01_part3_Keras.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "01_part3_Keras.ipynb",
      "provenance": [],
      "toc_visible": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/MatchLab-Imperial/deep-learning-course/blob/master/01_part3_Keras.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bb1Zl0tVSKCN"
      },
      "source": [
        "# **Starting with Keras**\n",
        "\n",
        "Keras is a high-level Deep Learning API  written in Python and capable of running on top of TensorFlow, CNTK, or Theano. It was developed with a focus on enabling fast experimentation. In other words, Keras is especially good for beginners, it is modular, minimalist, and it makes everything easy to get a neural network up and running in no time.\n",
        "\n",
        "The following code is based on [this tutorial](https://elitedatascience.com/keras-tutorial-deep-learning-in-python), and it is just a brief introduction to Keras. You can refer to [Keras documentation](https://keras.io/) to learn more details in the future about it."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rF1rW5C1dS4R"
      },
      "source": [
        "## Importing Libraries and Modules\n",
        "\n",
        "Google Colab already has everything you need installed. We highly recommend to use Colab for these tutorials, however, if you want to use your own environment, you will need to set up a Python ecosystem and install Keras on it.\n",
        "\n",
        "You can find multiple guides that will help you with the task of installing Keras on your machine. Here are some examples for [Windows](http://inmachineswetrust.com/posts/deep-learning-setup/) or [Ubuntu](https://medium.com/@naomi.fridman/install-conda-tensorflow-gpu-and-keras-on-ubuntu-18-04-1b403e740e25)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Wzb418Yih5SN"
      },
      "source": [
        "Let's start with an easy example. First of all, we will import NumPy and define the seed for the pseudorandom number generator. Specifying the seed is a common practice because different initializations might lead to varying results."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VTilKf6uRhML"
      },
      "source": [
        "import numpy as np\n",
        "np.random.seed(1234)  # for reproducibility"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ob8EYZ89h96k"
      },
      "source": [
        "Now we are ready to import some of the Keras' modules.\n",
        "\n",
        "First, we will import the `Sequential` module from Keras. A `Sequential` module is a stack of neural network layers. We will need it for creating our first feed-forward neural network."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tE570PePip7f"
      },
      "source": [
        "from keras.models import Sequential"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IC1aRifsj4Eu"
      },
      "source": [
        "As mentioned before, Keras is a wrapper that allows us to use either Theano or TensorFlow as a backend. It is easy to switch between them if required. However, newer versions of Keras will focus on the TensorFlow backend, dropping support for the rest of the backends. Thus, in this tutorial, we will use the TensorFlow backend. Similar to the NumPy pseudorandom number generator, we also fix Keras' pseudorandom number generator for reproducibility."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "81_cpwH9UhRn"
      },
      "source": [
        "import tensorflow as tf\n",
        "tf.random.set_seed(123)"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8zCje2VIWFIV"
      },
      "source": [
        "Next, we import two of the core layers used to build a network on Keras: [`Dense`](https://keras.io/layers/core/#dense) and [`Activation`](https://keras.io/layers/core/#activation). In short, the `Dense` defines a fully-connected network where all the neurons are connected to all of the elements of the input data, and the `Activation` defines a non-linear function applied to the network. We will explain more about those layers in the following tutorials."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ua1JaDKTj5XO"
      },
      "source": [
        "from keras.layers import Dense, Activation"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Vbm3UwLeKdDv"
      },
      "source": [
        "In addition, we need to import some utils, that will help us later with the visualisation and transformation of the dataset."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yUVkzIz3KtXt"
      },
      "source": [
        "from keras.utils import to_categorical\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZBQn3bnIkpOE"
      },
      "source": [
        "## Loading Datasets from Keras\n",
        "\n",
        "Besides the modules to build our neural networks, Keras library also contains a set of [standard datasets](https://keras.io/datasets/). The fact that those datasets are inside Keras is handy and can save us a lot of time when learning how to create our first neural network.\n",
        "\n",
        "Therefore, let's import the [MNIST](https://en.wikipedia.org/wiki/MNIST_database) dataset. MNIST is a popular dataset on the computer vision community that contains a large number of binary images of handwritten digits. Once loaded, Keras allows us to easily split the dataset into the commonly used train and test data splits."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "92F5A9XEIiVQ"
      },
      "source": [
        "from keras.datasets import mnist\n",
        "\n",
        "# the data, shuffled and split between train and test sets\n",
        "(X_train, y_train), (X_test, y_test) = mnist.load_data()"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nuHAmcCaImS8"
      },
      "source": [
        "Automatically, Keras has provided us with the `X_train` and `X_test` images for training and testing, respectively. Moreover, Keras also stores into `y_train` and `y_test` the corresponding numeric labels associated with `X_train` and `X_test`.\n",
        "\n",
        "We can now look into the shape of the imported data and visualise some examples:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l5o9qsfjJdNv",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 480
        },
        "outputId": "3ff13284-bdbf-4e8b-cd9c-a22dae8b8997"
      },
      "source": [
        "print('Image shape: {0}'.format(X_train.shape[1:]))\n",
        "print('Total number of training samples: {0}'.format(X_train.shape[0]))\n",
        "print('Total number of test samples: {0}'.format(X_test.shape[0]))\n",
        "\n",
        "N=5\n",
        "start_val = 0 # pick an element for the code to plot the following N**2 values\n",
        "fig, axes = plt.subplots(N,N)\n",
        "items = list(range(0, 10))\n",
        "for row in range(N):\n",
        "  for col in range(N):\n",
        "    idx = start_val+row+N*col\n",
        "    axes[row,col].imshow(X_train[idx], cmap='gray')\n",
        "    fig.subplots_adjust(hspace=0.5)\n",
        "    y_target = int(y_train[idx])\n",
        "    target = str(items[y_target])\n",
        "    axes[row,col].set_title(target)\n",
        "    axes[row,col].set_xticks([])\n",
        "    axes[row,col].set_yticks([])"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Image shape: (28, 28)\n",
            "Total number of training samples: 60000\n",
            "Total number of test samples: 10000\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 25 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WoNhnQtRNhHu"
      },
      "source": [
        "We have 60000 training samples and 10000 test samples, where each image has a shape of 28×28 pixels. Visualising the data before deploying any algorithm is always a good idea: it is a quick sanity check that can prevent avoidable mistakes."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DohVlsvsN9np"
      },
      "source": [
        "## Preprocessing Keras Datasets\n",
        "\n",
        "In this first example, we train a simple model to classify the digits on MNIST dataset.\n",
        "\n",
        "As will be studied in the lectures, convolutional layers are usually employed to classify images. However, for the time being, we use only `Dense` layers. Thus, we have to reshape the images into a 1D array to have a single dimension, and define an architecture that could be used in any 1D data, *e.g.*, time series. We will see on following weeks that when using Convolutional Neural Networks (CNNs) we must define a 3D shape for images (depth, width, height). CNNs usually have higher performance compared to models with only `Dense` layers when applied to 2D data.\n",
        "\n",
        "Let's transform the 2D images into 1D arrays."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "aa6-T9T9PABQ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "499de934-ea85-4d5b-835b-b2fbf9233a52"
      },
      "source": [
        "X_train_flatten = X_train.reshape(X_train.shape[0], X_train.shape[1] * X_train.shape[2])\n",
        "X_test_flatten = X_test.reshape(X_test.shape[0], X_test.shape[1] * X_test.shape[2])\n",
        "\n",
        "print('New X_train shape: {0}'.format(X_train_flatten.shape))"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "New X_train shape: (60000, 784)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aVLnq7URVwxH"
      },
      "source": [
        "A standard practise is to normalise the dataset to the range $[0, 1]$."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jbOITHGSV9WS"
      },
      "source": [
        "X_train_flatten = X_train_flatten.astype('float32')\n",
        "X_test_flatten = X_test_flatten.astype('float32')\n",
        "X_train_flatten /= 255\n",
        "X_test_flatten /= 255"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eZTwrNtOWNBe"
      },
      "source": [
        "We train a classifier in a supervised manner, and therefore, we need to process the labels so that our pipeline can use them. The `y_train` and `y_test` labels indicate the numerical values belonging to the `X_train` and `X_test` images.\n",
        "\n",
        "We use a Categorical CrossEntropy as the loss function. To feed the loss function with the labels, we need to transform them into a [one-hot encoded vector](https://en.wikipedia.org/wiki/One-hot)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "z-KifrbWXreU",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3d90a986-07e9-4b39-fd14-4d678bd964f1"
      },
      "source": [
        "Y_train_class = to_categorical(y_train, 10)\n",
        "Y_test_class = to_categorical(y_test, 10)\n",
        "\n",
        "print('New Y_train shape: {0}'.format(Y_train_class.shape))"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "New Y_train shape: (60000, 10)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jokxf-U8thD8"
      },
      "source": [
        "To see the difference between one-hot encoded and integer representation, let's print the first element of `y_train` and `Y_train_class`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0oYrx0kStdfy",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c931edbc-bc24-41f3-d739-c011eb207133"
      },
      "source": [
        "print('Label: {0}.    One-hot Vector: {1}'.format(y_train[0], Y_train_class[0]))"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Label: 5.    One-hot Vector: [0. 0. 0. 0. 0. 1. 0. 0. 0. 0.]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OcgAG0OZLZ3W"
      },
      "source": [
        "## Model Architecture\n",
        "\n",
        "\n",
        "Defining a model architecture is a critical step when trying to solve a machine learning problem. In this tutorial, we will understand how models can be built into Keras.\n",
        "\n",
        "First, we need to declare a `Sequential` module. As mentioned, the `Sequential` module allows us to combine neural network layers linearly. We create a model by passing a list of layer instances to the `Sequential` module:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9_sJo3MbS1Lz",
        "outputId": "234d702f-5b69-4eac-f324-5161f056a87d",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "model = Sequential([\n",
        "    Dense(10, input_shape=(784,)),\n",
        "    Activation('softmax'),\n",
        "])"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/keras/src/layers/core/dense.py:87: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.\n",
            "  super().__init__(activity_regularizer=activity_regularizer, **kwargs)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iEVpTbQ9TX_u"
      },
      "source": [
        "The previous model is composed of a dense layer and activation function (Softmax). Another way of defining architectures is to simply add layers via the `.add()` method:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w-i9u2fbTZCk"
      },
      "source": [
        "model = Sequential()\n",
        "model.add(Dense(10, input_dim=784))\n",
        "model.add(Activation('softmax'))"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xGsQ9NTBBHFC"
      },
      "source": [
        "We now have a simple model ready to go!\n",
        "\n",
        "Note that the model needs to know the shape of the input data. For this reason, the first layer in the Sequential model needs information about the input shape (the following layers can automatically infer the shape, and you do not need to specify it). In our case, we tell the network that the `input_shape` is 784. The 784 size vector comes from the flatten operation of our images (28 x 28). If we were using directly images, the input shape would have been (28, 28, 1), which corresponds to the (width, height, depth) of each digit image. The formatting (width, height, depth) is called `channels_last`, which is used by default in Keras. However, Keras also accepts the format (depth, width, height), that is called `channels_first` and it is commonly used in PyTorch. Keep in mind that you can change the formatting by passing the argument `data_format` when defining a layer.\n",
        "\n",
        "Let's print the model shape output:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "h5CZJUssUW1b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "2254380d-d281-46f6-b428-d863d616eab7"
      },
      "source": [
        "print(model.output_shape)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(None, 10)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Kpz5BNJ8UnuP"
      },
      "source": [
        "We define the output of our layer to be 10, that corresponds to the 10 different classes in MNIST (one, two, three, ..., nine.). Keras automatically handles the connections between layers, so there is no need for us to manually set up anything within the architecture.\n",
        "\n",
        "We see how our model looks like by using the `.summary()` method."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "a37n1PQzU3BX",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 193
        },
        "outputId": "de0c097b-4cee-4f2b-fc95-27a5d3c5bf1e"
      },
      "source": [
        "model.summary()"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mModel: \"sequential_1\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"sequential_1\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                        \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape               \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m        Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ dense_1 (\u001b[38;5;33mDense\u001b[0m)                      │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m10\u001b[0m)                  │           \u001b[38;5;34m7,850\u001b[0m │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ activation_1 (\u001b[38;5;33mActivation\u001b[0m)            │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m10\u001b[0m)                  │               \u001b[38;5;34m0\u001b[0m │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Layer (type)                         </span>┃<span style=\"font-weight: bold\"> Output Shape                </span>┃<span style=\"font-weight: bold\">         Param # </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ dense_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                      │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">10</span>)                  │           <span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ activation_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Activation</span>)            │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">10</span>)                  │               <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m7,850\u001b[0m (30.66 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> (30.66 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m7,850\u001b[0m (30.66 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> (30.66 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TPQEWWDuwCzk"
      },
      "source": [
        "In our example, the first row of the table has $784\\cdot 10+ 10=7850$ parameters, where the first term refers to the connections between input data and neurons,  and the second term to the output bias."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "q_zb2cNDVaJi"
      },
      "source": [
        "## Compilation\n",
        "\n",
        "Before training our model, we need to configure the learning process, which is done via the compile method. We must define three important parameters here:\n",
        "\n",
        "*   The loss function that the model will try to minimise.\n",
        "*   The optimiser used to minimise the loss function and update the weights of the network.\n",
        "*   The list of metrics you want the model to compute in every step.\n",
        "\n",
        "In our example, as we are doing classification, we will define the `categorical_crossentropy` as the loss function, and we will compute the accuracy metric. You can refer to Keras documentation to learn more about [losses](https://keras.io/losses/) or [optimisers](https://keras.io/optimizers/). Many other parameters could be tuned, e.g., learning rate, decay factor, or weight normalisation. We will explore in more detail all of these parameters in the following tutorials. For now, we will use the default values for them.\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kHasLrmvWtSf"
      },
      "source": [
        "model.compile(loss='categorical_crossentropy',\n",
        "              optimizer='adam',\n",
        "              metrics=['accuracy'])"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iqImvb8kWznJ"
      },
      "source": [
        "## Training\n",
        "\n",
        "To train the model, we call `.fit()` after compiling the model. We have to define the batch size and the number of total epochs the model is going to train. We are finally ready to start the learning of our classifier!"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "trVlIdWNXUwd",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e89dea3e-66f5-4525-ce7b-02a2a931922f"
      },
      "source": [
        "history = model.fit(X_train_flatten, Y_train_class, batch_size=32, epochs=10)"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 1ms/step - accuracy: 0.8190 - loss: 0.7082\n",
            "Epoch 2/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9136 - loss: 0.3100\n",
            "Epoch 3/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m5s\u001b[0m 1ms/step - accuracy: 0.9199 - loss: 0.2849\n",
            "Epoch 4/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 1ms/step - accuracy: 0.9239 - loss: 0.2735\n",
            "Epoch 5/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 1ms/step - accuracy: 0.9264 - loss: 0.2664\n",
            "Epoch 6/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 1ms/step - accuracy: 0.9276 - loss: 0.2615\n",
            "Epoch 7/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m5s\u001b[0m 1ms/step - accuracy: 0.9290 - loss: 0.2577\n",
            "Epoch 8/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 1ms/step - accuracy: 0.9295 - loss: 0.2548\n",
            "Epoch 9/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 1ms/step - accuracy: 0.9304 - loss: 0.2523\n",
            "Epoch 10/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 1ms/step - accuracy: 0.9310 - loss: 0.2502\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yj8MJvUnXzHj"
      },
      "source": [
        "Great, we have trained our first model!\n",
        "\n",
        "Imagine that we need to train the model until it converges, and we do not know how many epochs it will take. Then, we can define some mechanism to control the learning curves and stop them whenever we see fit. Those mechanisms are called callbacks in Keras. Thus, callbacks allow us to define early-stopping rules. Additionally, we use [callbacks methods](https://keras.io/callbacks/) to save the model's weights or log history along the training process, among many other things. An example of using the early stopping callback is given later in this tutorial.\n",
        "\n",
        "Most of the times, the training curves give important information about the training process. To visualise them, we can use the information saved on the training history:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OGy405W2boBB",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 489
        },
        "outputId": "c0010cce-40a6-4198-cf95-fad6671bfb7c"
      },
      "source": [
        "fig, axes = plt.subplots(2,1)\n",
        "axes[0].plot(history.history['accuracy'])\n",
        "axes[0].set_title('Model Accuracy')\n",
        "axes[0].set_ylabel('Accuracy')\n",
        "axes[0].set_xlabel('Epoch')\n",
        "\n",
        "fig.subplots_adjust(hspace=0.5)\n",
        "\n",
        "axes[1].plot(history.history['loss'], 'g')\n",
        "axes[1].set_title('Model Loss')\n",
        "axes[1].set_ylabel('Cross Entropy Loss')\n",
        "axes[1].set_xlabel('Epoch')"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Text(0.5, 0, 'Epoch')"
            ]
          },
          "metadata": {},
          "execution_count": 18
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1xMgl2FYaSA9"
      },
      "source": [
        "## Evaluating Model\n",
        "\n",
        "Finally, we check the metrics of our model on the test data by using the method `.evaluate()`:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Dac7-bRJab05",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "092571db-4ca5-4f49-e8e3-420305c19c54"
      },
      "source": [
        "score = model.evaluate(X_test_flatten, Y_test_class, verbose=0)\n",
        "print('Test loss:', score[0])\n",
        "print('Test accuracy:', score[1])"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Test loss: 0.2644534111022949\n",
            "Test accuracy: 0.9284999966621399\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Jzdmn3rCgo8l"
      },
      "source": [
        "# Training a Simple Multi-layer Perceptron\n",
        "\n",
        "Earlier, we showed how to create a simple network that maps directly from $784$ (input size) to $10$ (output size). In the following sections, we explain the basics of creating more complex models, models that combine different sequential layers to perform more accurate classifications.\n",
        "\n",
        "Hence, we start by defining a network with an extra layer and study how that affects the final test accuracy."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WRRqajaThGgf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 193
        },
        "outputId": "889862c7-6f6b-4c3d-cef2-fda28de2c0b0"
      },
      "source": [
        "model2 = Sequential()\n",
        "model2.add(Dense(100, activation='relu', input_shape=(784,)))\n",
        "model2.add(Dense(10, activation='softmax'))\n",
        "\n",
        "model2.summary()"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mModel: \"sequential_2\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"sequential_2\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                        \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape               \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m        Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ dense_2 (\u001b[38;5;33mDense\u001b[0m)                      │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m100\u001b[0m)                 │          \u001b[38;5;34m78,500\u001b[0m │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ dense_3 (\u001b[38;5;33mDense\u001b[0m)                      │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m10\u001b[0m)                  │           \u001b[38;5;34m1,010\u001b[0m │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Layer (type)                         </span>┃<span style=\"font-weight: bold\"> Output Shape                </span>┃<span style=\"font-weight: bold\">         Param # </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ dense_2 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                      │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">100</span>)                 │          <span style=\"color: #00af00; text-decoration-color: #00af00\">78,500</span> │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ dense_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                      │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">10</span>)                  │           <span style=\"color: #00af00; text-decoration-color: #00af00\">1,010</span> │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m79,510\u001b[0m (310.59 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">79,510</span> (310.59 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m79,510\u001b[0m (310.59 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">79,510</span> (310.59 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RK4s-S-ihT2v"
      },
      "source": [
        "We have increased the number of parameters from 7850 to 79510 by adding this layer. Remember that the number of units given to the last layer in Keras defines the dimensionality of the output space, thus, we need the last layer to have the same size that the total number of classes to classify.\n",
        "\n",
        "Note that we use ReLU (Rectified Linear Unit) activation function after the first dense layer. ReLU is a common activation function, we will give more details about it in future tutorials. However, for now, you can learn more about ReLU [here](https://keras.io/activations/#relu).\n",
        "\n",
        "We train our new model and visualise the model curves:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ni6ov50rkltw",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 837
        },
        "outputId": "9c3dcc68-a222-4770-bcd5-9a703c584d7b"
      },
      "source": [
        "model2.compile(loss='categorical_crossentropy',\n",
        "              optimizer='adam',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "history2 = model2.fit(X_train_flatten, Y_train_class, batch_size=32, epochs=10)\n",
        "\n",
        "fig, axes = plt.subplots(2,1)\n",
        "axes[0].plot(history2.history['accuracy'])\n",
        "axes[0].set_title('Model Accuracy')\n",
        "axes[0].set_ylabel('Accuracy')\n",
        "axes[0].set_xlabel('Epoch')\n",
        "\n",
        "fig.subplots_adjust(hspace=0.5)\n",
        "\n",
        "axes[1].plot(history2.history['loss'], 'g')\n",
        "axes[1].set_title('Model Loss')\n",
        "axes[1].set_ylabel('Cross Entropy Loss')\n",
        "axes[1].set_xlabel('Epoch')"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 3ms/step - accuracy: 0.8754 - loss: 0.4520\n",
            "Epoch 2/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9597 - loss: 0.1375\n",
            "Epoch 3/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m7s\u001b[0m 3ms/step - accuracy: 0.9732 - loss: 0.0906\n",
            "Epoch 4/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m5s\u001b[0m 2ms/step - accuracy: 0.9812 - loss: 0.0668\n",
            "Epoch 5/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9853 - loss: 0.0522\n",
            "Epoch 6/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 3ms/step - accuracy: 0.9885 - loss: 0.0408\n",
            "Epoch 7/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m5s\u001b[0m 2ms/step - accuracy: 0.9916 - loss: 0.0320\n",
            "Epoch 8/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9937 - loss: 0.0249\n",
            "Epoch 9/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 4ms/step - accuracy: 0.9951 - loss: 0.0196\n",
            "Epoch 10/10\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9961 - loss: 0.0159\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Text(0.5, 0, 'Epoch')"
            ]
          },
          "metadata": {},
          "execution_count": 21
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LDcMSVr_l6C3"
      },
      "source": [
        "We have now trained two different models. Let's plot their metrics to see how that extra layer affected the results in the network performance."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cnk6_wuemCXO",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e53d5501-be7c-4090-a080-91fad9a796d0"
      },
      "source": [
        "print('Old model:')\n",
        "print('Test loss:', score[0])\n",
        "print('Test accuracy:', score[1])\n",
        "\n",
        "print('New model:')\n",
        "score2 = model2.evaluate(X_test_flatten, Y_test_class, verbose=0)\n",
        "print('Test loss:', score2[0])\n",
        "print('Test accuracy:', score2[1])"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Old model:\n",
            "Test loss: 0.2644534111022949\n",
            "Test accuracy: 0.9284999966621399\n",
            "New model:\n",
            "Test loss: 0.08987264335155487\n",
            "Test accuracy: 0.9772999882698059\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "a-DHj7P7n1Eh"
      },
      "source": [
        "We have improved accuracy results on the test set. Even though there is still margin for parameter tuning, the reached accuracy is already pretty high. We could add more layers or change the number of neurons in each layer to see if we could boost even further the results.\n",
        "\n",
        "Let's now explore some techniques that will prove useful in the following tutorials."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9XyUx6ThaT47"
      },
      "source": [
        "## Saving/Loading Model\n",
        "\n",
        "Some networks require long training times (hours, days or even weeks), hence, it is essential to know how to save the models for using them in future times without the need of retraining them every time.\n",
        "\n",
        "We can save and load the trained model in different ways.\n",
        "\n",
        "The first way is to save everything into a single HDF5 file, which will contain:\n",
        "\n",
        "*   the architecture of the model, allowing to re-create the model\n",
        "*   the weights of the model\n",
        "*   the training configuration (loss, optimiser)\n",
        "*   the state of the optimiser, which permits us to resume training exactly where you left it off.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0KJp2Pj2aTOc",
        "outputId": "1b2006ff-26c7-41d3-803f-4f79228d929b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "from keras.models import load_model\n",
        "\n",
        "model.save('my_model.h5')  # creates a HDF5 file 'my_model.h5'\n",
        "del model  # deletes the existing model\n",
        "\n",
        "# Load a compiled model identical to the previous one\n",
        "model = load_model('my_model.h5')"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`. \n",
            "WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vD5h5qsUblji"
      },
      "source": [
        "An alternative method is to save the architecture and the weights independently.\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dJXTYfTDcaWs"
      },
      "source": [
        "model.save_weights('my_model.weights.h5') # creates a HDF5 file 'my_model_weights.h5'\n",
        "\n",
        "with open('my_model_architecture.json', 'w') as f:\n",
        "    f.write(model.to_json()) # creates a json file 'my_model_architecture.json'"
      ],
      "execution_count": 24,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wPgCrD4Leony"
      },
      "source": [
        "If we want to load the model, without having to define it, we can use the `model_from_json()`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gXr0L8nqeipI"
      },
      "source": [
        "from keras.models import model_from_json\n",
        "\n",
        "# Model reconstruction from JSON file\n",
        "with open('my_model_architecture.json', 'r') as f:\n",
        "    model = model_from_json(f.read())"
      ],
      "execution_count": 25,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NzmL07T_dh7Z"
      },
      "source": [
        "Once we have a model defined, we can load the weights with the method `load_weights(file_path)`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yBahatkweEZx",
        "outputId": "eddeb7de-56dd-41e4-e069-f7460a46f7ff",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "model.load_weights('my_model.weights.h5')"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/keras/src/saving/saving_lib.py:713: UserWarning: Skipping variable loading for optimizer 'adam', because it has 6 variables whereas the saved optimizer has 2 variables. \n",
            "  saveable.load_own_variables(weights_store.get(inner_path))\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UWyW0AMhGMDc"
      },
      "source": [
        "Additionally, you can refer to [Keras' documentation](https://keras.io/getting_started/faq/#how-can-i-regularly-save-keras-models-during-training) and check how to create a callback which saves our model every 100 steps."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mX2BJRnjIxmN"
      },
      "source": [
        "### Download Models / Save to Google Drive\n",
        "One of the problems you may face is that the models, or any other file, saved in Colab will not be there permanently. In some cases, you will want to store your model weights in a more lasting way. There are two ways to save your model. First, you can download any file to your computer. To do so, you can use the left-side menu in Colab, and follow the instructions on the image below.\n",
        "\n",
        "![Screenshot](https://i.ibb.co/yS5JgPd/Screenshot-2021-01-18-at-15-46-04.png)\n",
        "\n",
        "If you want to use any file you previously downloaded, you will need to manually upload your file using the Colab interface.\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ez688r6XKFrn"
      },
      "source": [
        "Another way to save your weights is by using your Google Drive storage. This can be more convenient, as you can quickly load the models again from your Drive without having to upload manually the file. To do so, you can click on the folder with the Drive symbol highlighted in the following image.\n",
        "\n",
        "![Screenshot](https://i.ibb.co/PFGw6QR/Screenshot-2021-01-18-at-16-16-44.png)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xZ_A8BuHL1pi"
      },
      "source": [
        "After clicking on the Drive folder, the code below will appear in your Colab notebook. You need to run it and follow the instructions to have access to your Google Drive folder."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I-tn782jL7lM",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "79751032-2fb6-43db-d918-743bc8975b06"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zLZ7pcurNgkK"
      },
      "source": [
        "After the `drive` folder is mounted, you will see a new folder on your files section in Colab as in the following image.\n",
        "\n",
        "![Screenshot](https://i.ibb.co/NSM1RFK/Screenshot-2021-01-18-at-16-17-48.png)\n",
        "\n",
        "Now, you can save the model to your personal Google Drive and also load any model from it."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C-0qOFZpNMmP"
      },
      "source": [
        "model.save_weights('/content/drive/MyDrive/my_model.weights.h5')\n",
        "model.load_weights('/content/drive/MyDrive/my_model.weights.h5')"
      ],
      "execution_count": 28,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nJVamZKmPDJU"
      },
      "source": [
        "Using your Google Drive storage is quite convenient. However, keep in mind that any file saved in your drive will also count towards your Google Drive storage limit."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-6v_S7oxfUQC"
      },
      "source": [
        "## Obtaining an Output of an Intermediate Layer\n",
        "\n",
        "Sometimes we need to check how features of intermediate layers look like. This can be used for extracting features, but also for debugging purposes. That is why we need to see how we can obtain information about any layer within the architecture.\n",
        "\n",
        "The easiest way is to design a new model that will have as final output our desired layer. We first define the model:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dZa6sDFLgXoC",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 257
        },
        "outputId": "813d7202-f7ac-422f-e94a-c9399758e9e2"
      },
      "source": [
        "from keras.models import Model\n",
        "from keras.layers import Input, Dense\n",
        "\n",
        "inputs = Input(shape=(784,), name='input_layer')\n",
        "first_dense = Dense(64, activation='relu', name='first_dense')(inputs)\n",
        "second_dense = Dense(128, activation='relu', name='second_dense')(first_dense)\n",
        "final_dense = Dense(10, activation='softmax', name='final_dense')(second_dense)\n",
        "model = Model(inputs=inputs, outputs=final_dense)\n",
        "model.summary()"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mModel: \"functional_9\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"functional_9\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                        \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape               \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m        Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ input_layer (\u001b[38;5;33mInputLayer\u001b[0m)             │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m784\u001b[0m)                 │               \u001b[38;5;34m0\u001b[0m │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ first_dense (\u001b[38;5;33mDense\u001b[0m)                  │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m64\u001b[0m)                  │          \u001b[38;5;34m50,240\u001b[0m │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ second_dense (\u001b[38;5;33mDense\u001b[0m)                 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m128\u001b[0m)                 │           \u001b[38;5;34m8,320\u001b[0m │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ final_dense (\u001b[38;5;33mDense\u001b[0m)                  │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m10\u001b[0m)                  │           \u001b[38;5;34m1,290\u001b[0m │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Layer (type)                         </span>┃<span style=\"font-weight: bold\"> Output Shape                </span>┃<span style=\"font-weight: bold\">         Param # </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ input_layer (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)             │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">784</span>)                 │               <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ first_dense (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                  │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)                  │          <span style=\"color: #00af00; text-decoration-color: #00af00\">50,240</span> │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ second_dense (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)                 │           <span style=\"color: #00af00; text-decoration-color: #00af00\">8,320</span> │\n",
              "├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤\n",
              "│ final_dense (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                  │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">10</span>)                  │           <span style=\"color: #00af00; text-decoration-color: #00af00\">1,290</span> │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m59,850\u001b[0m (233.79 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">59,850</span> (233.79 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m59,850\u001b[0m (233.79 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">59,850</span> (233.79 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hoXvhyLzhKU2"
      },
      "source": [
        "Now, we specify the name of the layer in which we are interested in obtaining the output. In our example, it is the second dense layer (\"second_dense\")."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8Zu29bschSps",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4d3f1dda-bdd4-463b-9d8c-c216835bc7ab"
      },
      "source": [
        "from keras.models import Model\n",
        "\n",
        "layer_name = 'second_dense'\n",
        "intermediate_layer_model = Model(inputs=model.input,\n",
        "                                 outputs=model.get_layer(layer_name).output)\n",
        "intermediate_output = intermediate_layer_model.predict(X_test_flatten)\n",
        "final_output = model.predict(X_test_flatten)\n",
        "\n",
        "print('Second dense layer output shape: {0}'.format(intermediate_output.shape))\n",
        "print('Final output shape: {0}'.format(final_output.shape))"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[1m313/313\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step\n",
            "\u001b[1m313/313\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 2ms/step\n",
            "Second dense layer output shape: (10000, 128)\n",
            "Final output shape: (10000, 10)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iV22i-GFj5np"
      },
      "source": [
        "## A bit more about Callback Functions\n",
        "\n",
        "We have already talked a bit about callback functions and how to use them. You can check the complete list of available callbacks [in the documentation](https://keras.io/callbacks/).\n",
        "\n",
        "Controlling and managing callbacks is essential when training big models. Sometimes we want to train the network until convergence. However, we are not able to determine the number of epochs it will take. Keras provides a handy method for controlling the training loops, the `EarlyStopping` callback. This callback checks when the network is not \"learning\" anymore and allows us to stop the training at any time.\n",
        "\n",
        "Let's see an example of how to stop the training when the validation loss is not decreasing anymore:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GNShEzC-kdKN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ba0b7ef0-f169-4816-d35b-3024697aa9f1"
      },
      "source": [
        "from keras.callbacks import EarlyStopping\n",
        "\n",
        "# Define the metric that will determine the earlyStopping.\n",
        "# 'patience' defines the number of epochs we wait before stopping training\n",
        "# when the monitored metric does not improve\n",
        "early_stopping = EarlyStopping(monitor='val_loss', patience=2)\n",
        "\n",
        "model.compile(loss='categorical_crossentropy',\n",
        "              optimizer='adam',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "model.fit(X_train_flatten, Y_train_class, validation_split=0.2,\n",
        "            epochs=100, batch_size=32, callbacks=[early_stopping])"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 3ms/step - accuracy: 0.8480 - loss: 0.5141 - val_accuracy: 0.9537 - val_loss: 0.1558\n",
            "Epoch 2/100\n",
            "\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 3ms/step - accuracy: 0.9563 - loss: 0.1456 - val_accuracy: 0.9628 - val_loss: 0.1224\n",
            "Epoch 3/100\n",
            "\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m5s\u001b[0m 3ms/step - accuracy: 0.9711 - loss: 0.0987 - val_accuracy: 0.9657 - val_loss: 0.1134\n",
            "Epoch 4/100\n",
            "\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m5s\u001b[0m 3ms/step - accuracy: 0.9782 - loss: 0.0734 - val_accuracy: 0.9694 - val_loss: 0.1040\n",
            "Epoch 5/100\n",
            "\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9846 - loss: 0.0540 - val_accuracy: 0.9694 - val_loss: 0.1061\n",
            "Epoch 6/100\n",
            "\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9869 - loss: 0.0435 - val_accuracy: 0.9709 - val_loss: 0.1082\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.src.callbacks.history.History at 0x79c28d290d30>"
            ]
          },
          "metadata": {},
          "execution_count": 31
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "meMroifllkIN"
      },
      "source": [
        "The model stops training without reaching the 100 epochs. How convenient, right?\n",
        "\n",
        "We could also combine multiple callbacks when calling the fit module, learn more about it [here](https://keras.io/callbacks/)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3w4AVBmLCDJe"
      },
      "source": [
        "## Freezing Keras Layers\n",
        "\n",
        "Keras also allows us to freeze some of the weights of specific layers. Freezing layers means that we can exclude them from training. A typical example where this proved useful is when fine-tuning a model. Remember that when fine-tuning a model, some layers are fixed, and normally only the last ones are trained to finetune the task-specific layers (e.g. classifiers) to the new task.\n",
        "\n",
        "To freeze layers we can set the trainable argument (Boolean) of the layer to be non-trainable or trainable:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Q6x8F5hZDLWH"
      },
      "source": [
        "frozen_layer = Dense(32, trainable=False)"
      ],
      "execution_count": 32,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3s-5Bm8ODNe0"
      },
      "source": [
        "Additionally, we can access the layer arguments after initialisation. Therefore, we can set the trainable argument to `True` or `False` at any time. However, for this to take effect, we must call the `compile()` method after modifying any properties of the model:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MTnjPeE7DjEs",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 861
        },
        "outputId": "889b1c9c-2b23-4f13-d871-451ef47e8fe1"
      },
      "source": [
        "from keras.models import clone_model\n",
        "\n",
        "model = Sequential()\n",
        "model.add(Dense(10, activation='softmax', input_shape=(784,)))\n",
        "\n",
        "# We clone model in frozen_model\n",
        "frozen_model = clone_model(model)\n",
        "frozen_model.layers[0].trainable = False\n",
        "\n",
        "# with this model the weights of the layer will be updated during training\n",
        "model.compile(loss='categorical_crossentropy',\n",
        "              optimizer='adam',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "# in the model below, the weights of `layer` will not be updated during training\n",
        "frozen_model.compile(loss='categorical_crossentropy',\n",
        "              optimizer='adam',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "model.summary()\n",
        "frozen_model.summary()\n",
        "\n",
        "score_untrained = model.evaluate(X_test_flatten, Y_test_class, verbose=0)\n",
        "score_frozen_untrained = frozen_model.evaluate(X_test_flatten, Y_test_class, verbose=0)\n",
        "\n",
        "print('\\nTrain Model:')\n",
        "model.fit(X_train_flatten, Y_train_class, batch_size=32, epochs=5)\n",
        "\n",
        "print('\\nTrain Frozen Model:')\n",
        "frozen_model.fit(X_train_flatten, Y_train_class, batch_size=32, epochs=5)\n",
        "\n",
        "score = model.evaluate(X_test_flatten, Y_test_class, verbose=0)\n",
        "score_frozen = frozen_model.evaluate(X_test_flatten, Y_test_class, verbose=0)\n",
        "\n",
        "print('\\nBefore Training: Trainable Model Test Accuracy:', score_untrained[1])\n",
        "print('After Training:  Trainable Model Test Accuracy:', score[1])\n",
        "\n",
        "print('\\nBefore Training: Frozen Model Test Accuracy:', score_frozen_untrained[1])\n",
        "print('After Training:  Frozen Model Test Accuracy:', score_frozen[1])"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mModel: \"sequential_3\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"sequential_3\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                        \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape               \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m        Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ dense_5 (\u001b[38;5;33mDense\u001b[0m)                      │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m10\u001b[0m)                  │           \u001b[38;5;34m7,850\u001b[0m │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Layer (type)                         </span>┃<span style=\"font-weight: bold\"> Output Shape                </span>┃<span style=\"font-weight: bold\">         Param # </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ dense_5 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                      │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">10</span>)                  │           <span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m7,850\u001b[0m (30.66 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> (30.66 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m7,850\u001b[0m (30.66 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> (30.66 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mModel: \"sequential_3\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"sequential_3\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                        \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape               \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m        Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ dense_5 (\u001b[38;5;33mDense\u001b[0m)                      │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m10\u001b[0m)                  │           \u001b[38;5;34m7,850\u001b[0m │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Layer (type)                         </span>┃<span style=\"font-weight: bold\"> Output Shape                </span>┃<span style=\"font-weight: bold\">         Param # </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩\n",
              "│ dense_5 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                      │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">10</span>)                  │           <span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> │\n",
              "└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m7,850\u001b[0m (30.66 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> (30.66 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m7,850\u001b[0m (30.66 KB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">7,850</span> (30.66 KB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Train Model:\n",
            "Epoch 1/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.8161 - loss: 0.7139\n",
            "Epoch 2/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 1ms/step - accuracy: 0.9140 - loss: 0.3102\n",
            "Epoch 3/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 1ms/step - accuracy: 0.9202 - loss: 0.2851\n",
            "Epoch 4/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9235 - loss: 0.2736\n",
            "Epoch 5/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9256 - loss: 0.2665\n",
            "\n",
            "Train Frozen Model:\n",
            "Epoch 1/5\n",
            "\u001b[1m  93/1875\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.0764 - loss: 2.4031"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/keras/src/backend/tensorflow/trainer.py:75: UserWarning: The model does not have any trainable weights.\n",
            "  warnings.warn(\"The model does not have any trainable weights.\")\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 1ms/step - accuracy: 0.0786 - loss: 2.4025\n",
            "Epoch 2/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 1ms/step - accuracy: 0.0786 - loss: 2.4025\n",
            "Epoch 3/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 1ms/step - accuracy: 0.0786 - loss: 2.4025\n",
            "Epoch 4/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 1ms/step - accuracy: 0.0786 - loss: 2.4025\n",
            "Epoch 5/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.0786 - loss: 2.4025\n",
            "\n",
            "Before Training: Trainable Model Test Accuracy: 0.06970000267028809\n",
            "After Training:  Trainable Model Test Accuracy: 0.9269999861717224\n",
            "\n",
            "Before Training: Frozen Model Test Accuracy: 0.08060000091791153\n",
            "After Training:  Frozen Model Test Accuracy: 0.08060000091791153\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mLmeoTdqhmhX"
      },
      "source": [
        "We see in the model description that the frozen model has 0 trainable parameters, while it has 7,850 non-trainable parameters. This fact indicates that we successfully froze that model and the training does not affect it. We can verify this using the results after training both networks, where the trainable model improves accuracy while the frozen one does not change at all."
      ]
    }
  ]
}