TP_ann_mlp_part_2_students.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RIjFqlgERHAq"
      },
      "source": [
        "# IMA-205\n",
        "# TP ANN (part B)\n",
        "\n",
        "## Coding a Multi-Layer Perceptron with Tensorflow"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oG_lwX5GRHAs"
      },
      "source": [
        "In the second part of this TP, we will be looking at the Multi-Layer Perceptron (MLP) using Tensorflow (http://www.tensorflow.org)\n",
        "\n",
        "We will be using the following packages :\n",
        "   \n",
        "   - Scikit-learn (http://scikit-learn.org/)\n",
        "   - Tensorflow (http://www.tensorflow.org)\n",
        "\n",
        "The following commands will make sure that you have all the necessary packages"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 33,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sCGFw05kRHAs",
        "outputId": "d5e9f183-18ba-41a9-ebc7-cd3548f00d7d"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "2.12.0\n"
          ]
        }
      ],
      "source": [
        "%matplotlib inline                      \n",
        "\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import os\n",
        "import pdb\n",
        "\n",
        "import sklearn  # scikit-learn\n",
        "import tensorflow as tf\n",
        "\n",
        "# import tensorflow models\n",
        "from tensorflow.keras.models import Sequential, Model\n",
        "from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten, Input\n",
        "from tensorflow.keras.layers import Conv2D, MaxPooling2D\n",
        "from tensorflow.keras import optimizers\n",
        "print(tf.keras.__version__)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dxgAcMrMRHAt"
      },
      "source": [
        "# 1 - Multi-Layer Perceptron with Tensorflow"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OjK-B5mkRHAt"
      },
      "source": [
        "## Introduction to TensorFlow"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gzVmdesCRHAu"
      },
      "source": [
        "Tensorflow is an environment written by Google which allows easy implementation of deep neural networks. In particular, it provides automatic differentiation so that the user does not have to determine gradients manually, which can be extremely long even for simple networks, as you have seen in the previous TP*.\n",
        "\n",
        "This, in turn allows a user to apply techniques such as stochastic gradient descent for training purposes. The differentiation graph is created symbolically when the network is created. Since a neural network is simply a cascade (or concatenation) of simple functions, the network in Tensorflow is simply a sequence of functions which are applied to one or several variables. The output of the final function is the output of the network.\n",
        "\n",
        "For instance, if the input were ```x``` and the network consisted of two functions ```a``` and ```b```, you could create the network simply by writing :\n",
        "\n",
        "```y = b(a(x))```\n",
        "\n",
        "However, it is clear that these functions and variables must be of some special type, so that the computer can figure out how to carry out the automatic differentiation for training. Tensorflow provides these functions and variables, which must be created with the Tensorflow package.\n",
        "\n",
        "Previously, Tensorflow required a strict separation between the creation (declaration) of variables and their execution (giving a numerical value, using the ```Session``` function). With Keras and Tensorflow 2, this separation has been removed, and the language has become simpler, in particular for creating and training networks.\n",
        "\n",
        "## Creating a network\n",
        "\n",
        "To create a network, there are two (main) methods :\n",
        "\n",
        "### Creating a model using the Sequential API\n",
        "\n",
        "Tensorflow has a simple way of adding layers to create a neural network. First, you can indicate to Tensorflow that the model is 'sequential' (a simple model, with not many tweaks). For this, you can use the following function :\n",
        "- ```model = Sequential()```\n",
        "\n",
        "After this, you can add layers with the function.\n",
        "\n",
        "- ```model.add()```\n",
        "\n",
        "You can then use the ```Dense``` (and other) functions to specify different layer types.  __Note that in the case of this approach, you will have to specify the input image size in the first layer of the network, inside the first layer function.__ So, for example, if the first layer is a dense layer with a relu activation, with n_out output neurons, and n_input neurons :\n",
        "\n",
        "- model = Sequential()\n",
        "- model.add(Dense(n_out,activation,input_shape=(n_in,),activation='relu'))\n",
        "\n",
        "Otherwise, the network does not know how many weights to create. __Be careful of this special case (the first layer)__ \n",
        "\n",
        "### Creating a model using the standard API\n",
        "\n",
        "Otherwise, another approach to creating the model is to explicitly create the input variable, and just cascade the different functions, as in Tensorflow. So, for the same example, we would have :\n",
        "\n",
        "- input = Input(shape=(n_in,))\n",
        "- output = Dense(n_in,n_out,activation='relu')(input)\n",
        "- model = Model(input, output)\n",
        "\n",
        "\n",
        "For now, let's use the Sequential API (however, if you want to try the standard API, go ahead).\n",
        "\n",
        "## Training and testing a network\n",
        "\n",
        "Tensorflow allows the easy training of a network with the following functions :\n",
        "\n",
        "- model.compile(loss='binary_crossentropy', optimizer=optimizers.Adam(lr=learning_rate)) : create the loss function and define the optimiser used. This function can do many other useful things (such as specifying different metrics to look at the model's performance\n",
        "\n",
        "- model.fit(...) : train the model\n",
        "- model.evaluate(...) : test the model\n",
        "- model.predict(...) : carry out a simple forward pass on the model\n",
        "\n",
        "Please look at the Tensorflow documenation for further details on these functions.\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UYFxzt5mRHAy"
      },
      "source": [
        "# MLP with TensorFlow\n",
        "\n",
        "We are now going to create an MLP with tensorflow. We will start with an MLP with one hidden layer. The network should consist of the following operations, in the following order :\n",
        "\n",
        "- Fully connected layer, with 50 output neurons\n",
        "- ReLU activation\n",
        "- Fully connected layer, with 50 output neurons\n",
        "- ReLU activation\n",
        "- Fully connected layer, with 1 output neuron (because we have a binary classification problem)\n",
        "- Sigmoid output activation\n",
        "\n",
        "We are going to be solving a binary classification \n",
        "problem, so the output of the network should be a scalar between 0 and 1 (thus the last layer is a Sigmoid activation).\n",
        "\n",
        "The loss function should be defined as the binary cross-entropy between the predicted class and the true class\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jdGVzE6NRHAy"
      },
      "source": [
        "First, we load the data."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 34,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 430
        },
        "id": "EfiJwuLmRHAy",
        "outputId": "d1b2fee2-741a-4e19-9469-3cdc98365fa4"
      },
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "from sklearn import datasets\n",
        "from sklearn.preprocessing import MinMaxScaler\n",
        "from sklearn.model_selection import train_test_split\n",
        "\n",
        "X, Y = datasets.make_circles(n_samples=1000, noise=0.05, factor=0.5)\n",
        "# X, y = datasets.make_moons(n_samples=1000, random_state=42)  # try with 2 moons\n",
        "X = MinMaxScaler().fit_transform(X)\n",
        "\n",
        "#show data in plot\n",
        "plt.plot(X[Y == 1, 0], X[Y == 1, 1], 'ro')\n",
        "plt.plot(X[Y == 0, 0], X[Y == 0, 1], 'bo')\n",
        "plt.grid('on')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cRL1c6UyRHAy"
      },
      "source": [
        "We are now going to define some useful auxillary functions.\n",
        "\n",
        "First, a function that shows the decision boundary of our network. This works only for 2D input data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 35,
      "metadata": {
        "id": "x0cUhY8xRHAz"
      },
      "outputs": [],
      "source": [
        "# THIS CODE IS GIVEN\n",
        "\n",
        "def plot_decision_function_2d(model_mlp, X, Y):\n",
        "   # create a mesh to plot in\n",
        "    h = .02  # step size in the mesh\n",
        "    offset = 0.1\n",
        "    x_min, x_max = X[:, 0].min() - offset, X[:, 0].max() + offset\n",
        "    y_min, y_max = X[:, 1].min() - offset, X[:, 1].max() + offset\n",
        "    xx, yy = np.meshgrid(np.arange(x_min, x_max, h),\n",
        "                         np.arange(y_min, y_max, h))\n",
        "\n",
        "    # Plot the decision boundary. For that, we will assign a color to each\n",
        "    # point in the mesh [x_min, x_max]x[y_min, y_max].\n",
        "    Z = model_mlp.predict(np.c_[xx.ravel(), yy.ravel()])\n",
        "    print(Z.shape)\n",
        "    print(Z)\n",
        "    \n",
        "    Z = Z<0.5\n",
        "    # Put the result into a color plot\n",
        "    Z = Z.reshape(xx.shape)\n",
        "    plt.figure()\n",
        "    plt.contourf(xx, yy, Z, cmap=plt.cm.Paired)\n",
        "\n",
        "    plt.plot(X[Y == 1, 0], X[Y == 1, 1], 'yo')\n",
        "    plt.plot(X[Y == 0, 0], X[Y == 0, 1], 'ko')\n",
        "\n",
        "    plt.title(\"Decision surface\")\n",
        "    plt.axis('tight')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CKtZJ5iQRHA0"
      },
      "source": [
        "## CREATING AND TRAINING THE MODEL\n",
        "\n",
        "We are now ready to create our network with a Tensorflow Session and to carry out training on our training dataset."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 36,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WO8dHtrGRHA0",
        "outputId": "91b9de65-3a83-44c4-9c7a-de0e33b8813e"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "(850, 2)\n",
            "Training\n",
            "Epoch 1/20\n",
            "14/14 - 1s - loss: 0.6902 - accuracy: 0.5000 - 1s/epoch - 81ms/step\n",
            "Epoch 2/20\n",
            "14/14 - 0s - loss: 0.6849 - accuracy: 0.6506 - 39ms/epoch - 3ms/step\n",
            "Epoch 3/20\n",
            "14/14 - 0s - loss: 0.6810 - accuracy: 0.7741 - 37ms/epoch - 3ms/step\n",
            "Epoch 4/20\n",
            "14/14 - 0s - loss: 0.6779 - accuracy: 0.7765 - 38ms/epoch - 3ms/step\n",
            "Epoch 5/20\n",
            "14/14 - 0s - loss: 0.6758 - accuracy: 0.6988 - 38ms/epoch - 3ms/step\n",
            "Epoch 6/20\n",
            "14/14 - 0s - loss: 0.6737 - accuracy: 0.6635 - 37ms/epoch - 3ms/step\n",
            "Epoch 7/20\n",
            "14/14 - 0s - loss: 0.6716 - accuracy: 0.6388 - 37ms/epoch - 3ms/step\n",
            "Epoch 8/20\n",
            "14/14 - 0s - loss: 0.6695 - accuracy: 0.6259 - 37ms/epoch - 3ms/step\n",
            "Epoch 9/20\n",
            "14/14 - 0s - loss: 0.6673 - accuracy: 0.6247 - 38ms/epoch - 3ms/step\n",
            "Epoch 10/20\n",
            "14/14 - 0s - loss: 0.6649 - accuracy: 0.6588 - 37ms/epoch - 3ms/step\n",
            "Epoch 11/20\n",
            "14/14 - 0s - loss: 0.6630 - accuracy: 0.7118 - 38ms/epoch - 3ms/step\n",
            "Epoch 12/20\n",
            "14/14 - 0s - loss: 0.6601 - accuracy: 0.7024 - 48ms/epoch - 3ms/step\n",
            "Epoch 13/20\n",
            "14/14 - 0s - loss: 0.6578 - accuracy: 0.6765 - 38ms/epoch - 3ms/step\n",
            "Epoch 14/20\n",
            "14/14 - 0s - loss: 0.6558 - accuracy: 0.6447 - 37ms/epoch - 3ms/step\n",
            "Epoch 15/20\n",
            "14/14 - 0s - loss: 0.6527 - accuracy: 0.6671 - 37ms/epoch - 3ms/step\n",
            "Epoch 16/20\n",
            "14/14 - 0s - loss: 0.6502 - accuracy: 0.6871 - 37ms/epoch - 3ms/step\n",
            "Epoch 17/20\n",
            "14/14 - 0s - loss: 0.6472 - accuracy: 0.6741 - 38ms/epoch - 3ms/step\n",
            "Epoch 18/20\n",
            "14/14 - 0s - loss: 0.6444 - accuracy: 0.6682 - 43ms/epoch - 3ms/step\n",
            "Epoch 19/20\n",
            "14/14 - 0s - loss: 0.6418 - accuracy: 0.6729 - 37ms/epoch - 3ms/step\n",
            "Epoch 20/20\n",
            "14/14 - 0s - loss: 0.6388 - accuracy: 0.6871 - 38ms/epoch - 3ms/step\n",
            "Testing\n",
            "5/5 - 0s - loss: 0.6354 - accuracy: 0.6800 - 112ms/epoch - 22ms/step\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "[0.6354228854179382, 0.6800000071525574]"
            ]
          },
          "execution_count": 36,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "# FILL IN CODE BY STUDENTS IN THIS SECTION\n",
        "\n",
        "X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.15)\n",
        "print(X_train.shape)\n",
        "\n",
        "# Layer sizes\n",
        "n_input = X.shape[1]  # Number of input features\n",
        "n_hidden = 50  # Number of hidden nodes\n",
        "\n",
        "# create the model (using the sequential API)\n",
        "\n",
        "model_mlp = Sequential([tf.keras.layers.Dense(n_hidden, activation='relu', input_shape=(n_input,)), tf.keras.layers.Dense(1, activation='sigmoid')])\n",
        "\n",
        "# create the loss and optimiser\n",
        "learning_rate = 0.01\n",
        "model_mlp.compile(loss='binary_crossentropy', optimizer=optimizers.Adam(), metrics=[\"accuracy\"])\n",
        "\n",
        "# Run optimisation algorithm\n",
        "n_epochs = 20\n",
        "batch_size = 64\n",
        "\n",
        "print('Training')\n",
        "model_mlp.fit(X_train, Y_train, epochs=n_epochs, batch_size=batch_size, verbose=2)\n",
        "\n",
        "print('Testing')\n",
        "model_mlp.evaluate(X_test, Y_test, verbose=2)\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 37,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 608
        },
        "id": "lpBaU1rcc0oY",
        "outputId": "3101c36f-1381-4cb8-cb7f-2c96110d3016"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "106/106 [==============================] - 0s 1ms/step\n",
            "(3363, 1)\n",
            "[[0.41434819]\n",
            " [0.41369072]\n",
            " [0.41297355]\n",
            " ...\n",
            " [0.5734925 ]\n",
            " [0.5751171 ]\n",
            " [0.57674015]]\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "plot_decision_function_2d(model_mlp, X_test, Y_test)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6oni6Q5vRHA0"
      },
      "source": [
        "# MLP with several hidden layers\n",
        "\n",
        "We are now going to create an MLP with more hidden layers. We are going to use a more complicated dataset : the MNIST dataset, which contains images of handwritten digits. There are 10 classes, one for each digit.\n",
        "\n",
        "We are going to implement the following architecture :\n",
        "\n",
        "- Fully connected layer\n",
        "- Relu activation\n",
        "- Fully connected layer\n",
        "- Relu activation\n",
        "- Fully connected layer\n",
        "- Relu activation\n",
        "- Fully connected layer\n",
        "- Softmax output activation\n",
        "\n",
        "__IMPORTANT Note__ The sigmoid layer has been replaced by a softmax layer (at the end). This is normal, since we have a multi-class problem.\n",
        "\n",
        "\n",
        "First, we load the MNIST dataset."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 38,
      "metadata": {
        "id": "Z0cGLMKRRHA0"
      },
      "outputs": [],
      "source": [
        "# Import MNIST data\n",
        "mnist = tf.keras.datasets.mnist\n",
        "(X_train, y_train),(X_test, y_test) = mnist.load_data()\n",
        "\n",
        "Y_train = tf.keras.utils.to_categorical(y_train)  # in order to convert y to a matrix with (num_examples, num_classes) (one-hot encoding)\n",
        "Y_test = tf.keras.utils.to_categorical(y_test)  # in order to convert y to a matrix with (num_examples, num_classes) (one-hot encoding)\n",
        "\n",
        "#reshape the input images : flatten the last two dimensions\n",
        "X_train = np.reshape(X_train,(X_train.shape[0],X_train.shape[1]*X_train.shape[2]))\n",
        "X_test = np.reshape(X_test,(X_test.shape[0],X_test.shape[1]*X_test.shape[2]))\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EzMV_9-NRHA1"
      },
      "source": [
        "Finally, fill in the following code to create and train your MLP"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 39,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2G7AToqyRHA1",
        "outputId": "9a8420b2-dd10-491f-ae2c-bf0cdc87363a"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Training\n",
            "Epoch 1/20\n",
            "938/938 - 4s - loss: 1.1350 - accuracy: 0.8878 - 4s/epoch - 5ms/step\n",
            "Epoch 2/20\n",
            "938/938 - 3s - loss: 0.1928 - accuracy: 0.9483 - 3s/epoch - 3ms/step\n",
            "Epoch 3/20\n",
            "938/938 - 2s - loss: 0.1378 - accuracy: 0.9599 - 2s/epoch - 3ms/step\n",
            "Epoch 4/20\n",
            "938/938 - 3s - loss: 0.1197 - accuracy: 0.9656 - 3s/epoch - 3ms/step\n",
            "Epoch 5/20\n",
            "938/938 - 3s - loss: 0.1039 - accuracy: 0.9705 - 3s/epoch - 3ms/step\n",
            "Epoch 6/20\n",
            "938/938 - 3s - loss: 0.0895 - accuracy: 0.9738 - 3s/epoch - 3ms/step\n",
            "Epoch 7/20\n",
            "938/938 - 3s - loss: 0.0844 - accuracy: 0.9751 - 3s/epoch - 3ms/step\n",
            "Epoch 8/20\n",
            "938/938 - 3s - loss: 0.0775 - accuracy: 0.9770 - 3s/epoch - 3ms/step\n",
            "Epoch 9/20\n",
            "938/938 - 3s - loss: 0.0696 - accuracy: 0.9803 - 3s/epoch - 3ms/step\n",
            "Epoch 10/20\n",
            "938/938 - 3s - loss: 0.0668 - accuracy: 0.9811 - 3s/epoch - 3ms/step\n",
            "Epoch 11/20\n",
            "938/938 - 3s - loss: 0.0592 - accuracy: 0.9839 - 3s/epoch - 3ms/step\n",
            "Epoch 12/20\n",
            "938/938 - 3s - loss: 0.0563 - accuracy: 0.9842 - 3s/epoch - 3ms/step\n",
            "Epoch 13/20\n",
            "938/938 - 3s - loss: 0.0507 - accuracy: 0.9858 - 3s/epoch - 3ms/step\n",
            "Epoch 14/20\n",
            "938/938 - 3s - loss: 0.0519 - accuracy: 0.9859 - 3s/epoch - 3ms/step\n",
            "Epoch 15/20\n",
            "938/938 - 3s - loss: 0.0408 - accuracy: 0.9894 - 3s/epoch - 3ms/step\n",
            "Epoch 16/20\n",
            "938/938 - 3s - loss: 0.0423 - accuracy: 0.9883 - 3s/epoch - 3ms/step\n",
            "Epoch 17/20\n",
            "938/938 - 3s - loss: 0.0412 - accuracy: 0.9887 - 3s/epoch - 3ms/step\n",
            "Epoch 18/20\n",
            "938/938 - 3s - loss: 0.0339 - accuracy: 0.9911 - 3s/epoch - 3ms/step\n",
            "Epoch 19/20\n",
            "938/938 - 3s - loss: 0.0343 - accuracy: 0.9908 - 3s/epoch - 3ms/step\n",
            "Epoch 20/20\n",
            "938/938 - 3s - loss: 0.0380 - accuracy: 0.9905 - 3s/epoch - 3ms/step\n",
            "Testing\n",
            "313/313 - 1s - loss: 0.1122 - accuracy: 0.9778 - 683ms/epoch - 2ms/step\n",
            "313/313 [==============================] - 1s 1ms/step\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "array([[3.86316817e-19, 1.73876647e-16, 3.17987498e-13, ...,\n",
              "        9.99999642e-01, 9.80711070e-22, 1.41017325e-13],\n",
              "       [8.41676284e-13, 2.32542435e-15, 1.00000000e+00, ...,\n",
              "        3.28381358e-14, 1.87620932e-15, 2.28283365e-28],\n",
              "       [2.15400326e-19, 1.00000000e+00, 1.44073472e-17, ...,\n",
              "        4.97222836e-12, 5.66341382e-17, 1.07278556e-16],\n",
              "       ...,\n",
              "       [7.68475131e-15, 9.50242671e-18, 6.17385382e-24, ...,\n",
              "        6.45250153e-22, 8.75315801e-15, 1.14269393e-15],\n",
              "       [4.43381613e-17, 5.65608129e-29, 0.00000000e+00, ...,\n",
              "        8.70715085e-26, 5.01645524e-24, 1.70466947e-22],\n",
              "       [5.03731397e-11, 8.75235587e-15, 2.62618421e-27, ...,\n",
              "        0.00000000e+00, 5.72856825e-20, 1.33738321e-29]], dtype=float32)"
            ]
          },
          "execution_count": 39,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "\n",
        "# Network Parameters\n",
        "n_hidden_1 = 256 # 1st layer number of neurons\n",
        "n_hidden_2 = 256 # 2nd layer number of neurons\n",
        "n_hidden_3 = 128 # 2nd layer number of neurons\n",
        "\n",
        "n_input = X_train.shape[1]\n",
        "n_classes = Y_train.shape[1] # MNIST total classes (0-9 digits)\n",
        "\n",
        "# TO CODE BY STUDENTS\n",
        "\n",
        "model_mlp_multi_layer = Sequential([tf.keras.layers.Dense(n_hidden_1, activation='relu', input_shape=(n_input,)), tf.keras.layers.Dense(n_hidden_2, activation='relu'), tf.keras.layers.Dense(n_hidden_3, activation='relu'), tf.keras.layers.Dense(n_classes, activation='softmax')])\n",
        "\n",
        "# create the loss and optimiser, use 'categorical_crossentropy' in loss\n",
        "loss = tf.keras.losses.CategoricalCrossentropy()\n",
        "model_mlp_multi_layer.compile(loss=loss, optimizer=tf.keras.optimizers.Adam(), metrics=[\"accuracy\"])\n",
        "\n",
        "# Run optimisation algorithm\n",
        "n_epochs = 20\n",
        "batch_size = 64\n",
        "\n",
        "print('Training')\n",
        "model_mlp_multi_layer.fit(X_train, Y_train, epochs=n_epochs, batch_size=batch_size, verbose=2)\n",
        "\n",
        "print('Testing')\n",
        "model_mlp_multi_layer.evaluate(X_test, Y_test, verbose=2)\n",
        "model_mlp_multi_layer.predict(X_test)   "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 40,
      "metadata": {
        "id": "s6Q-JntxluxG"
      },
      "outputs": [],
      "source": [
        "# THIS CODE IS GIVEN\n",
        "\n",
        "def test_mnist_images(model_mlp_multi_layer,X_test):\n",
        "  plt.figure(figsize=(10, 6))\n",
        "  for idx in range(0,10):\n",
        "      plt.subplot(2, 5, idx+1)\n",
        "      rand_ind = np.random.randint(0,X_test.shape[0])\n",
        "      plt.imshow(np.reshape(X_test[rand_ind,:],(28,28)),cmap='gray')\n",
        "      # get prediction\n",
        "      model_prediction = np.argmax(model_mlp_multi_layer.predict(np.expand_dims( X_test[rand_ind,:], axis=0)),axis=1)\n",
        "      plt.title(int(model_prediction))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 41,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 545
        },
        "id": "r2D9RE4apVWW",
        "outputId": "c1738c8d-5182-4898-b90a-56a4bee18d54"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "1/1 [==============================] - 0s 22ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n",
            "1/1 [==============================] - 0s 19ms/step\n",
            "1/1 [==============================] - 0s 21ms/step\n",
            "1/1 [==============================] - 0s 19ms/step\n",
            "1/1 [==============================] - 0s 20ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n",
            "1/1 [==============================] - 0s 16ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 1000x600 with 10 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "test_mnist_images(model_mlp_multi_layer,X_test)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lItdBVnfu7xK"
      },
      "source": [
        "You may observe some difficulties in getting good performance in this case. To improve training, we can turn to __batch_normalisation__ and __regularisation__"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Vgao_U5XRHA2"
      },
      "source": [
        "## BATCH NORMALISATION\n",
        "\n",
        "One approach to improving the convergence of neural network training which we have seen during the lesson is known as batch normalisation, which we have seen during the lesson. This can be implemented very simply in Tensorflow by adding the following layer :\n",
        "\n",
        "- ```BatchNormalization()```\n",
        "\n",
        "Change your model below, and implement this using your code above"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 42,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6Ela5-4XRHA2",
        "outputId": "b1ed75ed-0e37-495d-a325-3441d5fede2b"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Training\n",
            "Epoch 1/20\n",
            "938/938 - 8s - loss: 0.2005 - accuracy: 0.9376 - 8s/epoch - 8ms/step\n",
            "Epoch 2/20\n",
            "938/938 - 4s - loss: 0.0900 - accuracy: 0.9723 - 4s/epoch - 5ms/step\n",
            "Epoch 3/20\n",
            "938/938 - 5s - loss: 0.0681 - accuracy: 0.9782 - 5s/epoch - 5ms/step\n",
            "Epoch 4/20\n",
            "938/938 - 5s - loss: 0.0548 - accuracy: 0.9829 - 5s/epoch - 5ms/step\n",
            "Epoch 5/20\n",
            "938/938 - 4s - loss: 0.0463 - accuracy: 0.9849 - 4s/epoch - 5ms/step\n",
            "Epoch 6/20\n",
            "938/938 - 5s - loss: 0.0361 - accuracy: 0.9885 - 5s/epoch - 5ms/step\n",
            "Epoch 7/20\n",
            "938/938 - 4s - loss: 0.0351 - accuracy: 0.9887 - 4s/epoch - 5ms/step\n",
            "Epoch 8/20\n",
            "938/938 - 5s - loss: 0.0281 - accuracy: 0.9909 - 5s/epoch - 5ms/step\n",
            "Epoch 9/20\n",
            "938/938 - 5s - loss: 0.0284 - accuracy: 0.9903 - 5s/epoch - 5ms/step\n",
            "Epoch 10/20\n",
            "938/938 - 4s - loss: 0.0240 - accuracy: 0.9916 - 4s/epoch - 5ms/step\n",
            "Epoch 11/20\n",
            "938/938 - 5s - loss: 0.0198 - accuracy: 0.9934 - 5s/epoch - 5ms/step\n",
            "Epoch 12/20\n",
            "938/938 - 4s - loss: 0.0225 - accuracy: 0.9923 - 4s/epoch - 5ms/step\n",
            "Epoch 13/20\n",
            "938/938 - 4s - loss: 0.0182 - accuracy: 0.9939 - 4s/epoch - 5ms/step\n",
            "Epoch 14/20\n",
            "938/938 - 5s - loss: 0.0173 - accuracy: 0.9941 - 5s/epoch - 5ms/step\n",
            "Epoch 15/20\n",
            "938/938 - 5s - loss: 0.0147 - accuracy: 0.9949 - 5s/epoch - 5ms/step\n",
            "Epoch 16/20\n",
            "938/938 - 5s - loss: 0.0149 - accuracy: 0.9951 - 5s/epoch - 5ms/step\n",
            "Epoch 17/20\n",
            "938/938 - 4s - loss: 0.0147 - accuracy: 0.9952 - 4s/epoch - 5ms/step\n",
            "Epoch 18/20\n",
            "938/938 - 4s - loss: 0.0118 - accuracy: 0.9960 - 4s/epoch - 5ms/step\n",
            "Epoch 19/20\n",
            "938/938 - 5s - loss: 0.0120 - accuracy: 0.9959 - 5s/epoch - 5ms/step\n",
            "Epoch 20/20\n",
            "938/938 - 4s - loss: 0.0110 - accuracy: 0.9964 - 4s/epoch - 5ms/step\n",
            "Testing\n",
            "313/313 - 1s - loss: 0.0886 - accuracy: 0.9827 - 831ms/epoch - 3ms/step\n",
            "313/313 [==============================] - 1s 2ms/step\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "array([[3.3687331e-10, 5.1957358e-07, 4.1874073e-09, ..., 9.9999940e-01,\n",
              "        4.9096466e-10, 1.5164136e-09],\n",
              "       [6.7423093e-07, 1.5058970e-07, 9.9999893e-01, ..., 6.7709796e-12,\n",
              "        2.8689597e-09, 3.1643389e-11],\n",
              "       [3.0806652e-10, 1.0000000e+00, 3.5449973e-11, ..., 2.1962022e-08,\n",
              "        8.5687191e-09, 1.0727157e-09],\n",
              "       ...,\n",
              "       [2.5625418e-11, 2.1216136e-09, 2.7141177e-13, ..., 3.7880454e-10,\n",
              "        3.9175083e-10, 3.9464699e-11],\n",
              "       [4.4582515e-10, 6.4589427e-09, 5.6565176e-11, ..., 6.3154975e-09,\n",
              "        2.4774932e-05, 8.8180999e-12],\n",
              "       [3.9831256e-11, 1.3012791e-11, 3.4673831e-12, ..., 1.0899139e-13,\n",
              "        1.6864683e-11, 1.9137272e-09]], dtype=float32)"
            ]
          },
          "execution_count": 42,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "from tensorflow.keras.layers import Dense, BatchNormalization\n",
        "\n",
        "# TO CODE BY STUDENTS\n",
        "\n",
        "model_mlp_multi_layer_batchnorm = Sequential([tf.keras.layers.Dense(n_hidden_1, activation='relu', input_shape=(n_input,)), tf.keras.layers.BatchNormalization(), tf.keras.layers.Dense(n_hidden_2, activation='relu'), tf.keras.layers.BatchNormalization(), tf.keras.layers.Dense(n_hidden_3, activation='relu'), tf.keras.layers.BatchNormalization(), tf.keras.layers.Dense(n_classes, activation='softmax')])\n",
        "\n",
        "# create the loss and optimiser, use 'categorical_crossentropy' in loss\n",
        "loss = tf.keras.losses.CategoricalCrossentropy()\n",
        "model_mlp_multi_layer_batchnorm.compile(loss=loss, optimizer=tf.keras.optimizers.Adam(), metrics=[\"accuracy\"])\n",
        "\n",
        "# Run optimisation algorithm\n",
        "n_epochs = 20\n",
        "batch_size = 64\n",
        "\n",
        "print('Training')\n",
        "model_mlp_multi_layer_batchnorm.fit(X_train, Y_train, epochs=n_epochs, batch_size=batch_size, verbose=2)\n",
        "\n",
        "print('Testing')\n",
        "model_mlp_multi_layer_batchnorm.evaluate(X_test, Y_test, verbose=2)\n",
        "model_mlp_multi_layer_batchnorm.predict(X_test)   \n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 43,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 545
        },
        "id": "cSxGzvQ3pg3u",
        "outputId": "dd34ccc8-489e-4a9e-d0b2-3c6256754cb2"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "1/1 [==============================] - 0s 26ms/step\n",
            "1/1 [==============================] - 0s 25ms/step\n",
            "1/1 [==============================] - 0s 26ms/step\n",
            "1/1 [==============================] - 0s 28ms/step\n",
            "1/1 [==============================] - 0s 29ms/step\n",
            "1/1 [==============================] - 0s 29ms/step\n",
            "1/1 [==============================] - 0s 29ms/step\n",
            "1/1 [==============================] - 0s 25ms/step\n",
            "1/1 [==============================] - 0s 31ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 1000x600 with 10 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "test_mnist_images(model_mlp_multi_layer,X_test)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iQPJDop31Nci"
      },
      "source": [
        "## Dropout\n",
        "\n",
        "Often in the process of training neural networks, there is a difference between the training accuracy and the testing accuracy. This is the problem of overfitting. To alleviate this problem, we can turn to _dropout_:\n",
        "\n",
        "```tf.keras.layers.Dropout(rate,...)```\n",
        "\n",
        "where rate is the probability that an input neuron to a layer will get set to 0. You can do this in any layer. Use the previous architecture and try out the dropout. Set the dropout rate to 0.1 on the first layer as a test. Then try it out on all layers (you may not see a great difference here since the mnist database is relatively simple, this is just to show you how it is done). "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 44,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "x-BF-fpo1PpW",
        "outputId": "b401f6f0-7fa7-4dc6-ee84-a09a3dbc2030"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Training\n",
            "Epoch 1/20\n",
            "938/938 - 7s - loss: 0.2421 - accuracy: 0.9262 - 7s/epoch - 8ms/step\n",
            "Epoch 2/20\n",
            "938/938 - 5s - loss: 0.1181 - accuracy: 0.9636 - 5s/epoch - 6ms/step\n",
            "Epoch 3/20\n",
            "938/938 - 5s - loss: 0.0921 - accuracy: 0.9714 - 5s/epoch - 5ms/step\n",
            "Epoch 4/20\n",
            "938/938 - 5s - loss: 0.0748 - accuracy: 0.9758 - 5s/epoch - 5ms/step\n",
            "Epoch 5/20\n",
            "938/938 - 5s - loss: 0.0668 - accuracy: 0.9783 - 5s/epoch - 5ms/step\n",
            "Epoch 6/20\n",
            "938/938 - 5s - loss: 0.0569 - accuracy: 0.9814 - 5s/epoch - 5ms/step\n",
            "Epoch 7/20\n",
            "938/938 - 5s - loss: 0.0504 - accuracy: 0.9836 - 5s/epoch - 5ms/step\n",
            "Epoch 8/20\n",
            "938/938 - 5s - loss: 0.0466 - accuracy: 0.9846 - 5s/epoch - 5ms/step\n",
            "Epoch 9/20\n",
            "938/938 - 6s - loss: 0.0420 - accuracy: 0.9858 - 6s/epoch - 7ms/step\n",
            "Epoch 10/20\n",
            "938/938 - 5s - loss: 0.0397 - accuracy: 0.9868 - 5s/epoch - 5ms/step\n",
            "Epoch 11/20\n",
            "938/938 - 4s - loss: 0.0360 - accuracy: 0.9877 - 4s/epoch - 5ms/step\n",
            "Epoch 12/20\n",
            "938/938 - 5s - loss: 0.0336 - accuracy: 0.9890 - 5s/epoch - 5ms/step\n",
            "Epoch 13/20\n",
            "938/938 - 4s - loss: 0.0320 - accuracy: 0.9891 - 4s/epoch - 5ms/step\n",
            "Epoch 14/20\n",
            "938/938 - 5s - loss: 0.0291 - accuracy: 0.9902 - 5s/epoch - 6ms/step\n",
            "Epoch 15/20\n",
            "938/938 - 4s - loss: 0.0265 - accuracy: 0.9911 - 4s/epoch - 5ms/step\n",
            "Epoch 16/20\n",
            "938/938 - 5s - loss: 0.0267 - accuracy: 0.9913 - 5s/epoch - 5ms/step\n",
            "Epoch 17/20\n",
            "938/938 - 5s - loss: 0.0246 - accuracy: 0.9915 - 5s/epoch - 5ms/step\n",
            "Epoch 18/20\n",
            "938/938 - 5s - loss: 0.0239 - accuracy: 0.9920 - 5s/epoch - 5ms/step\n",
            "Epoch 19/20\n",
            "938/938 - 5s - loss: 0.0212 - accuracy: 0.9929 - 5s/epoch - 5ms/step\n",
            "Epoch 20/20\n",
            "938/938 - 5s - loss: 0.0208 - accuracy: 0.9928 - 5s/epoch - 5ms/step\n",
            "Testing\n",
            "313/313 - 1s - loss: 0.0676 - accuracy: 0.9823 - 813ms/epoch - 3ms/step\n",
            "313/313 [==============================] - 1s 2ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n",
            "1/1 [==============================] - 0s 19ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n",
            "1/1 [==============================] - 0s 19ms/step\n",
            "1/1 [==============================] - 0s 16ms/step\n",
            "1/1 [==============================] - 0s 16ms/step\n",
            "1/1 [==============================] - 0s 18ms/step\n",
            "1/1 [==============================] - 0s 17ms/step\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 1000x600 with 10 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "from tensorflow.keras.layers import Dense, BatchNormalization, Dropout\n",
        "\n",
        "# TO CODE BY STUDENTS\n",
        "\n",
        "model_mlp_multi_layer_batchnorm_dropout = Sequential([tf.keras.layers.Dense(n_hidden_1, activation='relu', input_shape=(n_input,)), tf.keras.layers.BatchNormalization(), tf.keras.layers.Dropout(0.1), tf.keras.layers.Dense(n_hidden_2, activation='relu'), tf.keras.layers.BatchNormalization(), tf.keras.layers.Dropout(0.1), tf.keras.layers.Dense(n_hidden_3, activation='relu'), tf.keras.layers.BatchNormalization(), tf.keras.layers.Dropout(0.1), tf.keras.layers.Dense(n_classes, activation='softmax')])\n",
        "\n",
        "# create the loss and optimiser, use 'categorical_crossentropy' in loss\n",
        "loss = tf.keras.losses.CategoricalCrossentropy()\n",
        "model_mlp_multi_layer_batchnorm_dropout.compile(loss=loss, optimizer=tf.keras.optimizers.Adam(), metrics=[\"accuracy\"])\n",
        "\n",
        "# Run optimisation algorithm\n",
        "n_epochs = 20\n",
        "batch_size = 64\n",
        "\n",
        "print('Training')\n",
        "model_mlp_multi_layer_batchnorm_dropout.fit(X_train, Y_train, epochs=n_epochs, batch_size=batch_size, verbose=2)\n",
        "\n",
        "print('Testing')\n",
        "model_mlp_multi_layer_batchnorm_dropout.evaluate(X_test, Y_test, verbose=2)\n",
        "model_mlp_multi_layer_batchnorm_dropout.predict(X_test)   \n",
        "\n",
        "\n",
        "test_mnist_images(model_mlp_multi_layer_batchnorm_dropout,X_test)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Y2CWgKZeRHA2"
      },
      "source": [
        "#### Documentation:\n",
        "\n",
        "  - http://www.tensorflow.org\n",
        "  - http://www.deeplearningbook.org/"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.9.15"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}