop=EarlyStopping(monitor='val_loss', patience=3)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "MnQLfrvvU0w2",
        "outputId": "189e67b3-7182-4caf-c028-f8b162ca09f2"
      },
      "source": [
        "history = model.fit(train_images, \n",
        "                    train_labels,\n",
        "                    validation_split=0.3,\n",
        "                    batch_size=128,\n",
        "                    epochs=25,\n",
        "                    callbacks=[reduce_lr])\n",
        "\n",
        "acc = history.history['accuracy']\n",
        "val_acc = history.history['val_accuracy']\n",
        "loss = history.history['loss']\n",
        "val_loss = history.history['val_loss']\n",
        "epochs = range(1, len(acc) + 1)\n",
        "\n",
        "plt.figure(figsize=(15,4))\n",
        "\n",
        "ax1 = plt.subplot(1, 2, 1)\n",
        "plt.plot(epochs, loss, label='Training loss')\n",
        "plt.plot(epochs, val_loss, label='Validation loss')\n",
        "plt.fill_between(epochs, loss,val_loss,color='g',alpha=.1)\n",
        "\n",
        "plt.title('Training and validation loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "\n",
        "ax2 = plt.subplot(1, 2, 2)\n",
        "plt.plot(epochs, acc, label='Training accuracy')\n",
        "plt.plot(epochs, val_acc, label='Validation accuracy')\n",
        "plt.fill_between(epochs, acc,val_acc,color='g',alpha=.1)\n",
        "plt.title('Training and validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/100\n",
            "329/329 [==============================] - 52s 70ms/step - loss: 1.0947 - accuracy: 0.6246 - val_loss: 1.3367 - val_accuracy: 0.5534\n",
            "Epoch 2/100\n",
            "329/329 [==============================] - 22s 67ms/step - loss: 0.4972 - accuracy: 0.8154 - val_loss: 0.4240 - val_accuracy: 0.8452\n",
            "Epoch 3/100\n",
            "329/329 [==============================] - 22s 67ms/step - loss: 0.4028 - accuracy: 0.8521 - val_loss: 0.3441 - val_accuracy: 0.8746\n",
            "Epoch 4/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.3590 - accuracy: 0.8681 - val_loss: 0.3285 - val_accuracy: 0.8833\n",
            "Epoch 5/100\n",
            "329/329 [==============================] - 22s 67ms/step - loss: 0.3219 - accuracy: 0.8797 - val_loss: 0.2983 - val_accuracy: 0.8931\n",
            "Epoch 6/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.3065 - accuracy: 0.8843 - val_loss: 0.2875 - val_accuracy: 0.8979\n",
            "Epoch 7/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.2882 - accuracy: 0.8936 - val_loss: 0.2667 - val_accuracy: 0.9036\n",
            "Epoch 8/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.2714 - accuracy: 0.8995 - val_loss: 0.2554 - val_accuracy: 0.9086\n",
            "Epoch 9/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.2468 - accuracy: 0.9090 - val_loss: 0.2667 - val_accuracy: 0.9034\n",
            "Epoch 10/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.2360 - accuracy: 0.9105 - val_loss: 0.2393 - val_accuracy: 0.9129\n",
            "Epoch 11/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.2232 - accuracy: 0.9167 - val_loss: 0.2293 - val_accuracy: 0.9162\n",
            "Epoch 12/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.2190 - accuracy: 0.9147 - val_loss: 0.2319 - val_accuracy: 0.9196\n",
            "Epoch 13/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.2096 - accuracy: 0.9210 - val_loss: 0.2134 - val_accuracy: 0.9251\n",
            "Epoch 14/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1919 - accuracy: 0.9282 - val_loss: 0.2519 - val_accuracy: 0.9072\n",
            "Epoch 15/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1861 - accuracy: 0.9286 - val_loss: 0.2220 - val_accuracy: 0.9242\n",
            "Epoch 16/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1873 - accuracy: 0.9296 - val_loss: 0.2230 - val_accuracy: 0.9202\n",
            "Epoch 17/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1731 - accuracy: 0.9345 - val_loss: 0.2342 - val_accuracy: 0.9182\n",
            "Epoch 18/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1694 - accuracy: 0.9362 - val_loss: 0.2088 - val_accuracy: 0.9277\n",
            "Epoch 19/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1573 - accuracy: 0.9410 - val_loss: 0.2213 - val_accuracy: 0.9254\n",
            "Epoch 20/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1549 - accuracy: 0.9412 - val_loss: 0.2041 - val_accuracy: 0.9285\n",
            "Epoch 21/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1469 - accuracy: 0.9444 - val_loss: 0.2258 - val_accuracy: 0.9255\n",
            "Epoch 22/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1464 - accuracy: 0.9428 - val_loss: 0.2071 - val_accuracy: 0.9317\n",
            "Epoch 23/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1385 - accuracy: 0.9472 - val_loss: 0.2203 - val_accuracy: 0.9278\n",
            "Epoch 24/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1313 - accuracy: 0.9502 - val_loss: 0.2192 - val_accuracy: 0.9299\n",
            "Epoch 25/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1228 - accuracy: 0.9518 - val_loss: 0.2136 - val_accuracy: 0.9335\n",
            "Epoch 26/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1197 - accuracy: 0.9540 - val_loss: 0.2092 - val_accuracy: 0.9323\n",
            "Epoch 27/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1108 - accuracy: 0.9578 - val_loss: 0.2117 - val_accuracy: 0.9327\n",
            "Epoch 28/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1151 - accuracy: 0.9559 - val_loss: 0.2092 - val_accuracy: 0.9328\n",
            "Epoch 29/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1042 - accuracy: 0.9594 - val_loss: 0.2402 - val_accuracy: 0.9266\n",
            "Epoch 30/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.1062 - accuracy: 0.9583 - val_loss: 0.2228 - val_accuracy: 0.9304\n",
            "Epoch 31/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0964 - accuracy: 0.9630 - val_loss: 0.2092 - val_accuracy: 0.9342\n",
            "Epoch 32/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0880 - accuracy: 0.9669 - val_loss: 0.2471 - val_accuracy: 0.9274\n",
            "Epoch 33/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0927 - accuracy: 0.9632 - val_loss: 0.2265 - val_accuracy: 0.9321\n",
            "Epoch 34/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0835 - accuracy: 0.9686 - val_loss: 0.2279 - val_accuracy: 0.9299\n",
            "Epoch 35/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0783 - accuracy: 0.9704 - val_loss: 0.2384 - val_accuracy: 0.9297\n",
            "Epoch 36/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0801 - accuracy: 0.9681 - val_loss: 0.2353 - val_accuracy: 0.9334\n",
            "Epoch 37/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0802 - accuracy: 0.9705 - val_loss: 0.2524 - val_accuracy: 0.9332\n",
            "Epoch 38/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0732 - accuracy: 0.9718 - val_loss: 0.2416 - val_accuracy: 0.9358\n",
            "Epoch 39/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0699 - accuracy: 0.9741 - val_loss: 0.2273 - val_accuracy: 0.9368\n",
            "Epoch 40/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0734 - accuracy: 0.9706 - val_loss: 0.2567 - val_accuracy: 0.9326\n",
            "Epoch 41/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0623 - accuracy: 0.9760 - val_loss: 0.2586 - val_accuracy: 0.9314\n",
            "Epoch 42/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0592 - accuracy: 0.9773 - val_loss: 0.2445 - val_accuracy: 0.9358\n",
            "Epoch 43/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0561 - accuracy: 0.9783 - val_loss: 0.2591 - val_accuracy: 0.9337\n",
            "Epoch 44/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0609 - accuracy: 0.9776 - val_loss: 0.2540 - val_accuracy: 0.9339\n",
            "Epoch 45/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0536 - accuracy: 0.9789 - val_loss: 0.2670 - val_accuracy: 0.9357\n",
            "Epoch 46/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0528 - accuracy: 0.9804 - val_loss: 0.2577 - val_accuracy: 0.9359\n",
            "Epoch 47/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0521 - accuracy: 0.9803 - val_loss: 0.2677 - val_accuracy: 0.9339\n",
            "Epoch 48/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0479 - accuracy: 0.9815 - val_loss: 0.2807 - val_accuracy: 0.9367\n",
            "Epoch 49/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0457 - accuracy: 0.9824 - val_loss: 0.2654 - val_accuracy: 0.9358\n",
            "Epoch 50/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0463 - accuracy: 0.9837 - val_loss: 0.2612 - val_accuracy: 0.9389\n",
            "Epoch 51/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0475 - accuracy: 0.9823 - val_loss: 0.2696 - val_accuracy: 0.9381\n",
            "Epoch 52/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0444 - accuracy: 0.9839 - val_loss: 0.2649 - val_accuracy: 0.9378\n",
            "Epoch 53/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0398 - accuracy: 0.9849 - val_loss: 0.2823 - val_accuracy: 0.9345\n",
            "Epoch 54/100\n",
            "329/329 [==============================] - 23s 68ms/step - loss: 0.0382 - accuracy: 0.9858 - val_loss: 0.2774 - val_accuracy: 0.9360\n",
            "Epoch 55/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0428 - accuracy: 0.9834 - val_loss: 0.2624 - val_accuracy: 0.9373\n",
            "Epoch 56/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0409 - accuracy: 0.9846 - val_loss: 0.2823 - val_accuracy: 0.9365\n",
            "Epoch 57/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0358 - accuracy: 0.9866 - val_loss: 0.3286 - val_accuracy: 0.9269\n",
            "Epoch 58/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0372 - accuracy: 0.9856 - val_loss: 0.2867 - val_accuracy: 0.9364\n",
            "Epoch 59/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0348 - accuracy: 0.9863 - val_loss: 0.3089 - val_accuracy: 0.9365\n",
            "Epoch 60/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0380 - accuracy: 0.9870 - val_loss: 0.2885 - val_accuracy: 0.9356\n",
            "Epoch 61/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0317 - accuracy: 0.9881 - val_loss: 0.2912 - val_accuracy: 0.9362\n",
            "Epoch 62/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0321 - accuracy: 0.9883 - val_loss: 0.2980 - val_accuracy: 0.9321\n",
            "Epoch 63/100\n",
            "329/329 [==============================] - 23s 69ms/step - loss: 0.0309 - accuracy: 0.9890 - val_loss: 0.3029 - val_accuracy: 0.9364\n",
            "Epoch 64/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0276 - accuracy: 0.9897 - val_loss: 0.2824 - val_accuracy: 0.9388\n",
            "Epoch 65/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0303 - accuracy: 0.9887 - val_loss: 0.3375 - val_accuracy: 0.9338\n",
            "Epoch 66/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0284 - accuracy: 0.9897 - val_loss: 0.2942 - val_accuracy: 0.9394\n",
            "Epoch 67/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0276 - accuracy: 0.9906 - val_loss: 0.3084 - val_accuracy: 0.9359\n",
            "Epoch 68/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0311 - accuracy: 0.9886 - val_loss: 0.3149 - val_accuracy: 0.9364\n",
            "Epoch 69/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0273 - accuracy: 0.9900 - val_loss: 0.2946 - val_accuracy: 0.9371\n",
            "Epoch 70/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0223 - accuracy: 0.9921 - val_loss: 0.3142 - val_accuracy: 0.9371\n",
            "Epoch 71/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0230 - accuracy: 0.9917 - val_loss: 0.3287 - val_accuracy: 0.9361\n",
            "Epoch 72/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0279 - accuracy: 0.9896 - val_loss: 0.3206 - val_accuracy: 0.9369\n",
            "Epoch 73/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0272 - accuracy: 0.9909 - val_loss: 0.3101 - val_accuracy: 0.9377\n",
            "Epoch 74/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0231 - accuracy: 0.9918 - val_loss: 0.3190 - val_accuracy: 0.9369\n",
            "Epoch 75/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0266 - accuracy: 0.9906 - val_loss: 0.3228 - val_accuracy: 0.9389\n",
            "Epoch 76/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0199 - accuracy: 0.9929 - val_loss: 0.3420 - val_accuracy: 0.9355\n",
            "Epoch 77/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0251 - accuracy: 0.9903 - val_loss: 0.3526 - val_accuracy: 0.9341\n",
            "Epoch 78/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0230 - accuracy: 0.9916 - val_loss: 0.3192 - val_accuracy: 0.9369\n",
            "Epoch 79/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0210 - accuracy: 0.9927 - val_loss: 0.3184 - val_accuracy: 0.9385\n",
            "Epoch 80/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0209 - accuracy: 0.9926 - val_loss: 0.3400 - val_accuracy: 0.9393\n",
            "Epoch 81/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0216 - accuracy: 0.9919 - val_loss: 0.3353 - val_accuracy: 0.9362\n",
            "Epoch 82/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0176 - accuracy: 0.9934 - val_loss: 0.3368 - val_accuracy: 0.9382\n",
            "Epoch 83/100\n",
            "329/329 [==============================] - 23s 68ms/step - loss: 0.0210 - accuracy: 0.9924 - val_loss: 0.3843 - val_accuracy: 0.9309\n",
            "Epoch 84/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0208 - accuracy: 0.9926 - val_loss: 0.3218 - val_accuracy: 0.9384\n",
            "Epoch 85/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0194 - accuracy: 0.9931 - val_loss: 0.3429 - val_accuracy: 0.9380\n",
            "Epoch 86/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0184 - accuracy: 0.9935 - val_loss: 0.3474 - val_accuracy: 0.9376\n",
            "Epoch 87/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0186 - accuracy: 0.9934 - val_loss: 0.3380 - val_accuracy: 0.9357\n",
            "Epoch 88/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0169 - accuracy: 0.9945 - val_loss: 0.3534 - val_accuracy: 0.9353\n",
            "Epoch 89/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0188 - accuracy: 0.9934 - val_loss: 0.3575 - val_accuracy: 0.9334\n",
            "Epoch 90/100\n",
            "329/329 [==============================] - 23s 69ms/step - loss: 0.0182 - accuracy: 0.9935 - val_loss: 0.3249 - val_accuracy: 0.9378\n",
            "Epoch 91/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0188 - accuracy: 0.9934 - val_loss: 0.3310 - val_accuracy: 0.9352\n",
            "Epoch 92/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0167 - accuracy: 0.9938 - val_loss: 0.3482 - val_accuracy: 0.9387\n",
            "Epoch 93/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0173 - accuracy: 0.9937 - val_loss: 0.3567 - val_accuracy: 0.9357\n",
            "Epoch 94/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0155 - accuracy: 0.9944 - val_loss: 0.3513 - val_accuracy: 0.9374\n",
            "Epoch 95/100\n",
            "329/329 [==============================] - 23s 69ms/step - loss: 0.0160 - accuracy: 0.9941 - val_loss: 0.3457 - val_accuracy: 0.9379\n",
            "Epoch 96/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0146 - accuracy: 0.9949 - val_loss: 0.3788 - val_accuracy: 0.9336\n",
            "Epoch 97/100\n",
            "329/329 [==============================] - 23s 68ms/step - loss: 0.0145 - accuracy: 0.9948 - val_loss: 0.3837 - val_accuracy: 0.9354\n",
            "Epoch 98/100\n",
            "329/329 [==============================] - 23s 69ms/step - loss: 0.0169 - accuracy: 0.9940 - val_loss: 0.3621 - val_accuracy: 0.9369\n",
            "Epoch 99/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0150 - accuracy: 0.9947 - val_loss: 0.3689 - val_accuracy: 0.9372\n",
            "Epoch 100/100\n",
            "329/329 [==============================] - 22s 68ms/step - loss: 0.0166 - accuracy: 0.9942 - val_loss: 0.3771 - val_accuracy: 0.9395\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "l9Y6ZDYzVJb8",
        "outputId": "2431c3c0-1bff-4efe-f540-07e475948f30"
      },
      "source": [
        "test_loss, test_accuracy = model.evaluate(test_images, test_labels, steps=math.ceil(len(test_labels)/32))\n",
        "print('Accuracy on test dataset after using CNN:', test_accuracy)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "313/313 [==============================] - 3s 9ms/step - loss: 0.4265 - accuracy: 0.9324\n",
            "Accuracy on test dataset after layer specific weight regularisation: 0.9323999881744385\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-wkBB_vm1g1-"
      },
      "source": [
        "After using a CNN model, the accuracy over the test set increases to 0.9324 without much hyperparameter tuning."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qbhr44DnMbKZ"
      },
      "source": [
        "\n",
        "### Task 1.3 Build an input pipeline for data augmentation\n",
        "\n",
        "*(weight ~3%)*\n",
        "\n",
        "Build a data preprocessing pipeline to perform data augmentation. (You may use Keras ImageDataGenerator or write your own transformations.)\n",
        "\n",
        "- Report the new classification accuracy. Make sure that you use the same number of training epochs as in Task 1.2.\n",
        "\n",
        "- (Optional) Profile your input pipeline to identify the most time-consuming operation. What actions have you taken to address that slow operation? (*Hint: You may use the [TensorFlow Profiler](https://github.com/tensorflow/profiler).*)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4YtlnztEpzUl"
      },
      "source": [
        "# data augumentation, using Keras ImageDataGenerator\n",
        "\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "\n",
        "model.set_weights(init_weights)\n",
        "\n",
        "\n",
        "train_datagen = ImageDataGenerator( \n",
        "        rotation_range=10,        # randomly rotate between 0-rotation_range angle\n",
        "        width_shift_range=0.2,    # randomly shift horizontally by this much\n",
        "        height_shift_range=0.2,   # randomly shift vertically by this much\n",
        "        shear_range=0.2,          # randomly shear by this much\n",
        "        zoom_range=0.2,           # randomly zoom (80% - 120%)\n",
        "        horizontal_flip=True,\n",
        "        fill_mode='nearest'       # fill any pixels lost in xform with nearest\n",
        "    )\n",
        "\n",
        "val_datagen = ImageDataGenerator()\n",
        "test_datagen = ImageDataGenerator()\n",
        "\n",
        "#train_datagen.fit(X_train)\n",
        "\n",
        "# fit the data generator on training data & validate on\n",
        "num_train_batches = len(X_train) // 32\n",
        "num_train_batches += (0 if len(X_train) % 32 == 0 else 1)     \n",
        "num_val_batches = len(X_val) // 32\n",
        "num_val_batches += (0 if len(X_val) % 32 == 0 else 1)     \n",
        "num_test_batches = len(test_images) // 32\n",
        "num_test_batches += (0 if len(test_images) % 32 == 0 else 1)    \n",
        "\n",
        "train_generator = train_datagen.flow(X_train, y_train, \n",
        "                                      batch_size=32,\n",
        "                                      seed=123)\n",
        "val_generator = val_datagen.flow(X_val, y_val, \n",
        "                                      batch_size=32,\n",
        "                                      seed=123)\n",
        "test_generator = test_datagen.flow(test_images, test_labels, \n",
        "                                    batch_size=32,\n",
        "                                    seed=123)\n",
        "\n",
        "# history = model.fit_generator(train_datagen.flow(X_train, y_train, batch_size=128),\n",
        "#                     validation_data =  (X_val,y_val),\n",
        "#                     epochs=100,\n",
        "#                     callbacks=[reduce_lr],\n",
        "#                     shuffle=True)  "
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lUrVT-qN5tmf"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "44zzOd-isggz"
      },
      "source": [
        "AUTOTUNE = tf.data.experimental.AUTOTUNE\n",
        "\n",
        "def augment(image,label):\n",
        "  image = tf.image.convert_image_dtype(image, tf.float32) \n",
        "  image = tf.image.random_flip_left_right(image)\n",
        "  image = tf.image.resize(image, (28, 28))\n",
        "  image = tf.image.resize_with_pad(image, 28 + 6, 28 + 6) # Add pixels of padding\n",
        "  image = tf.image.random_crop(image, size=[28, 28, 3]) # Random crop back to IMAGE_SIZExIMAGE_SIZE\n",
        "  return image, label\n",
        "\n",
        "augmented_train_batches = (\n",
        "    fmnist_train_ds\n",
        "    .cache()\n",
        "    # The augmentation is added here.\n",
        "    .map(augment, num_parallel_calls=AUTOTUNE)\n",
        "    .batch(32)\n",
        "    .prefetch(AUTOTUNE)\n",
        ")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hAk4NU_YXwK8"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "SPItzZz2X78z",
        "outputId": "cbbf5757-3ced-474f-b35d-646828a5d2b1"
      },
      "source": [
        "!pip install -U tensorboard_plugin_profile\n"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting tensorboard_plugin_profile\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d1/4e/0bf160776e5dacdba5105a580aa57a2bd37c1cc4faa1bd7695d82e7d6ae7/tensorboard_plugin_profile-2.4.0-py3-none-any.whl (1.2MB)\n",
            "\r\u001b[K     |▎                               | 10kB 25.3MB/s eta 0:00:01\r\u001b[K     |▋                               | 20kB 30.0MB/s eta 0:00:01\r\u001b[K     |▉                               | 30kB 34.0MB/s eta 0:00:01\r\u001b[K     |█▏                              | 40kB 29.6MB/s eta 0:00:01\r\u001b[K     |█▍                              | 51kB 28.3MB/s eta 0:00:01\r\u001b[K     |█▊                              | 61kB 30.8MB/s eta 0:00:01\r\u001b[K     |██                              | 71kB 26.3MB/s eta 0:00:01\r\u001b[K     |██▎                             | 81kB 26.6MB/s eta 0:00:01\r\u001b[K     |██▌                             | 92kB 23.3MB/s eta 0:00:01\r\u001b[K     |██▉                             | 102kB 24.6MB/s eta 0:00:01\r\u001b[K     |███▏                            | 112kB 24.6MB/s eta 0:00:01\r\u001b[K     |███▍                            | 122kB 24.6MB/s eta 0:00:01\r\u001b[K     |███▊                            | 133kB 24.6MB/s eta 0:00:01\r\u001b[K     |████                            | 143kB 24.6MB/s eta 0:00:01\r\u001b[K     |████▎                           | 153kB 24.6MB/s eta 0:00:01\r\u001b[K     |████▌                           | 163kB 24.6MB/s eta 0:00:01\r\u001b[K     |████▉                           | 174kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████                           | 184kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████▍                          | 194kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████▊                          | 204kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████                          | 215kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████▎                         | 225kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████▌                         | 235kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████▉                         | 245kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████                         | 256kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████▍                        | 266kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████▋                        | 276kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████                        | 286kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████▎                       | 296kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████▌                       | 307kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████▉                       | 317kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████                       | 327kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████▍                      | 337kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████▋                      | 348kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████                      | 358kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████▏                     | 368kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████▌                     | 378kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████▊                     | 389kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████                     | 399kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████▍                    | 409kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████▋                    | 419kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████                    | 430kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████▏                   | 440kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████▌                   | 450kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████▊                   | 460kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████                   | 471kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████▎                  | 481kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████▋                  | 491kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████                  | 501kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████▏                 | 512kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████▌                 | 522kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████▊                 | 532kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████                 | 542kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████▎                | 552kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████▋                | 563kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████▉                | 573kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████▏               | 583kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████▌               | 593kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████▊               | 604kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████               | 614kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████▎              | 624kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████▋              | 634kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████▉              | 645kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████▏             | 655kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████▍             | 665kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████▊             | 675kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████             | 686kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████▎            | 696kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████▋            | 706kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████▉            | 716kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████▏           | 727kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████▍           | 737kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████▊           | 747kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████           | 757kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████▎          | 768kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████▌          | 778kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████▉          | 788kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████▏         | 798kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████▍         | 808kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████▊         | 819kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████         | 829kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████▎        | 839kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████▌        | 849kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████▉        | 860kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████████        | 870kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████████▍       | 880kB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████████▊       | 890kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████       | 901kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▎      | 911kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▌      | 921kB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▉      | 931kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████████      | 942kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████████▍     | 952kB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████████▋     | 962kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████     | 972kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▎    | 983kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▌    | 993kB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▉    | 1.0MB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████████████    | 1.0MB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████████████▍   | 1.0MB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████████████▋   | 1.0MB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████   | 1.0MB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████▏  | 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████▌  | 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████▉  | 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████  | 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████▍ | 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████▋ | 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████ | 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▏| 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▌| 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▊| 1.1MB 24.6MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 1.2MB 24.6MB/s \n",
            "\u001b[?25hRequirement already satisfied, skipping upgrade: six>=1.10.0 in /usr/local/lib/python3.7/dist-packages (from tensorboard_plugin_profile) (1.15.0)\n",
            "Requirement already satisfied, skipping upgrade: werkzeug>=0.11.15 in /usr/local/lib/python3.7/dist-packages (from tensorboard_plugin_profile) (2.0.0)\n",
            "Collecting gviz-api>=1.9.0\n",
            "  Downloading https://files.pythonhosted.org/packages/8c/8f/c6f16235a16b3dc4efdcf34dbc93b3b6f678b88176dbd6a36c75d678888f/gviz_api-1.9.0-py2.py3-none-any.whl\n",
            "Requirement already satisfied, skipping upgrade: protobuf>=3.6.0 in /usr/local/lib/python3.7/dist-packages (from tensorboard_plugin_profile) (3.12.4)\n",
            "Requirement already satisfied, skipping upgrade: setuptools>=41.0.0 in /usr/local/lib/python3.7/dist-packages (from tensorboard_plugin_profile) (56.1.0)\n",
            "Installing collected packages: gviz-api, tensorboard-plugin-profile\n",
            "Successfully installed gviz-api-1.9.0 tensorboard-plugin-profile-2.4.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DUEbMOEnYQR8",
        "outputId": "0a6bb496-45c5-4383-da0e-13ebb43dd547"
      },
      "source": [
        "# Profiling the input pipeline to identify the most time-consuming operation\n",
        "# Training the model for the same number of epochs\n",
        "\n",
        "!rm -rf ./logs/\n",
        "\n",
        "from datetime import datetime\n",
        "import os\n",
        "\n",
        "root_logdir = \"logs\"\n",
        "run_id = datetime.now().strftime(\"%Y%m%d-%H%M%S\")\n",
        "logdir = os.path.join(root_logdir, run_id)\n",
        "\n",
        "callbacks = [\n",
        "    tf.keras.callbacks.TensorBoard(\n",
        "        log_dir=logdir,\n",
        "        histogram_freq=1,\n",
        "         write_graph=True, \n",
        "         write_images=True,\n",
        "         profile_batch = '15,25'\n",
        "    )\n",
        "]\n",
        "\n",
        "\n",
        "# reset the training\n",
        "model.set_weights(init_weights)\n",
        "\n",
        "history = model.fit(train_generator,\n",
        "                    steps_per_epoch = num_train_batches, \n",
        "                    epochs = 125,\n",
        "                    validation_data = (val_generator),\n",
        "                    validation_steps = num_val_batches,\n",
        "                    batch_size=512,\n",
        "                    callbacks=[callbacks],\n",
        "                    shuffle=True)  \n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/125\n",
            "1407/1407 [==============================] - 22s 16ms/step - loss: 0.8901 - accuracy: 0.6657 - val_loss: 0.5394 - val_accuracy: 0.7903\n",
            "Epoch 2/125\n",
            "1407/1407 [==============================] - 21s 15ms/step - loss: 0.6364 - accuracy: 0.7546 - val_loss: 0.4852 - val_accuracy: 0.8144\n",
            "Epoch 3/125\n",
            "1407/1407 [==============================] - 21s 15ms/step - loss: 0.5608 - accuracy: 0.7860 - val_loss: 0.4368 - val_accuracy: 0.8281\n",
            "Epoch 4/125\n",
            "1406/1407 [============================>.] - ETA: 0s - loss: 0.5116 - accuracy: 0.8065"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gK01SWsGpUbt"
      },
      "source": [
        "acc = history.history['accuracy']\n",
        "val_acc = history.history['val_accuracy']\n",
        "loss = history.history['loss']\n",
        "val_loss = history.history['val_loss']\n",
        "epochs = range(1, len(acc) + 1)\n",
        "\n",
        "plt.figure(figsize=(15,4))\n",
        "\n",
        "ax1 = plt.subplot(1, 2, 1)\n",
        "plt.plot(epochs, loss, label='Training loss')\n",
        "plt.plot(epochs, val_loss, label='Validation loss')\n",
        "plt.fill_between(epochs, loss,val_loss,color='g',alpha=.1)\n",
        "\n",
        "plt.title('Training and validation loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "\n",
        "ax2 = plt.subplot(1, 2, 2)\n",
        "plt.plot(epochs, acc, label='Training accuracy')\n",
        "plt.plot(epochs, val_acc, label='Validation accuracy')\n",
        "plt.fill_between(epochs, acc,val_acc,color='g',alpha=.1)\n",
        "plt.title('Training and validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yjZrEi9_D07x"
      },
      "source": [
        "test_loss, test_accuracy = model.evaluate(test_images, test_labels, steps=math.ceil(len(test_labels)/32))\n",
        "print('Accuracy on test dataset after data augmentation:', test_accuracy)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Xr5Dqcqtq7ky"
      },
      "source": [
        "After using data augumentation, for the training process, the accuracy of the model on the test set becomes: "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AKa4t3HEYmsA"
      },
      "source": [
        "%reload_ext tensorboard\n",
        "%tensorboard --logdir=logs"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "prctXU4BswKK"
      },
      "source": [
        "### Task 1.4 Fashion-MNIST with transfer learning\n",
        "\n",
        "*(weight ~3%)*\n",
        "\n",
        "Use a pretrained model as the convolutional base to improve the classification performance. (Hint: You may use models in Keras Applications or those in the TensorFlow Hub.)\n",
        "\n",
        "- Try both with fine-tuning and without fine-tuning.\n",
        "- Report the model performance as before.\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4LOg5ekrfdY-"
      },
      "source": [
        "# using VGG16 for transfer learning\n",
        "# reshaping data as that required for VGG16\n",
        "\n",
        "from keras.utils import to_categorical\n",
        "\n",
        "\n",
        "# Convert the images into 3 channels\n",
        "train_images=np.dstack([train_images] * 3)\n",
        "test_images=np.dstack([test_images]* 3)\n",
        "\n",
        "train_images = train_images.reshape(-1, 28,28,3)\n",
        "test_images= test_images.reshape (-1,28,28,3)\n",
        "\n",
        "# Resize the images 48*48 as required by VGG16\n",
        "from keras.preprocessing.image import img_to_array, array_to_img\n",
        "\n",
        "train_images = np.asarray([img_to_array(array_to_img(im, scale=False).resize((48,48))) for im in train_images])\n",
        "test_images = np.asarray([img_to_array(array_to_img(im, scale=False).resize((48,48))) for im in test_images])\n",
        "\n",
        "\n",
        "#train_images= train_images / 255.\n",
        "#test_images = test_images / 255.\n",
        "train_images = train_images.astype('float32')\n",
        "test_images = test_images.astype('float32')\n",
        "\n",
        "train_labels_one_hot = to_categorical(train_labels)\n",
        "test_labels_one_hot = to_categorical(test_labels)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9qszPeRmfdzq"
      },
      "source": [
        "# initialising the model\n",
        "from keras import optimizers\n",
        "from keras.applications.vgg16 import VGG16\n",
        "from keras.layers import Activation, Dense\n",
        "\n",
        "IMG_WIDTH = 48\n",
        "IMG_HEIGHT = 48\n",
        "IMG_DEPTH = 3\n",
        "BATCH_SIZE = 16\n",
        "\n",
        "model_vgg16= VGG16(weights='imagenet',\n",
        "                  include_top=False, \n",
        "                  input_shape=(IMG_HEIGHT, IMG_WIDTH, IMG_DEPTH)\n",
        "                 )\n",
        "model_vgg16.summary()\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Dh98SV9imc96"
      },
      "source": [
        "for layer in model_vgg16.layers[:-4]:\n",
        "    layer.trainable = False\n",
        "\n",
        "# Check the trainable status of the individual layers\n",
        "for layer in model_vgg16.layers:\n",
        "    print(layer, layer.trainable)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wQmk-9-7mdAv"
      },
      "source": [
        "# training on VGG16 without fine-tuning\n",
        "\n",
        "model= Sequential()\n",
        "\n",
        "model.add(model_vgg16)\n",
        "\n",
        "model.add(tf.keras.layers.Flatten())\n",
        "model.add(tf.keras.layers.Dense(1024, activation='relu'))\n",
        "model.add(tf.keras.layers.Dropout(0.5))\n",
        "model.add(tf.keras.layers.Dense(10, activation='softmax'))\n",
        "\n",
        "model.summary()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CeXUyLoGmruC"
      },
      "source": [
        "model.compile(\n",
        "    loss='categorical_crossentropy',\n",
        "    optimizer=Adam(),\n",
        "    metrics=['acc'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AMP0k3memrwr"
      },
      "source": [
        "history = model.fit(train_images, train_labels_one_hot,\n",
        "          batch_size=128,\n",
        "          epochs=100,\n",
        "          verbose=1,\n",
        "          validation_split=0.3)\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uVFtisqCfE1O"
      },
      "source": [
        "acc = history.history['acc']\n",
        "val_acc = history.history['val_acc']\n",
        "loss = history.history['loss']\n",
        "val_loss = history.history['val_loss']\n",
        "epochs = range(1, len(acc) + 1)\n",
        "\n",
        "plt.figure(figsize=(15,4))\n",
        "\n",
        "ax1 = plt.subplot(1, 2, 1)\n",
        "plt.plot(epochs, loss, label='Training loss')\n",
        "plt.plot(epochs, val_loss, label='Validation loss')\n",
        "plt.fill_between(epochs, loss,val_loss,color='g',alpha=.1)\n",
        "\n",
        "plt.title('Training and validation loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "\n",
        "ax2 = plt.subplot(1, 2, 2)\n",
        "plt.plot(epochs, acc, label='Training accuracy')\n",
        "plt.plot(epochs, val_acc, label='Validation accuracy')\n",
        "plt.fill_between(epochs, acc,val_acc,color='g',alpha=.1)\n",
        "plt.title('Training and validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend()\n",
        "plt.show()\n",
        "\n",
        "\n",
        "test_loss, test_accuracy = model.evaluate(test_images, test_labels, steps=math.ceil(len(test_labels)/32))\n",
        "print('Accuracy on test dataset after using Transfer learning (VGG16) and without fine tuning:', test_accuracy)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VxZg34dCoY6a"
      },
      "source": [
        "# using VGG16 with fine-tuning\n",
        "\n",
        "model_vgg16.trainable = True\n",
        "\n",
        "set_trainable = False\n",
        "for layer in model_vgg16.layers:\n",
        "    if layer.name == 'block5_conv1':\n",
        "        set_trainable = True\n",
        "    if set_trainable:\n",
        "        layer.trainable = True\n",
        "    else:\n",
        "        layer.trainable = False"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "z3FmXowtoY9D"
      },
      "source": [
        "model= Sequential()\n",
        "\n",
        "model.add(model_vgg16)\n",
        "\n",
        "model.add(tf.keras.layers.Flatten())\n",
        "model.add(tf.keras.layers.Dense(1024, activation='relu'))\n",
        "model.add(tf.keras.layers.Dropout(0.5))\n",
        "model.add(tf.keras.layers.Dense(10, activation='softmax'))\n",
        "\n",
        "model.summary()\n",
        "model.compile(loss='categorical_crossentropy',\n",
        "              optimizer=optimizers.RMSprop(lr=1e-5),\n",
        "              metrics=['accuracy'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "background_save": true
        },
        "id": "stVEj6EKoY_s"
      },
      "source": [
        "ft_history = model.fit(train_images, train_labels_one_hot,\n",
        "          batch_size=128,\n",
        "          epochs=50,\n",
        "          verbose=1,\n",
        "          validation_split=0.3)\n",
        "\n",
        "acc = ft_history.history['accuracy']\n",
        "val_acc = ft_history.history['val_accuracy']\n",
        "loss = ft_history.history['loss']\n",
        "val_loss = ft_history.history['val_loss']\n",
        "epochs = range(1, len(acc) + 1)\n",
        "\n",
        "plt.figure(figsize=(15,4))\n",
        "\n",
        "ax1 = plt.subplot(1, 2, 1)\n",
        "plt.plot(epochs, loss, label='Training loss')\n",
        "plt.plot(epochs, val_loss, label='Validation loss')\n",
        "plt.fill_between(epochs, loss,val_loss,color='g',alpha=.1)\n",
        "\n",
        "plt.title('Training and validation loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "\n",
        "ax2 = plt.subplot(1, 2, 2)\n",
        "plt.plot(epochs, acc, label='Training accuracy')\n",
        "plt.plot(epochs, val_acc, label='Validation accuracy')\n",
        "plt.fill_between(epochs, acc,val_acc,color='g',alpha=.1)\n",
        "plt.title('Training and validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "URFbigbEg3Qt"
      },
      "source": [
        "test_loss, test_accuracy = model.evaluate(test_images, test_labels, steps=math.ceil(len(test_labels)/32))\n",
        "print('Accuracy on test dataset after using Transfer learning (VGG16) and with fine tuning:', test_accuracy)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UaHLKDLas_dF"
      },
      "source": [
        "### Task 1.5 Performance comparison\n",
        "\n",
        "*(weight ~3%)*\n",
        "\n",
        "How many parameters are trainable in each of the two settings (with and without fine-tuning)? How does the difference impact the training time?\n",
        "\n",
        "\n",
        "Which setting achieved higher accuracy? Why did it work better for this problem? \n",
        "\n",
        "Have we benefitted from using the pretrained model?\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QNdU05bisTX1"
      },
      "source": [
        "The neural network was trained using transfer learning. For the purpose of transfer learning, VGG16 was used which is a .... it containes 10 classes and has been trained on 1999 images. It was chosen for this problem as it has a similar  single-label multiclass classification problem.\n",
        "First setting: Using VGG16 without fine-tuning, number of trainable parameters: \n",
        "Second setting: Using VGG16 with fine-tuning, number of trainable parameters: "
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ouK5NY-_pLDK"
      },
      "source": [
        "## Task 2 Fast training of deep networks\n",
        "\n",
        "*(weight ~20%)*"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LgoOE2W1pdfN"
      },
      "source": [
        "### Task 2.1 Train a highly accurate network for CIFAR10\n",
        "\n",
        "*(weight ~7%)*\n",
        "\n",
        "In this task, you will train deep neural networks on the [CIFAR10 dataset](https://www.cs.toronto.edu/~kriz/cifar.html). Compared with the datasets that you have worked on so far, CIFAR10 represents a relatively larger multi-class classification problem and presents a great opportunity for you to solve a \"harder\" problem.\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IaD5oqj3lhuI"
      },
      "source": [
        "#### Task 2.1.1 Document the hardware used\n",
        "\n",
        "Before you start, write down your hardware specifications, including \n",
        "\n",
        "- the GPU model, the number of GPUs, and the GPU memory\n",
        "- the CPU model, the number of CPUs, and the CPU clock speed\n",
        "\n",
        "(Hint: you may find commands like `nvidia-smi`, `lscpu` or `psutil` useful.)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FkGuv1zAavu2",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "df0b726e-ea43-4e65-bfc1-1e284a3718bd"
      },
      "source": [
        "import os\n",
        "\n",
        "print(tf.__version__)\n",
        "print(tf.keras.__version__)\n",
        "if 'COLAB_TPU_ADDR' in os.environ: \n",
        "    print('Connected to TPU') \n",
        "elif tf.test.gpu_device_name() is not '': \n",
        "    print('Connected to GPU ' + tf.test.gpu_device_name())\n",
        "else:\n",
        "    print('Neither connected to a TPU nor a GPU')\n",
        "\n",
        "gpu_info = !nvidia-smi\n",
        "gpu_info = '\\n'.join(gpu_info)\n",
        "if gpu_info.find('failed') >= 0:\n",
        "    print('Select the Runtime → \"Change runtime type\" menu to enable a GPU accelerator, ')\n",
        "    print('and then re-execute this cell.')\n",
        "else:\n",
        "    print(gpu_info)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2.4.1\n",
            "2.4.0\n",
            "Connected to GPU /device:GPU:0\n",
            "Fri May 14 14:53:39 2021       \n",
            "+-----------------------------------------------------------------------------+\n",
            "| NVIDIA-SMI 465.19.01    Driver Version: 460.32.03    CUDA Version: 11.2     |\n",
            "|-------------------------------+----------------------+----------------------+\n",
            "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n",
            "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n",
            "|                               |                      |               MIG M. |\n",
            "|===============================+======================+======================|\n",
            "|   0  Tesla K80           Off  | 00000000:00:04.0 Off |                    0 |\n",
            "| N/A   41C    P0    58W / 149W |    124MiB / 11441MiB |      0%      Default |\n",
            "|                               |                      |                  N/A |\n",
            "+-------------------------------+----------------------+----------------------+\n",
            "                                                                               \n",
            "+-----------------------------------------------------------------------------+\n",
            "| Processes:                                                                  |\n",
            "|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |\n",
            "|        ID   ID                                                   Usage      |\n",
            "|=============================================================================|\n",
            "+-----------------------------------------------------------------------------+\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "adN9Tq-6lyG-"
      },
      "source": [
        "#### Task 2.1.2 Train a \"shallow\" ConvNet\n",
        "\n",
        "Build a ConvNet with fewer than 10 layers. Train the network until it converges. You will use this network as a baseline for the later experiments. \n",
        "\n",
        "- Plot the training and validation history. \n",
        "- Report the testing accuracy. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "F8Our-hai998",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "f49c18d3-ba9c-4679-ec65-8053c8de4e6b"
      },
      "source": [
        "from keras.datasets import cifar10\n",
        "from keras.utils import to_categorical\n",
        "\n",
        "(trainX, trainY), (testX, testY) = cifar10.load_data()\n",
        "\t# one hot encode target values\n",
        "trainY = to_categorical(trainY)\n",
        "testY = to_categorical(testY)\n",
        "\n",
        "trainX = trainX.astype('float32')\n",
        "testX = testX.astype('float32')\n",
        "\n",
        "trainX = trainX/255.0\n",
        "testX = testX/255.0\n",
        "\n",
        "\n",
        "print (\"number of training examples = \" + str(trainX.shape[0]))\n",
        "print (\"number of test examples = \" + str(testX.shape[0]))\n",
        "print (\"X_train shape: \" + str(trainX.shape))\n",
        "print (\"Y_train shape: \" + str(trainY.shape))\n",
        "print (\"X_test shape: \" + str(testX.shape))\n",
        "print (\"Y_test shape: \" + str(testY.shape))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "number of training examples = 50000\n",
            "number of test examples = 10000\n",
            "X_train shape: (50000, 32, 32, 3)\n",
            "Y_train shape: (50000, 10)\n",
            "X_test shape: (10000, 32, 32, 3)\n",
            "Y_test shape: (10000, 10)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "aERbt7FFiqoP"
      },
      "source": [
        "model = tf.keras.Sequential()\n",
        "model.add(Conv2D(32, (3, 3), activation='relu', kernel_initializer='he_uniform', padding='same', input_shape=(32, 32, 3)))\n",
        "model.add(Conv2D(32, (3, 3), activation='relu', kernel_initializer='he_uniform', padding='same'))\n",
        "model.add(MaxPooling2D((2, 2)))\n",
        "model.add(Flatten())\n",
        "model.add(Dense(128, activation='relu', kernel_initializer='he_uniform'))\n",
        "model.add(Dense(10, activation='softmax'))\n",
        "# compile model\n",
        "opt = SGD(lr=0.001, momentum=0.9)\n",
        "model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "Pv1fzvAXYP_P",
        "outputId": "97762906-af6b-4770-fc23-ca1c981ab678"
      },
      "source": [
        "history = model.fit(trainX, trainY,\n",
        "          batch_size=128,\n",
        "          epochs=50,\n",
        "          verbose=1,\n",
        "          validation_split=0.3)\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "274/274 [==============================] - 34s 19ms/step - loss: 2.1008 - accuracy: 0.2378 - val_loss: 1.7581 - val_accuracy: 0.3923\n",
            "Epoch 2/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.7008 - accuracy: 0.3992 - val_loss: 1.6232 - val_accuracy: 0.4289\n",
            "Epoch 3/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.5581 - accuracy: 0.4479 - val_loss: 1.5096 - val_accuracy: 0.4663\n",
            "Epoch 4/50\n",
            "274/274 [==============================] - 5s 16ms/step - loss: 1.4585 - accuracy: 0.4889 - val_loss: 1.4383 - val_accuracy: 0.4935\n",
            "Epoch 5/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.3784 - accuracy: 0.5148 - val_loss: 1.3992 - val_accuracy: 0.5089\n",
            "Epoch 6/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.3054 - accuracy: 0.5409 - val_loss: 1.3190 - val_accuracy: 0.5327\n",
            "Epoch 7/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.2443 - accuracy: 0.5643 - val_loss: 1.3110 - val_accuracy: 0.5393\n",
            "Epoch 8/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.1843 - accuracy: 0.5846 - val_loss: 1.2423 - val_accuracy: 0.5656\n",
            "Epoch 9/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.1427 - accuracy: 0.5966 - val_loss: 1.2233 - val_accuracy: 0.5726\n",
            "Epoch 10/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.1015 - accuracy: 0.6181 - val_loss: 1.1784 - val_accuracy: 0.5929\n",
            "Epoch 11/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.0625 - accuracy: 0.6331 - val_loss: 1.2039 - val_accuracy: 0.5867\n",
            "Epoch 12/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 1.0364 - accuracy: 0.6402 - val_loss: 1.1563 - val_accuracy: 0.6033\n",
            "Epoch 13/50\n",
            "274/274 [==============================] - 5s 16ms/step - loss: 0.9867 - accuracy: 0.6576 - val_loss: 1.1206 - val_accuracy: 0.6097\n",
            "Epoch 14/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.9618 - accuracy: 0.6701 - val_loss: 1.1465 - val_accuracy: 0.6089\n",
            "Epoch 15/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.9361 - accuracy: 0.6779 - val_loss: 1.0934 - val_accuracy: 0.6214\n",
            "Epoch 16/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.8988 - accuracy: 0.6926 - val_loss: 1.1008 - val_accuracy: 0.6189\n",
            "Epoch 17/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.8739 - accuracy: 0.7008 - val_loss: 1.1014 - val_accuracy: 0.6180\n",
            "Epoch 18/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.8429 - accuracy: 0.7115 - val_loss: 1.0579 - val_accuracy: 0.6351\n",
            "Epoch 19/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.8151 - accuracy: 0.7233 - val_loss: 1.0825 - val_accuracy: 0.6317\n",
            "Epoch 20/50\n",
            "274/274 [==============================] - 5s 16ms/step - loss: 0.8047 - accuracy: 0.7238 - val_loss: 1.0599 - val_accuracy: 0.6365\n",
            "Epoch 21/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.7601 - accuracy: 0.7410 - val_loss: 1.0537 - val_accuracy: 0.6391\n",
            "Epoch 22/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.7353 - accuracy: 0.7546 - val_loss: 1.0454 - val_accuracy: 0.6425\n",
            "Epoch 23/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.7096 - accuracy: 0.7615 - val_loss: 1.0699 - val_accuracy: 0.6365\n",
            "Epoch 24/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.6883 - accuracy: 0.7674 - val_loss: 1.0741 - val_accuracy: 0.6371\n",
            "Epoch 25/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.6649 - accuracy: 0.7759 - val_loss: 1.0331 - val_accuracy: 0.6483\n",
            "Epoch 26/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.6446 - accuracy: 0.7849 - val_loss: 1.0483 - val_accuracy: 0.6493\n",
            "Epoch 27/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.6167 - accuracy: 0.7940 - val_loss: 1.0430 - val_accuracy: 0.6509\n",
            "Epoch 28/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.5929 - accuracy: 0.8024 - val_loss: 1.0686 - val_accuracy: 0.6463\n",
            "Epoch 29/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.5797 - accuracy: 0.8067 - val_loss: 1.0562 - val_accuracy: 0.6525\n",
            "Epoch 30/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.5353 - accuracy: 0.8225 - val_loss: 1.1359 - val_accuracy: 0.6323\n",
            "Epoch 31/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.5242 - accuracy: 0.8266 - val_loss: 1.0776 - val_accuracy: 0.6501\n",
            "Epoch 32/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.4952 - accuracy: 0.8382 - val_loss: 1.0881 - val_accuracy: 0.6495\n",
            "Epoch 33/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.4776 - accuracy: 0.8457 - val_loss: 1.1073 - val_accuracy: 0.6471\n",
            "Epoch 34/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.4590 - accuracy: 0.8531 - val_loss: 1.1038 - val_accuracy: 0.6517\n",
            "Epoch 35/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.4277 - accuracy: 0.8656 - val_loss: 1.1042 - val_accuracy: 0.6531\n",
            "Epoch 36/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.4117 - accuracy: 0.8731 - val_loss: 1.1343 - val_accuracy: 0.6484\n",
            "Epoch 37/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.3790 - accuracy: 0.8853 - val_loss: 1.1723 - val_accuracy: 0.6446\n",
            "Epoch 38/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.3641 - accuracy: 0.8896 - val_loss: 1.1755 - val_accuracy: 0.6492\n",
            "Epoch 39/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.3431 - accuracy: 0.8984 - val_loss: 1.1899 - val_accuracy: 0.6481\n",
            "Epoch 40/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.3074 - accuracy: 0.9126 - val_loss: 1.2310 - val_accuracy: 0.6489\n",
            "Epoch 41/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.3016 - accuracy: 0.9123 - val_loss: 1.2233 - val_accuracy: 0.6503\n",
            "Epoch 42/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.2748 - accuracy: 0.9218 - val_loss: 1.2853 - val_accuracy: 0.6390\n",
            "Epoch 43/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.2608 - accuracy: 0.9281 - val_loss: 1.3242 - val_accuracy: 0.6383\n",
            "Epoch 44/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.2380 - accuracy: 0.9361 - val_loss: 1.3496 - val_accuracy: 0.6383\n",
            "Epoch 45/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.2199 - accuracy: 0.9423 - val_loss: 1.3110 - val_accuracy: 0.6477\n",
            "Epoch 46/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.2038 - accuracy: 0.9480 - val_loss: 1.3332 - val_accuracy: 0.6493\n",
            "Epoch 47/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.1805 - accuracy: 0.9580 - val_loss: 1.3704 - val_accuracy: 0.6469\n",
            "Epoch 48/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.1600 - accuracy: 0.9660 - val_loss: 1.3886 - val_accuracy: 0.6476\n",
            "Epoch 49/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.1479 - accuracy: 0.9690 - val_loss: 1.4280 - val_accuracy: 0.6501\n",
            "Epoch 50/50\n",
            "274/274 [==============================] - 5s 17ms/step - loss: 0.1414 - accuracy: 0.9691 - val_loss: 1.4712 - val_accuracy: 0.6469\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "error",
          "ename": "NameError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-13-cfa1e4a3d923>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      5\u001b[0m           validation_split=0.3)\n\u001b[1;32m      6\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 7\u001b[0;31m \u001b[0macc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mft_history\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mhistory\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'accuracy'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      8\u001b[0m \u001b[0mval_acc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mft_history\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mhistory\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'val_accuracy'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      9\u001b[0m \u001b[0mloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mft_history\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mhistory\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'loss'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mNameError\u001b[0m: name 'ft_history' is not defined"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        },
        "id": "9OuyMYnYZ9cX",
        "outputId": "9ac9ff84-da06-4ba6-f5ab-adf5524d5f99"
      },
      "source": [
        "acc = history.history['accuracy']\n",
        "val_acc = history.history['val_accuracy']\n",
        "loss = history.history['loss']\n",
        "val_loss = history.history['val_loss']\n",
        "epochs = range(1, len(acc) + 1)\n",
        "\n",
        "plt.figure(figsize=(15,4))\n",
        "\n",
        "ax1 = plt.subplot(1, 2, 1)\n",
        "plt.plot(epochs, loss, label='Training loss')\n",
        "plt.plot(epochs, val_loss, label='Validation loss')\n",
        "plt.fill_between(epochs, loss,val_loss,color='g',alpha=.1)\n",
        "\n",
        "plt.title('Training and validation loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "\n",
        "ax2 = plt.subplot(1, 2, 2)\n",
        "plt.plot(epochs, acc, label='Training accuracy')\n",
        "plt.plot(epochs, val_acc, label='Validation accuracy')\n",
        "plt.fill_between(epochs, acc,val_acc,color='g',alpha=.1)\n",
        "plt.title('Training and validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9NTjVIUkmv7S"
      },
      "source": [
        "#### Task 2.1.3 Train a ResNet\n",
        "\n",
        "Train a residual neural network (ResNet) on the CIFAR10 training data and report the test accuracy and the training time.\n",
        "\n",
        "The ResNet is a popular network architecture for image classification. You may find more information about how ResNet works by reading this [paper](https://arxiv.org/abs/1512.03385).\n",
        "\n",
        "\n",
        "*(You may implement a resnet model or use an existing implementation. In either case, you should not use pretrained network weights.)*"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ls8FzX76hCtu"
      },
      "source": [
        "from keras.layers import Input, Add, Dense, Activation, ZeroPadding2D, BatchNormalization, Flatten, Conv2D, AveragePooling2D, MaxPooling2D, GlobalMaxPooling2D\n",
        "from keras.models import Model, load_model\n",
        "from keras.preprocessing import image\n",
        "from keras.utils import layer_utils\n",
        "from keras.utils.data_utils import get_file\n",
        "from keras.applications.imagenet_utils import preprocess_input\n",
        "import pydot\n",
        "from IPython.display import SVG\n",
        "from keras.utils.vis_utils import model_to_dot\n",
        "from keras.utils import plot_model\n",
        "#from resnets_utils import *\n",
        "from keras.initializers import glorot_uniform\n",
        "import scipy.misc\n",
        "from matplotlib.pyplot import imshow\n",
        "%matplotlib inline\n",
        "\n",
        "import keras.backend as K"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l0VC_8udgSJ7"
      },
      "source": [
        "# model has been taken from\n",
        "\n",
        "def identity_block(X, f, filters, stage, block):\n",
        "    \"\"\"\n",
        "    Implementation of the identity block as defined in Figure 3\n",
        "    \n",
        "    Arguments:\n",
        "    X -- input tensor of shape (m, n_H_prev, n_W_prev, n_C_prev)\n",
        "    f -- integer, specifying the shape of the middle CONV's window for the main path\n",
        "    filters -- python list of integers, defining the number of filters in the CONV layers of the main path\n",
        "    stage -- integer, used to name the layers, depending on their position in the network\n",
        "    block -- string/character, used to name the layers, depending on their position in the network\n",
        "    \n",
        "    Returns:\n",
        "    X -- output of the identity block, tensor of shape (n_H, n_W, n_C)\n",
        "    \"\"\"\n",
        "    \n",
        "    # defining name basis\n",
        "    conv_name_base = 'res' + str(stage) + block + '_branch'\n",
        "    bn_name_base = 'bn' + str(stage) + block + '_branch'\n",
        "    \n",
        "    # Retrieve Filters\n",
        "    F1, F2, F3 = filters\n",
        "    \n",
        "    # Save the input value. You'll need this later to add back to the main path. \n",
        "    X_shortcut = X\n",
        "    \n",
        "    # First component of main path\n",
        "    X = Conv2D(filters = F1, kernel_size = (1, 1), strides = (1,1), padding = 'same', name = conv_name_base + '2a', kernel_initializer = glorot_uniform(seed=0))(X)\n",
        "    X = BatchNormalization(axis = 3, name = bn_name_base + '2a')(X)\n",
        "    X = Activation('relu')(X)\n",
        "\n",
        "    \n",
        "    # Second component of main path (≈3 lines)\n",
        "    X = Conv2D(filters = F2, kernel_size = (f, f), strides = (1,1), padding = 'same', name = conv_name_base + '2b', kernel_initializer = glorot_uniform(seed=0))(X)\n",
        "    X = BatchNormalization(axis = 3, name = bn_name_base + '2b')(X)\n",
        "    X = Activation('relu')(X)\n",
        "\n",
        "    # Third component of main path (≈2 lines)\n",
        "    X = Conv2D(filters = F3, kernel_size = (1, 1), strides = (1,1), padding = 'same', name = conv_name_base + '2c', kernel_initializer = glorot_uniform(seed=0))(X)\n",
        "    X = BatchNormalization(axis = 3, name = bn_name_base + '2c')(X)\n",
        "\n",
        "    # Final step: Add shortcut value to main path, and pass it through a RELU activation (≈2 lines)\n",
        "    X = Add()([X, X_shortcut])\n",
        "    X = Activation('relu')(X)\n",
        "    \n",
        "    \n",
        "    return X"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "crC_G4D7gSM0"
      },
      "source": [
        "def convolutional_block(X, f, filters, stage, block, s = 2):\n",
        "    \"\"\"\n",
        "    Implementation of the convolutional block as defined in Figure 4\n",
        "    \n",
        "    Arguments:\n",
        "    X -- input tensor of shape (m, n_H_prev, n_W_prev, n_C_prev)\n",
        "    f -- integer, specifying the shape of the middle CONV's window for the main path\n",
        "    filters -- python list of integers, defining the number of filters in the CONV layers of the main path\n",
        "    stage -- integer, used to name the layers, depending on their position in the network\n",
        "    block -- string/character, used to name the layers, depending on their position in the network\n",
        "    s -- Integer, specifying the stride to be used\n",
        "    \n",
        "    Returns:\n",
        "    X -- output of the convolutional block, tensor of shape (n_H, n_W, n_C)\n",
        "    \"\"\"\n",
        "    \n",
        "    # defining name basis\n",
        "    conv_name_base = 'res' + str(stage) + block + '_branch'\n",
        "    bn_name_base = 'bn' + str(stage) + block + '_branch'\n",
        "    \n",
        "    # Retrieve Filters\n",
        "    F1, F2, F3 = filters\n",
        "    \n",
        "    # Save the input value\n",
        "    X_shortcut = X\n",
        "\n",
        "\n",
        "    ##### MAIN PATH #####\n",
        "    # First component of main path \n",
        "    X = Conv2D(F1, (1, 1), strides = (s,s), name = conv_name_base + '2a', padding = 'same', kernel_initializer = glorot_uniform(seed=0))(X)\n",
        "    X = BatchNormalization(axis = 3, name = bn_name_base + '2a')(X)\n",
        "    X = Activation('relu')(X)\n",
        "\n",
        "    # Second component of main path (≈3 lines)\n",
        "    X = Conv2D(filters = F2, kernel_size = (f, f), strides = (1,1), padding = 'same', name = conv_name_base + '2b', kernel_initializer = glorot_uniform(seed=0))(X)\n",
        "    X = BatchNormalization(axis = 3, name = bn_name_base + '2b')(X)\n",
        "    X = Activation('relu')(X)\n",
        "\n",
        "\n",
        "    # Third component of main path (≈2 lines)\n",
        "    X = Conv2D(filters = F3, kernel_size = (1, 1), strides = (1,1), padding = 'same', name = conv_name_base + '2c', kernel_initializer = glorot_uniform(seed=0))(X)\n",
        "    X = BatchNormalization(axis = 3, name = bn_name_base + '2c')(X)\n",
        "\n",
        "\n",
        "    ##### SHORTCUT PATH #### (≈2 lines)\n",
        "    X_shortcut = Conv2D(filters = F3, kernel_size = (1, 1), strides = (s,s), padding = 'same', name = conv_name_base + '1',\n",
        "                        kernel_initializer = glorot_uniform(seed=0))(X_shortcut)\n",
        "    X_shortcut = BatchNormalization(axis = 3, name = bn_name_base + '1')(X_shortcut)\n",
        "\n",
        "    # Final step: Add shortcut value to main path, and pass it through a RELU activation (≈2 lines)\n",
        "    X = Add()([X, X_shortcut])\n",
        "    X = Activation('relu')(X)\n",
        "    \n",
        "    \n",
        "    return X"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0A5R3qWvgSPE"
      },
      "source": [
        "def ResNet50(input_shape=(32, 32, 3), classes=10):\n",
        "    \"\"\"\n",
        "    Implementation of the popular ResNet50 the following architecture:\n",
        "    CONV2D -> BATCHNORM -> RELU -> MAXPOOL -> CONVBLOCK -> IDBLOCK*2 -> CONVBLOCK -> IDBLOCK*3\n",
        "    -> CONVBLOCK -> IDBLOCK*5 -> CONVBLOCK -> IDBLOCK*2 -> AVGPOOL -> TOPLAYER\n",
        "\n",
        "    Arguments:\n",
        "    input_shape -- shape of the images of the dataset\n",
        "    classes -- integer, number of classes\n",
        "\n",
        "    Returns:\n",
        "    model -- a Model() instance in Keras\n",
        "    \"\"\"\n",
        "\n",
        "    # Define the input as a tensor with shape input_shape\n",
        "    X_input = Input(input_shape)\n",
        "\n",
        "    # Zero-Padding\n",
        "    X = ZeroPadding2D((3, 3))(X_input)\n",
        "\n",
        "    # Stage 1\n",
        "    X = Conv2D(64, (7, 7), strides=(2, 2), name='conv1', padding = 'same', kernel_initializer=glorot_uniform(seed=0))(X)\n",
        "    X = BatchNormalization(axis=3, name='bn_conv1')(X)\n",
        "    X = Activation('relu')(X)\n",
        "    X = MaxPooling2D((3, 3), strides=(2, 2))(X)\n",
        "\n",
        "    # Stage 2\n",
        "    X = convolutional_block(X, f=3, filters=[64, 64, 256], stage=2, block='a', s=1)\n",
        "    X = identity_block(X, 3, [64, 64, 256], stage=2, block='b')\n",
        "    X = identity_block(X, 3, [64, 64, 256], stage=2, block='c')\n",
        "\n",
        "    ### START CODE HERE ###\n",
        "\n",
        "    # Stage 3 (≈4 lines)\n",
        "    X = convolutional_block(X, f = 3, filters = [128, 128, 512], stage = 3, block='a', s = 2)\n",
        "    X = identity_block(X, 3, [128, 128, 512], stage=3, block='b')\n",
        "    X = identity_block(X, 3, [128, 128, 512], stage=3, block='c')\n",
        "    X = identity_block(X, 3, [128, 128, 512], stage=3, block='d')\n",
        "\n",
        "    # Stage 4 (≈6 lines)\n",
        "    X = convolutional_block(X, f = 3, filters = [256, 256, 1024], stage = 4, block='a', s = 2)\n",
        "    X = identity_block(X, 3, [256, 256, 1024], stage=4, block='b')\n",
        "    X = identity_block(X, 3, [256, 256, 1024], stage=4, block='c')\n",
        "    X = identity_block(X, 3, [256, 256, 1024], stage=4, block='d')\n",
        "    X = identity_block(X, 3, [256, 256, 1024], stage=4, block='e')\n",
        "    X = identity_block(X, 3, [256, 256, 1024], stage=4, block='f')\n",
        "\n",
        "    # Stage 5 (≈3 lines)\n",
        "    X = convolutional_block(X, f = 3, filters = [512, 512, 2048], stage = 5, block='a', s = 2)\n",
        "    X = identity_block(X, 3, [512, 512, 2048], stage=5, block='b')\n",
        "    X = identity_block(X, 3, [512, 512, 2048], stage=5, block='c')\n",
        "\n",
        "    # AVGPOOL (≈1 line). Use \"X = AveragePooling2D(...)(X)\"\n",
        "    X = AveragePooling2D((2,2), name=\"avg_pool\")(X)\n",
        "\n",
        "    ### END CODE HERE ###\n",
        "\n",
        "    # output layer\n",
        "    X = Flatten()(X)\n",
        "    X = Dense(classes, activation='softmax', name='fc' + str(classes), kernel_initializer = glorot_uniform(seed=0))(X)\n",
        "    \n",
        "    \n",
        "    # Create model\n",
        "    model = Model(inputs = X_input, outputs = X, name='ResNet50')\n",
        "\n",
        "    return model"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2NPhPC5ygSRy"
      },
      "source": [
        "model = ResNet50(input_shape = (32, 32, 3), classes = 10)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TuZz_42rgxDT"
      },
      "source": [
        "model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2JP6a4AwgxFf",
        "outputId": "328a30bd-9ac3-4ac0-b14b-8b5e8e7f28be"
      },
      "source": [
        "history = model.fit(trainX, trainY, epochs = 25, batch_size = 128, validation_split=0.3)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/25\n",
            "274/274 [==============================] - 34s 105ms/step - loss: 2.6798 - accuracy: 0.2709 - val_loss: 9.2174 - val_accuracy: 0.1005\n",
            "Epoch 2/25\n",
            "274/274 [==============================] - 28s 102ms/step - loss: 1.7045 - accuracy: 0.4206 - val_loss: 2.9059 - val_accuracy: 0.4373\n",
            "Epoch 3/25\n",
            "274/274 [==============================] - 27s 100ms/step - loss: 1.4073 - accuracy: 0.5152 - val_loss: 1.6513 - val_accuracy: 0.4373\n",
            "Epoch 4/25\n",
            "274/274 [==============================] - 28s 100ms/step - loss: 1.3752 - accuracy: 0.5265 - val_loss: 1.4069 - val_accuracy: 0.5225\n",
            "Epoch 5/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 1.0728 - accuracy: 0.6206 - val_loss: 1.0812 - val_accuracy: 0.6157\n",
            "Epoch 6/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.9405 - accuracy: 0.6697 - val_loss: 1.1847 - val_accuracy: 0.6005\n",
            "Epoch 7/25\n",
            "274/274 [==============================] - 28s 100ms/step - loss: 0.7962 - accuracy: 0.7250 - val_loss: 1.0408 - val_accuracy: 0.6544\n",
            "Epoch 8/25\n",
            "274/274 [==============================] - 28s 100ms/step - loss: 0.6811 - accuracy: 0.7606 - val_loss: 2.2654 - val_accuracy: 0.4123\n",
            "Epoch 9/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.6272 - accuracy: 0.7812 - val_loss: 1.3110 - val_accuracy: 0.5890\n",
            "Epoch 10/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.5730 - accuracy: 0.8017 - val_loss: 4.4399 - val_accuracy: 0.3232\n",
            "Epoch 11/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.5966 - accuracy: 0.7923 - val_loss: 1.0659 - val_accuracy: 0.6667\n",
            "Epoch 12/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.4666 - accuracy: 0.8372 - val_loss: 1.0928 - val_accuracy: 0.6718\n",
            "Epoch 13/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.4716 - accuracy: 0.8386 - val_loss: 1.0515 - val_accuracy: 0.6827\n",
            "Epoch 14/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.3412 - accuracy: 0.8819 - val_loss: 1.2080 - val_accuracy: 0.6653\n",
            "Epoch 15/25\n",
            "274/274 [==============================] - 28s 100ms/step - loss: 0.2345 - accuracy: 0.9192 - val_loss: 1.2977 - val_accuracy: 0.6665\n",
            "Epoch 16/25\n",
            "274/274 [==============================] - 28s 100ms/step - loss: 0.1771 - accuracy: 0.9388 - val_loss: 1.3935 - val_accuracy: 0.6554\n",
            "Epoch 17/25\n",
            "274/274 [==============================] - 27s 100ms/step - loss: 0.1515 - accuracy: 0.9482 - val_loss: 1.3044 - val_accuracy: 0.6755\n",
            "Epoch 18/25\n",
            "274/274 [==============================] - 27s 100ms/step - loss: 0.1974 - accuracy: 0.9322 - val_loss: 1.3454 - val_accuracy: 0.6910\n",
            "Epoch 19/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.1491 - accuracy: 0.9503 - val_loss: 1.5466 - val_accuracy: 0.6673\n",
            "Epoch 20/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.1246 - accuracy: 0.9567 - val_loss: 1.5633 - val_accuracy: 0.6633\n",
            "Epoch 21/25\n",
            "274/274 [==============================] - 27s 100ms/step - loss: 0.1147 - accuracy: 0.9599 - val_loss: 1.5183 - val_accuracy: 0.6827\n",
            "Epoch 22/25\n",
            "274/274 [==============================] - 28s 101ms/step - loss: 0.1017 - accuracy: 0.9652 - val_loss: 1.3489 - val_accuracy: 0.7005\n",
            "Epoch 23/25\n",
            "274/274 [==============================] - 27s 100ms/step - loss: 0.0928 - accuracy: 0.9679 - val_loss: 1.4820 - val_accuracy: 0.6829\n",
            "Epoch 24/25\n",
            "274/274 [==============================] - 27s 100ms/step - loss: 0.0923 - accuracy: 0.9686 - val_loss: 1.4839 - val_accuracy: 0.6787\n",
            "Epoch 25/25\n",
            "274/274 [==============================] - 27s 100ms/step - loss: 0.0823 - accuracy: 0.9724 - val_loss: 1.5507 - val_accuracy: 0.6880\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 232
        },
        "id": "Bd-C3s2TgxIE",
        "outputId": "7f29e8e4-ae39-4a1c-d248-806f40a14ba4"
      },
      "source": [
        "acc = history.history['accuracy']\n",
        "val_acc = history.history['val_accuracy']\n",
        "loss = history.history['loss']\n",
        "val_loss = history.history['val_loss']\n",
        "epochs = range(1, len(acc) + 1)\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "error",
          "ename": "NameError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-50-73cd0c6116b6>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0mepochs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0macc\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      6\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 7\u001b[0;31m \u001b[0mplt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfigure\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfigsize\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m15\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      8\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      9\u001b[0m \u001b[0max1\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mplt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msubplot\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mNameError\u001b[0m: name 'plt' is not defined"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        },
        "id": "FaC62xjqtQOA",
        "outputId": "0550b481-b8c0-4d2a-9d4e-ceb5b70c892b"
      },
      "source": [
        "plt.figure(figsize=(15,4))\n",
        "\n",
        "ax1 = plt.subplot(1, 2, 1)\n",
        "plt.plot(epochs, loss, label='Training loss')\n",
        "plt.plot(epochs, val_loss, label='Validation loss')\n",
        "plt.fill_between(epochs, loss,val_loss,color='g',alpha=.1)\n",
        "\n",
        "plt.title('Training and validation loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "\n",
        "ax2 = plt.subplot(1, 2, 2)\n",
        "plt.plot(epochs, acc, label='Training accuracy')\n",
        "plt.plot(epochs, val_acc, label='Validation accuracy')\n",
        "plt.fill_between(epochs, acc,val_acc,color='g',alpha=.1)\n",
        "plt.title('Training and validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AH6ZBiECzS75"
      },
      "source": [
        "### Task 2.2 Fast training of ResNet\n",
        "\n",
        "*(weight ~10%)*\n",
        "\n",
        "In this task, you will experiment with different ways to reduce the time for training your ResNet on CIFAR10. There are different ways to speed up neural network training; below are two ideas. Please select at least one idea to implement. Explain the experiment steps and report the final performance and training time.\n",
        "\n",
        "#### Option 1. Learning rate schedule\n",
        "\n",
        "Use a learning rate schedule for the training. Some popular learning rate schedules include \n",
        "\n",
        "- the Step Decay learning rate (e.g., see [here](https://github.com/kuangliu/pytorch-cifar))\n",
        "- [Cyclical learning rates](https://arxiv.org/abs/1506.01186)\n",
        "- [The exponential learning rate](https://openreview.net/forum?id=rJg8TeSFDH) \n",
        "\n",
        "Also, Keras provides [some convenient functions](https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/schedules) that you can use.\n",
        "\n",
        "\n",
        "#### Option 2. Look ahead optimiser\n",
        "\n",
        "Read [this paper](https://arxiv.org/abs/1907.08610) and implement the Lookahead optimiser."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MrgTps8IrU2M"
      },
      "source": [
        "initial_learning_rate = 0.1\n",
        "lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(\n",
        "    initial_learning_rate,\n",
        "    decay_steps=100000,\n",
        "    decay_rate=0.96,\n",
        "    staircase=True)\n",
        "\n",
        "# model.compile(optimizer=tf.keras.optimizers.SGD(learning_rate=lr_schedule),\n",
        "#               loss='sparse_categorical_crossentropy',\n",
        "#               metrics=['accuracy'])\n",
        "\n",
        "model.compile(optimizer=tf.keras.optimizers.SGD(learning_rate=lr_schedule), loss='categorical_crossentropy', metrics=['accuracy'])\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "KWTRZ1Lytyxq",
        "outputId": "560ee9e4-84fe-4386-95bc-44c97ca36036"
      },
      "source": [
        "history = model.fit(trainX, trainY, epochs = 50, batch_size = 128, validation_split=0.3)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "274/274 [==============================] - 31s 102ms/step - loss: 4.3708 - accuracy: 0.1995 - val_loss: 2.1470 - val_accuracy: 0.2313\n",
            "Epoch 2/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 1.6856 - accuracy: 0.3980 - val_loss: 1.6104 - val_accuracy: 0.4349\n",
            "Epoch 3/50\n",
            "274/274 [==============================] - 27s 97ms/step - loss: 1.4685 - accuracy: 0.4684 - val_loss: 1.7151 - val_accuracy: 0.4005\n",
            "Epoch 4/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 1.3491 - accuracy: 0.5142 - val_loss: 1.4295 - val_accuracy: 0.4850\n",
            "Epoch 5/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 1.2393 - accuracy: 0.5515 - val_loss: 1.5368 - val_accuracy: 0.4947\n",
            "Epoch 6/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 1.1344 - accuracy: 0.5907 - val_loss: 1.6536 - val_accuracy: 0.4622\n",
            "Epoch 7/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 1.0299 - accuracy: 0.6307 - val_loss: 1.4013 - val_accuracy: 0.5437\n",
            "Epoch 8/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.9269 - accuracy: 0.6703 - val_loss: 1.4689 - val_accuracy: 0.5201\n",
            "Epoch 9/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.8334 - accuracy: 0.7053 - val_loss: 1.5372 - val_accuracy: 0.5409\n",
            "Epoch 10/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.7474 - accuracy: 0.7320 - val_loss: 1.3329 - val_accuracy: 0.5657\n",
            "Epoch 11/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.6561 - accuracy: 0.7648 - val_loss: 2.2400 - val_accuracy: 0.4361\n",
            "Epoch 12/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.5721 - accuracy: 0.7981 - val_loss: 1.5241 - val_accuracy: 0.5633\n",
            "Epoch 13/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.4763 - accuracy: 0.8317 - val_loss: 1.6112 - val_accuracy: 0.5339\n",
            "Epoch 14/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.4314 - accuracy: 0.8476 - val_loss: 1.7161 - val_accuracy: 0.5393\n",
            "Epoch 15/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.3503 - accuracy: 0.8765 - val_loss: 2.1237 - val_accuracy: 0.5199\n",
            "Epoch 16/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.3085 - accuracy: 0.8903 - val_loss: 1.7149 - val_accuracy: 0.5921\n",
            "Epoch 17/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.2650 - accuracy: 0.9076 - val_loss: 2.3394 - val_accuracy: 0.5327\n",
            "Epoch 18/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.2350 - accuracy: 0.9166 - val_loss: 2.2167 - val_accuracy: 0.5297\n",
            "Epoch 19/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.1922 - accuracy: 0.9330 - val_loss: 1.8854 - val_accuracy: 0.6058\n",
            "Epoch 20/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.1813 - accuracy: 0.9363 - val_loss: 2.4047 - val_accuracy: 0.5598\n",
            "Epoch 21/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.1646 - accuracy: 0.9415 - val_loss: 2.2882 - val_accuracy: 0.5646\n",
            "Epoch 22/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.1340 - accuracy: 0.9541 - val_loss: 2.9475 - val_accuracy: 0.5280\n",
            "Epoch 23/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.1413 - accuracy: 0.9517 - val_loss: 2.3100 - val_accuracy: 0.5841\n",
            "Epoch 24/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.1249 - accuracy: 0.9591 - val_loss: 2.3929 - val_accuracy: 0.5863\n",
            "Epoch 25/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.1111 - accuracy: 0.9621 - val_loss: 2.2426 - val_accuracy: 0.6146\n",
            "Epoch 26/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0964 - accuracy: 0.9661 - val_loss: 2.4595 - val_accuracy: 0.5669\n",
            "Epoch 27/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0984 - accuracy: 0.9665 - val_loss: 2.2999 - val_accuracy: 0.6080\n",
            "Epoch 28/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0786 - accuracy: 0.9732 - val_loss: 5.3214 - val_accuracy: 0.4201\n",
            "Epoch 29/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.1147 - accuracy: 0.9650 - val_loss: 2.3267 - val_accuracy: 0.6041\n",
            "Epoch 30/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0752 - accuracy: 0.9741 - val_loss: 2.5161 - val_accuracy: 0.5811\n",
            "Epoch 31/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0735 - accuracy: 0.9746 - val_loss: 2.2109 - val_accuracy: 0.6294\n",
            "Epoch 32/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0634 - accuracy: 0.9768 - val_loss: 2.4684 - val_accuracy: 0.6162\n",
            "Epoch 33/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0622 - accuracy: 0.9783 - val_loss: 2.8884 - val_accuracy: 0.5435\n",
            "Epoch 34/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0905 - accuracy: 0.9704 - val_loss: 2.5157 - val_accuracy: 0.6135\n",
            "Epoch 35/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0576 - accuracy: 0.9809 - val_loss: 2.4354 - val_accuracy: 0.6079\n",
            "Epoch 36/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0481 - accuracy: 0.9841 - val_loss: 2.8333 - val_accuracy: 0.5971\n",
            "Epoch 37/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0550 - accuracy: 0.9813 - val_loss: 2.6592 - val_accuracy: 0.5895\n",
            "Epoch 38/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0594 - accuracy: 0.9796 - val_loss: 2.5134 - val_accuracy: 0.6117\n",
            "Epoch 39/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0517 - accuracy: 0.9837 - val_loss: 2.8524 - val_accuracy: 0.5913\n",
            "Epoch 40/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0637 - accuracy: 0.9770 - val_loss: 3.2078 - val_accuracy: 0.5687\n",
            "Epoch 41/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0559 - accuracy: 0.9819 - val_loss: 2.7068 - val_accuracy: 0.6105\n",
            "Epoch 42/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0444 - accuracy: 0.9848 - val_loss: 2.7780 - val_accuracy: 0.6097\n",
            "Epoch 43/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0415 - accuracy: 0.9856 - val_loss: 2.5118 - val_accuracy: 0.6359\n",
            "Epoch 44/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0343 - accuracy: 0.9883 - val_loss: 2.7502 - val_accuracy: 0.6152\n",
            "Epoch 45/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0360 - accuracy: 0.9876 - val_loss: 2.6630 - val_accuracy: 0.6227\n",
            "Epoch 46/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0382 - accuracy: 0.9862 - val_loss: 3.1498 - val_accuracy: 0.5680\n",
            "Epoch 47/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0442 - accuracy: 0.9846 - val_loss: 3.2287 - val_accuracy: 0.5698\n",
            "Epoch 48/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0427 - accuracy: 0.9844 - val_loss: 2.8921 - val_accuracy: 0.6103\n",
            "Epoch 49/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0398 - accuracy: 0.9863 - val_loss: 2.8975 - val_accuracy: 0.6131\n",
            "Epoch 50/50\n",
            "274/274 [==============================] - 27s 98ms/step - loss: 0.0387 - accuracy: 0.9868 - val_loss: 2.6665 - val_accuracy: 0.6224\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "y44c8M4xqqew"
      },
      "source": [
        "from optimizers import Lookahead\n",
        "opt = tf.keras.optimizers.SGD(learning_rate)\n",
        "opt = tfa.optimizers.Lookahead(opt, k = 5, alpha = 0.5)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "C8cKfAOjpn7c"
      },
      "source": [
        "### Task 2.3 Performance comparison\n",
        "\n",
        "*(weight ~3%)*\n",
        "\n",
        "Based on the above experiments, which method or which combination of methods result in the best accuracy with the same training time."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eWQm8ZfBxDH9"
      },
      "source": [
        "## Task 3 (**HD level task**) Research on new models\n",
        "*(weight ~10%)*\n",
        "\n",
        "Today, ResNet has become a very mature ConvNet architecture. In this task, you will research one recent ConvNet architecture. You may choose an architecture from the reference list below.\n",
        "\n",
        "Write a short report for your research, covering these points:\n",
        "\n",
        "- Identify the main issues that your chosen architecture aims to address. (For example, does it try to reduce the number of parameters or to speed up the training?)\n",
        "- What measures the architecture used to reduce the number of parameters, or reducing the training cost, or improving the model performance?\n",
        "\n",
        "Implement the architecture and compare its performance on CIFAR10 with ResNet. You may include your implementation, experiments, and analyses here in this notebook.\n",
        "\n",
        "**References**:\n",
        "\n",
        "1. Huang G, Liu Z, Van Der Maaten L, Weinberger KQ. *Densely connected convolutional networks*. In Proceedings of the IEEE conference on computer vision and pattern recognition 2017 (pp. 4700-4708).\n",
        "2. Zhang X, Zhou X, Lin M, Sun J. *Shufflenet: An extremely efficient convolutional neural network for mobile devices*. In Proceedings of the IEEE conference on computer vision and pattern recognition 2018 (pp. 6848-6856).\n",
        "3. Tan M, Le Q. *Efficientnet: Rethinking model scaling for convolutional neural networks*. In International Conference on Machine Learning 2019 May 24 (pp. 6105-6114). PMLR.\n",
        "4. Hu J, Shen L, Sun G. *Squeeze-and-excitation networks*. In Proceedings of the IEEE conference on computer vision and pattern recognition 2018 (pp. 7132-7141).\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hUV0wuZ01DNA"
      },
      "source": [
        "---\n",
        "**END OF ASSIGNMENT TWO**"
      ]
    }
  ]
}