02_CNN_Introduction.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "02_CNN_Introduction.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "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/02_CNN_Introduction.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5uoxdNjkDw2K"
      },
      "source": [
        "# Introduction to Convolutional Neural Networks\n",
        "\n",
        "This tutorial addresses the basic concepts regarding Convolutional Neural Networks and their implementation using the Keras framework.\n",
        "Convolutional Neural Networks (CNNs) are a class of feed-forward artificial neural architecture. They are applied to analyse visual 2D imagery, meaning that we can feed images directly into a CNN without the need to flatten them into a 1D vector as done in the previous tutorial.\n",
        "CNNs have revolutionised the field of computer vision in the last decade. In 2012 Alex Krizhevsky introduced the AlexNet architecture to win the ImageNet Challenge (one of the most important competitions on image classification within the Computer Vision community), by reducing the top-5 error more than 10 percentage points, which was an incredible improvement at that time. As of now, CNNs are used not only on image classification but in many other computer vision tasks.\n",
        "\n",
        "![](https://cdn-5f733ed3c1ac190fbc56ef88.closte.com/wp-content/uploads/2017/03/alexnet_small-1.png)\n",
        "\n",
        "The image above is from [cv-tricks' blog](https://cv-tricks.com/cnn/understand-resnet-alexnet-vgg-inception/) and shows the proposed AlexNet architecture. It is composed of 5 convolutional layers followed by 3 fully connected layers. Nowadays, we can find much deeper and more complex architectures, which outperform AlexNet on the ImageNet Challenge. \n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dJ2d_oQ2UEPz"
      },
      "source": [
        "## CNN Structure\n",
        "\n",
        "The basic pipeline of common CNNs consists of an image as input and a stack of convolutional layers that extract a feature representation from the input image. The final shape of the image representation is conditioned on the type of problem/task that the architecture is facing. For instance, the output of the last layer in a classification problem is a probability vector. Each dimension of the probability vector represents how likely is that the input image belongs to a specific class. However, the architecture design is up to us, and therefore, we could code a network that outputs a single value for regression problems, or that generates a new image map for semantic segmentation. Now, let's dig in a bit into CNN and introduce some layers that are widely used. "
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wvNKve9GZL4M"
      },
      "source": [
        "### 2D Convolutional Layer\n",
        "\n",
        "The most common layer in any CNN architecture is the 2D convolutional layer. Convolutional layers are specifically designed to extract features from images or even extract features from previously extracted features. As shown in the following illustration, 2D convolutions apply the same filter to the full image. And therefore, due to its nature, 2D convolutional filters exploit the local information presented in images, making them a powerful tool for image analysis. \n",
        "\n",
        "![](https://cdn-images-1.medium.com/max/800/1*Fw-ehcNBR9byHtho-Rxbtw.gif)\n",
        "\n",
        "Image [source](https://towardsdatascience.com/intuitively-understanding-convolutions-for-deep-learning-1f6f42faee1).\n",
        "\n",
        "The latest deep learning frameworks have made possible the integration of convolutional layers easily on our architectures with only a single line of code. We will address here how 2D convolutions work since full understanding is needed to comprehend how any CNN operates. The following images and some explanations can be further explored on the original [Irhum Shafkat's blog](https://towardsdatascience.com/intuitively-understanding-convolutions-for-deep-learning-1f6f42faee1) or in the [Neural Networks and Deep Learning](http://neuralnetworksanddeeplearning.com/) book. Both are strongly recommended.\n",
        "\n",
        "To understand 2D convolutions, we need to define first what a kernel is. Kernels are simply matrices of numbers. The numbers on the kernels are the so-called weights, and the weights on the kernels change as we train the network. Hence, network training aims to optimise the network's parameters (update the kernel weights) such that the cost function is minimised. \n",
        "\n",
        "The 2D convolution operation takes the network's kernels and \"slides\" them over the input image (alike to a sliding window) as in the following image from [PyImageSearch](https://www.pyimagesearch.com/2015/03/23/sliding-windows-for-object-detection-with-python-and-opencv/) blog:\n",
        "\n",
        "![](https://pyimagesearch.com/wp-content/uploads/2014/10/sliding_window_example.gif)\n",
        "\n",
        "In each step, the network performs an element-wise multiplication with the elements that are currently on. The results of this elementwise multiplication are added to obtain the output value of the operation. CNNs repeat previous step for all the positions of the sliding window, composing at the end the feature map. This generated feature map can go through another 2D convolutional layer and create more powerful features. \n",
        "\n",
        "> \n",
        "\n",
        "![](https://cdn-images-1.medium.com/max/800/1*Zx-ZMLKab7VOCQTxdZ1OAw.gif)\n",
        "\n",
        "> \n",
        "The previous image shows the 2D convolution operation. Thus, the new feature values are the weighted sum of all the elements in the sliding window after the elementwise multiplication between input and kernel.  The bigger the size of the kernel is, the more feature elements contribute to the final output value. In contrast to fully connected layers, where a new feature value is a weighted sum over **all** input values, as mentioned, 2D convolutions compute features based on local areas. In other words, instead of looking at every input component, they consider only features coming from close locations. \n",
        "\n",
        "In the above example, the input image on the left has a size of 5x5 and the dimension of the resulting feature map is 3x3, showing that the size of the output maps is not always equal to the input. Hence, the output size can be computed by doing:\n",
        "\n",
        "$O = W - K + 1$,\n",
        "\n",
        "where $O$ is the output height/length, $W$ is the input height/length and $K$ is the kernel size. The output size is not only conditioned on the input size but also on the kernel size. Check in the following code cell how the output feature map shape changes as you increase the kernel size. In Keras, we define the layer by using `Conv2D` from `keras.layers` (documentation [here](https://keras.io/layers/convolutional/#conv2d)).\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ez90vQEksR9b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7d234cf8-89b3-47b4-d2af-5f24ce22cd00"
      },
      "source": [
        "import numpy as np\n",
        "import keras\n",
        "import tensorflow as tf\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Conv2D\n",
        "\n",
        "# Generate dummy data\n",
        "input_feature = np.random.random((1, 100, 100, 1))\n",
        "\n",
        "# input: 100x100 image with 1 channels -> (100, 100, 1) tensor.\n",
        "# this applies 1 convolution filter of size 3x3 each.\n",
        "model = Sequential()\n",
        "model.add(Conv2D(1, (3, 3), input_shape=(100, 100, 1)))\n",
        "\n",
        "output_feature = model.predict(input_feature)\n",
        "\n",
        "print('Input size: ({:}, {:})'.format(input_feature.shape[1], input_feature.shape[2]))\n",
        "print('Output size: ({:}, {:})'.format(output_feature.shape[1], output_feature.shape[2]))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Input size: (100, 100)\n",
            "Output size: (98, 98)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mNOhth4FuNob"
      },
      "source": [
        "Moreover, the kernel size and the input size are not the only parameters affecting the output size. We are going to introduce two extra elements that change the size of the output map: the padding and the stride."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mT6f_n5QiU8X"
      },
      "source": [
        "### Adding Padding to Input Features\n",
        "\n",
        "In some tasks, such as [image translation](https://arxiv.org/pdf/1611.07004.pdf), we need the output size to be equal to the input size. The solution to that is using padding, where extra edges are added to the input features so that the dimension is not reduced after the convolutional layer. Normally those pixels have $0$ value (termed zero-padding), but depending on the application other methods could be used, e.g., reflection or symmetric padding.\n",
        "\n",
        "![](https://cdn-images-1.medium.com/max/800/1*1okwhewf5KCtIPaFib4XaA.gif)\n",
        "\n",
        "If padding is used, the new output size can be computed by doing:\n",
        "\n",
        "$O = W - K + 2P+ 1$, \n",
        "\n",
        "where $P$ is the padding value. $P$ must be set in concordance with the kernel size if dimensionality wants to be preserved. Padding can be added to the `Conv2d` layer in Keras by using the padding argument and selecting one of the two settings: `valid`, which is the default value and means no padding; and `same`, which adapts the padding value to have the same output size as input size."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X7Z5G2O00Etd",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8becfaa7-c04c-47b9-a0d0-b21531197025"
      },
      "source": [
        "import numpy as np\n",
        "import keras\n",
        "import tensorflow as tf\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Conv2D\n",
        "\n",
        "# Generate dummy data\n",
        "input_feature = np.random.random((1, 100, 100, 1))\n",
        "\n",
        "# input: 100x100 image with 1 channels -> (100, 100, 1) tensor.\n",
        "# this applies 1 convolution filter of size 3x3 each.\n",
        "# attribute padding='same' applies zero-padding to the input feature map\n",
        "model = Sequential()\n",
        "model.add(Conv2D(1, (3, 3), input_shape=(100, 100, 1), padding='same'))\n",
        "\n",
        "output_feature = model.predict(input_feature)\n",
        "\n",
        "print('Input size: ({:}, {:})'.format(input_feature.shape[1], input_feature.shape[2]))\n",
        "print('Output size: ({:}, {:})'.format(output_feature.shape[1], output_feature.shape[2]))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Input size: (100, 100)\n",
            "Output size: (100, 100)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tdB4jMeswXVA"
      },
      "source": [
        "### Using Stride in Convolutional Layers\n",
        "\n",
        "The stride operation allows the convolutional layers to skip some of the sliding windows explained above. Hence, instead of jumping one pixel apart, we can define the number of skipped elements before computing the weighting sum between the kernel's weights and input features. A stride of 1 means that features will be extracted from all windows a pixel apart, so basically, every single window is computed. A stride of 2 means that we are selecting windows 2 pixels apart, skipping every other window in the process. Strides reduce the number of computations and consequently the size of the output map. In practice, as we go deeper into the CNN, the spatial size of the feature map gets smaller while the number of channels increases. Moreover, we can further reduce the size of the feature map using pooling operations, which we introduce later in this tutorial.\n",
        "\n",
        "![](https://cdn-images-1.medium.com/max/800/1*BMngs93_rm2_BpJFH2mS0Q.gif)\n",
        "\n",
        "If strides are used, the new output size can be computed as:\n",
        "\n",
        "$O = \\dfrac{W - K + 2P}{S}+ 1$, \n",
        "\n",
        "where $S$ is the stride value. The stride is set in the layer by using the `strides` argument."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Y8GjztiH0hnw",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5cc501ec-2071-4430-9760-ba9f99e7b6db"
      },
      "source": [
        "import numpy as np\n",
        "import keras\n",
        "import tensorflow as tf\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Conv2D\n",
        "\n",
        "# Generate dummy data\n",
        "input_feature = np.random.random((1, 100, 100, 1))\n",
        "\n",
        "# input: 100x100 image with 1 channels -> (100, 100, 1) tensor.\n",
        "# this applies 1 convolution filter of size 3x3 each.\n",
        "# attribute padding='same' applies zero-padding to the input feature map\n",
        "# attribute strides=2 applies applies stride of 2\n",
        "model = Sequential()\n",
        "model.add(Conv2D(1, (3, 3), input_shape=(100, 100, 1), padding='same', strides=2))\n",
        "\n",
        "output_feature = model.predict(input_feature)\n",
        "\n",
        "print('Input size: ({:}, {:})'.format(input_feature.shape[1], input_feature.shape[2]))\n",
        "print('Output size: ({:}, {:})'.format(output_feature.shape[1], output_feature.shape[2]))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Input size: (100, 100)\n",
            "Output size: (50, 50)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "f-fKvgJ3H2gr"
      },
      "source": [
        "## Differences Between Kernel and Filters\n",
        "\n",
        "The examples above take as input a single-channel image and compute a feature map with also one channel. However, when dealing with RGB images or feature maps, the input is no longer a single-channel map but, instead, they can have multiple channels. In the case of an RGB image, for each 2D convolution, we will need to define 3 kernels to interact with each of the image's channel colours. This group of kernels is called a filter. Thus, a filter is a collection of kernels that produces a single output.\n",
        "\n",
        "As a regular practice when defining Deep Learning models, we increase the number of filters in each convolutional layer as we go deeper into the model. Due to the element-wise multiplication, the number of kernels on each filter must be the same that the number of channels in the input feature map. Keras already deals with the number of kernels inside each filter. Keras keeps track of the input size in each convolutional layer unlike other frameworks (Pytorch or TensorFlow). Hence, in Keras, we must only decide the number of filters (output channels) in each layer.\n",
        "\n",
        "The next figure shows how the convolution is performed when having three input channels. First, one filter uses its three independent kernels to convolve with the RGB channels of the input image:\n",
        "\n",
        "![](https://cdn-images-1.medium.com/max/1000/1*8dx6nxpUh2JqvYWPadTwMQ.gif)\n",
        "\n",
        "Next, each of the processed feature maps is added together to obtain a single channel:\n",
        "\n",
        "![](https://cdn-images-1.medium.com/max/1000/1*CYB2dyR3EhFs1xNLK8ewiA.gif)\n",
        "\n",
        "Finally, we add the bias term to obtain the feature map. There is a single bias for the full output channel map. This operation is repeated for all the filters inside the convolutional layer.\n",
        "\n",
        "Now, we show how to use a `Conv2D` layer that takes an input image with 3 channels and generates an output map with 32 channels.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nqiEwpdROxga",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "05f02226-48b5-41fd-f1b0-c1e50db62341"
      },
      "source": [
        "import numpy as np\n",
        "import keras\n",
        "import tensorflow as tf\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Conv2D\n",
        "\n",
        "# Generate dummy data\n",
        "input_feature = np.random.random((1, 100, 100, 3))\n",
        "\n",
        "# input: 100x100 image with 3 channels -> (100, 100, 3) tensor.\n",
        "# this applies 32 convolution filters of size 3x3 each.\n",
        "# attribute padding='same' applies zero-padding to the input feature map\n",
        "# attribute strides=2 applies applies stride of 2\n",
        "model = Sequential()\n",
        "model.add(Conv2D(32, (3, 3), input_shape=(100, 100, 3), padding='same', strides=2))\n",
        "\n",
        "output_feature = model.predict(input_feature)\n",
        "\n",
        "print('Input size: ({:}, {:}, {:})'.format(input_feature.shape[1], input_feature.shape[2], input_feature.shape[3]))\n",
        "print('Output size: ({:}, {:}, {:})'.format(output_feature.shape[1], output_feature.shape[2], output_feature.shape[3]))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Input size: (100, 100, 3)\n",
            "Output size: (50, 50, 32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "S50cYwkhPqka"
      },
      "source": [
        "## Activation Functions\n",
        "\n",
        "As seen in previous tutorials, after a `Dense` layer we usually can find an activation function. We introduce here how to use them after `Conv2D` layers. Those activation functions are a set of operators that maps the feature values to a new set of values, the mapping depends on the function at hand. The main reason for using activation functions is that they add non-linearities to the network, giving more expressive power to the network, which will be able to reproduce more complex functions. \n",
        "\n",
        "\n",
        "The complete list of activation functions that Keras offers can be found [here](https://keras.io/activations/). We introduce here some of them:\n",
        "\n",
        "*  **Sigmoid Function** sets the output in the range (0, 1). The sigmoid function is widely used in binary classification problems since its output can be taken as a probability value. `keras.activations.sigmoid(x)`:\n",
        "\n",
        ">![](https://i.ibb.co/Ph8dsTv/sigmoid.png)\n",
        "\n",
        "*  **Tanh Function** is a logistic function as sigmoid, but the range of the tanh function is (-1, 1). Contrary to sigmoid function, where the values close to 0 are set around 0.5, in the tanh function they will be still mapped around the 0 value. `keras.activations.tanh(x)`:\n",
        "\n",
        ">![](https://i.ibb.co/68g7LpL/tanh.png)\n",
        "\n",
        "*  **ReLU Function** is the most common activation function you can find in any current CNN as in general works better than the rest. The range of this function is in \\[0, inf). It sets all negative values to 0 and hence is computationally easy to implement. As a drawback, during training some neurons *die*, meaning that the output is 0 for all available data points and no gradient is propagated there. `keras.activations.relu(x, alpha=0.0, max_value=None, threshold=0.0)`:\n",
        "\n",
        ">![](https://i.ibb.co/Zd9H8Z4/relu.png)\n",
        "\n",
        "*  **LeakyReLU Function** is a modified version of the ReLU activation above, which attempts to solve the problem of dying neurons that ReLU has. While ReLU does not backpropagate negative values, Leaky ReLU smooths those values without setting them to 0. That allows the gradients to backpropagate through the network even for negative values. `keras.layers.LeakyReLU(alpha=0.3)`:\n",
        "\n",
        ">![](https://i.ibb.co/dmnJ6h1/leakyrelu.png)\n",
        "\n",
        "*  **Softmax Function** is another widely activation function for multi-class classification problems and usually is employed as the last activation function in the classification model. This function sets all of the output elements to the range (0, 1). However, the softmax function does not take independently the input values to map it into its probability value. Softmax Function takes an un-normalized vector, $s$, and normalizes it into a probability distribution, $p$, following the softmax expression. As the output is a probability, the output elements add up to 1. `keras.activations.softmax(x, axis=-1)`. Thus, the output value $p_i$ is computed as:\n",
        "\n",
        "> $p_{i} = \\dfrac{e^{s_i}}{\\sum_{\\substack{j}}^{N} e^{s_j}}$"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-MWo2VsYr5T2"
      },
      "source": [
        "The following example shows the feature maps before and after of the ReLU activation function. All values that are negative are set to 0 after the activation function.\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lp1HH0VOqN0N",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1884817f-59b6-41f8-8889-474d9d865180"
      },
      "source": [
        "import numpy as np\n",
        "import keras\n",
        "import tensorflow as tf\n",
        "tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)\n",
        "from keras.models import Sequential, Model\n",
        "from keras.layers import Conv2D, Activation\n",
        "\n",
        "# Generate dummy data\n",
        "input_feature = np.random.random((1, 3, 3, 1)) - 0.5\n",
        "\n",
        "# input: 3x3 image with 1 channel -> (3, 3, 1) tensor.\n",
        "# This applies a 1 convolution filter of size 3x3 each.\n",
        "# This applies a ReLU activation function\n",
        "model = Sequential()\n",
        "model.add(Conv2D(1, (3, 3), input_shape=(3, 3, 1), padding='same', strides=1, name='conv'))\n",
        "model.add(Activation('relu'))\n",
        "\n",
        "model_before_ReLU = Model(inputs=model.input, outputs=model.get_layer('conv').output)\n",
        "\n",
        "output_feature = model_before_ReLU.predict(input_feature)\n",
        "output_ReLu_feature = model.predict(input_feature)\n",
        "\n",
        "print('Output Network without activation function')\n",
        "print(output_feature)\n",
        "\n",
        "print('')\n",
        "print('Output Network after ReLU activation function')\n",
        "print(output_ReLu_feature)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Output Network without activation function\n",
            "[[[[ 0.23390736]\n",
            "   [ 0.2609628 ]\n",
            "   [-0.10049579]]\n",
            "\n",
            "  [[ 0.47150436]\n",
            "   [ 0.01826818]\n",
            "   [-0.0576808 ]]\n",
            "\n",
            "  [[ 0.08542764]\n",
            "   [ 0.02355585]\n",
            "   [ 0.09083541]]]]\n",
            "\n",
            "Output Network after ReLU activation function\n",
            "[[[[0.23390736]\n",
            "   [0.2609628 ]\n",
            "   [0.        ]]\n",
            "\n",
            "  [[0.47150436]\n",
            "   [0.01826818]\n",
            "   [0.        ]]\n",
            "\n",
            "  [[0.08542764]\n",
            "   [0.02355585]\n",
            "   [0.09083541]]]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ys6ZB1DOTLH6"
      },
      "source": [
        "## Pooling Layer\n",
        "\n",
        "It is a common practice to insert a pooling layer between convolutional layers in CNNs. In a standard CNN architecture, we set the feature sizes to become smaller progressively to reduce the computation in the networks, and to merge the information from different spatial locations. To reduce the feature map sizes, we can either use bigger stride size in the convolutional layers or we can use pooling layers. Pooling layers perform a spatial sliding window and apply an operation to reduce the spatial size. Those operations vary depending on the architecture, being the max, mean and min pooling the most typical ones. Here, we will explain the max pooling, although all the others work similarly. Max pooling keeps only the max value in a neighbourhood, where the neighbourhood is defined by the size of the kernel. Let's visualise it, the next example shows the result of a Max Pooling layer with a 2x2 kernel and a stride of 2.\n",
        "\n",
        "![](https://i.ibb.co/Xp454S4/MaxPool.png)\n",
        "\n",
        "As in convolutional layers, the final size is conditioned to the stride size of the pooling layer. However, contrary to convolutional layers, pooling layers operate independently on each of the input channels, without modifying the depth of the feature maps. To add max pooling to our model we need to import `MaxPooling2d` from `keras.layers` and define the stride and pooling size."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HzoTGOwDmQm4",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6a6bfeb3-3c9e-4947-a090-df10a5c5e662"
      },
      "source": [
        "import numpy as np\n",
        "import keras\n",
        "import tensorflow as tf\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Conv2D, Activation, MaxPooling2D\n",
        "\n",
        "# Generate dummy data\n",
        "input_feature = np.random.random((1, 100, 100, 3))\n",
        "\n",
        "# input: 100x100 image with 3 channels -> (100, 100, 3) tensor.\n",
        "# this applies 32 convolution filters of size 3x3 each.\n",
        "# attribute padding='same' applies zero-padding to the input feature map\n",
        "# attribute strides=1 applies applies stride of 1\n",
        "model = Sequential()\n",
        "model.add(Conv2D(32, (3, 3), input_shape=(100, 100, 3), padding='same', strides=1))\n",
        "model.add(Activation('relu'))\n",
        "model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same'))\n",
        "\n",
        "output_feature = model.predict(input_feature)\n",
        "\n",
        "print('Input size: ({:}, {:}, {:})'.format(input_feature.shape[1], input_feature.shape[2], input_feature.shape[3]))\n",
        "print('Output size: ({:}, {:}, {:})'.format(output_feature.shape[1], output_feature.shape[2], output_feature.shape[3]))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Input size: (100, 100, 3)\n",
            "Output size: (50, 50, 32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LKU4-GbZHw0R"
      },
      "source": [
        "# Example: Classification on MNIST\n",
        "\n",
        "As opposed to the *Introduction to Keras* tutorial, in this section, we show how to perform image classification when the input data is a 2D image instead of a flat 1D vector. \n",
        "\n",
        "As discussed above, Convolutional Neural Networks aim to extract and exploit the local relationships on 2D maps, hence, CNNs are much more convenient for images than Multi-layer Perceptron models. \n",
        "\n",
        "First of all, we load the MNIST dataset from Keras' framework. The definition of the data is almost identical than in the Keras tutorial, although this time we are not reshaping the input images into a 1D vector."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3t0Wq9ArL4EZ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "62154a13-9ea8-4404-e8a1-96ea2fe16abe"
      },
      "source": [
        "import tensorflow as tf\n",
        "tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Dense, Dropout, Activation, BatchNormalization\n",
        "from keras.utils import np_utils\n",
        "from keras.datasets import mnist\n",
        "from keras.optimizers import RMSprop\n",
        "from keras.layers import Conv2D, MaxPooling2D, Flatten\n",
        "import numpy as np\n",
        "\n",
        "# the data, shuffled and split between train and test sets\n",
        "# Here we are using the official test set as our validation set, in further\n",
        "# tutorials, test and validation splits will be explained properly.\n",
        "# Hence, even though the variables are `x_test` and `y_test`, they represent our validation set\n",
        "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
        "\n",
        "shape = x_train.shape\n",
        "\n",
        "# Normalize and reshape the input images\n",
        "x_train = np.expand_dims(x_train.astype('float32'), -1)\n",
        "x_test = np.expand_dims(x_test.astype('float32'), -1)\n",
        "\n",
        "x_train /= 255\n",
        "x_test /= 255\n",
        "\n",
        "y_train_class = np_utils.to_categorical(y_train, 10)\n",
        "y_test_class = np_utils.to_categorical(y_test, 10)\n",
        "\n",
        "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 validation samples: {0}'.format(x_test.shape[0]))"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz\n",
            "11493376/11490434 [==============================] - 0s 0us/step\n",
            "Image shape: (28, 28, 1)\n",
            "Total number of training samples: 60000\n",
            "Total number of validation samples: 10000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4Wp29GszMO-j"
      },
      "source": [
        "As illustrated in the cell, this time the input image before the model is 28x28x1. \n",
        "\n",
        "Now we can define a model composed of convolutional layers, activation functions, and maxpool operators:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "B78NUAqwMeFR"
      },
      "source": [
        "model = Sequential()\n",
        "model.add(Conv2D(16, (3,3), padding='same', input_shape=x_train.shape[1:]))\n",
        "model.add(Activation('relu'))\n",
        "model.add(MaxPooling2D(pool_size=(2,2)))\n",
        "model.add(Conv2D(16, (3,3), padding='same'))\n",
        "model.add(Activation('relu'))\n",
        "model.add(MaxPooling2D(pool_size=(2,2)))"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BX_KAFlBMnVZ"
      },
      "source": [
        "As explained in the previous tutorial, in a classification problem, the output of the model is a probability vector. Each dimension of the vector indicates how likely is that the input image belongs to a specific class. \n",
        "\n",
        "Up to now, the resulting feature map of the model is a map with the shape *Batch x Weight' x Height' x Channel*, and it needs to be mapped into a vector with shape *Batch x Num Classes*. A common technique to process this mapping is to add a Flatten layer that will reshape the feature map to *Batch x (Weight' * Height' * Channel)*. Following the Flatten layer, we add a dense layer which maps this new feature map to the desired output size. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3CAc1r2vOJQv"
      },
      "source": [
        "model.add(Flatten())\n",
        "model.add(Dense(10))\n",
        "model.add(Activation('softmax'))"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aHInEJ-aON5v"
      },
      "source": [
        "Let's visualize the output shape in each layer."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UD4NZtIxOTbY",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "92c15390-188f-4d7e-c498-cfe4cd56e4e2"
      },
      "source": [
        "model.summary()"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d (Conv2D)              (None, 28, 28, 16)        160       \n",
            "_________________________________________________________________\n",
            "activation (Activation)      (None, 28, 28, 16)        0         \n",
            "_________________________________________________________________\n",
            "max_pooling2d (MaxPooling2D) (None, 14, 14, 16)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_1 (Conv2D)            (None, 14, 14, 16)        2320      \n",
            "_________________________________________________________________\n",
            "activation_1 (Activation)    (None, 14, 14, 16)        0         \n",
            "_________________________________________________________________\n",
            "max_pooling2d_1 (MaxPooling2 (None, 7, 7, 16)          0         \n",
            "_________________________________________________________________\n",
            "flatten (Flatten)            (None, 784)               0         \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 10)                7850      \n",
            "_________________________________________________________________\n",
            "activation_2 (Activation)    (None, 10)                0         \n",
            "=================================================================\n",
            "Total params: 10,330\n",
            "Trainable params: 10,330\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_F66ds58OTBu"
      },
      "source": [
        "Finally, we can train our CNN and check its performance on MNIST. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BmeolsG_Pi8A",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ce992818-f5c4-49a2-f27b-1500b222e839"
      },
      "source": [
        "# initiate RMSprop optimizer\n",
        "opt = RMSprop(lr=0.0001, decay=1e-6)\n",
        "\n",
        "model.compile(loss='categorical_crossentropy',\n",
        "              optimizer=opt,\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "model.fit(x_train, y_train_class, batch_size=32, epochs=10,  verbose=0)\n",
        "\n",
        "score = model.evaluate(x_test, y_test_class, verbose=0)\n",
        "print('Validation loss:', score[0])\n",
        "print('Validation accuracy:', score[1])"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Validation loss: 0.08394194394350052\n",
            "Validation accuracy: 0.9743000268936157\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Q0ZorPBtpLlZ"
      },
      "source": [
        "# Coursework\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2gkM3dPZbNcX"
      },
      "source": [
        "## Task 1: Classification\n",
        "\n",
        "At this point, we know what is a CNN, how they work, and the components needed to design them. In this first task, we want you to create a CNN that is able to outperform the Multi-layer Perceptron model from Tutorial 1. For the first part of the coursework, we train on CIFAR10, a  classical dataset for image classification. Note that in these tutorials, we mainly use the official test sets of several standard datasets as our validation data. The reason we use the given test sets as validation data for the tutorials is that is an easy way to make sure that we all work with the same split and report results using the same data. However, in a proper machine learning setup, your validation set should be separate from the test set, so you can tune the model/parameters on the validation set and then check the final performance in the test set. Thus, even though the variables are `x_test` and `y_test`, they represent our validation set.\n",
        "\n",
        "Let's first load the dataset and visualise some examples:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zKWvv0VwZPH3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 322
        },
        "outputId": "8771ffa0-3dd3-45c3-8766-06c729c92028"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Dense, Dropout, Activation\n",
        "from keras.utils import np_utils\n",
        "from keras.datasets import cifar10\n",
        "from keras.optimizers import RMSprop\n",
        "\n",
        "# the data, shuffled and split between train and test sets\n",
        "# Here we are using the official test set as our validation set, in further\n",
        "# tutorials, test and validation splits will be explained properly.\n",
        "# Hence, even though the variables are `x_test` and `y_test`, they represent our validation set\n",
        "(X_train, y_train), (X_test, y_test) = cifar10.load_data()\n",
        "\n",
        "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 validation samples: {0}'.format(X_test.shape[0]))\n",
        "\n",
        "# Let's visualize some examples\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",
          "text": [
            "Image shape: (32, 32, 3)\n",
            "Total number of training samples: 50000\n",
            "Total number of validation samples: 10000\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 25 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-p9uVP3pccOW"
      },
      "source": [
        "CIFAR10 contains 50,000 32x32 color training images, labeled over 10 categories and 10,000 test images. As mentioned, we use those test images as our validation set in this tutorial. We need to preprocess the images for the 1D neural network defined in the previous tutorial:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bFurVCqpclnm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5226fee6-23bc-43fd-c722-617ff31feec6"
      },
      "source": [
        "# Preprocess dataset for a 1D neural network\n",
        "X_train = X_train.astype('float32')\n",
        "X_test = X_test.astype('float32')\n",
        "\n",
        "# Normalize the image\n",
        "X_train /= 255\n",
        "X_test /= 255\n",
        "\n",
        "X_train_flatten = X_train.reshape(X_train.shape[0], X_train.shape[1] * X_train.shape[2]*3)\n",
        "X_test_flatten = X_test.reshape(X_test.shape[0], X_test.shape[1] * X_test.shape[2]*3)\n",
        "\n",
        "print('New X_train shape: {0}'.format(X_train_flatten.shape))\n",
        "\n",
        "Y_train_class = np_utils.to_categorical(y_train, 10)\n",
        "Y_test_class = np_utils.to_categorical(y_test, 10)\n",
        "\n",
        "print('New Y_train shape: {0}'.format(Y_train_class.shape))\n"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "New X_train shape: (50000, 3072)\n",
            "New Y_train shape: (50000, 10)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mxFx2s3Fc6wM"
      },
      "source": [
        "Now, we are ready to define the Multi-layer Perceptron model and train it."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KEZW-1pxbKlC",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c146e6d1-d6b0-4ece-def7-c7348c30551f"
      },
      "source": [
        "model = Sequential()\n",
        "\n",
        "model.add(Dense(32, activation='relu', input_shape=(X_train_flatten.shape[1:])))\n",
        "model.add(Dense(64, activation='relu'))\n",
        "model.add(Dense(128, activation='relu'))\n",
        "model.add(Dense(256, activation='relu'))\n",
        "model.add(Dense(512, activation='relu'))\n",
        "model.add(Dense(10, activation='softmax'))\n",
        "\n",
        "# initiate RMSprop optimizer\n",
        "opt = RMSprop(lr=0.0001, decay=1e-6)\n",
        "\n",
        "# Let's train the model using RMSprop\n",
        "model.compile(loss='categorical_crossentropy',\n",
        "              optimizer=opt,\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "model.fit(X_train_flatten, Y_train_class, batch_size=32, epochs=20)\n",
        "\n",
        "score = model.evaluate(X_test_flatten, Y_test_class, verbose=0)\n",
        "print('Validation loss:', score[0])\n",
        "print('Validation accuracy:', score[1])"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 2.0479 - accuracy: 0.2362\n",
            "Epoch 2/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.7793 - accuracy: 0.3563\n",
            "Epoch 3/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.6936 - accuracy: 0.3909\n",
            "Epoch 4/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.6421 - accuracy: 0.4085\n",
            "Epoch 5/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.5967 - accuracy: 0.4245\n",
            "Epoch 6/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.5629 - accuracy: 0.4399\n",
            "Epoch 7/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.5393 - accuracy: 0.4471\n",
            "Epoch 8/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.5119 - accuracy: 0.4595\n",
            "Epoch 9/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.4797 - accuracy: 0.4699\n",
            "Epoch 10/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.4686 - accuracy: 0.4746\n",
            "Epoch 11/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.4503 - accuracy: 0.4822\n",
            "Epoch 12/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.4237 - accuracy: 0.4901\n",
            "Epoch 13/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.4241 - accuracy: 0.4942\n",
            "Epoch 14/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.4023 - accuracy: 0.4950\n",
            "Epoch 15/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.3772 - accuracy: 0.5062\n",
            "Epoch 16/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.3757 - accuracy: 0.5073\n",
            "Epoch 17/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.3564 - accuracy: 0.5168\n",
            "Epoch 18/20\n",
            "1563/1563 [==============================] - 5s 3ms/step - loss: 1.3609 - accuracy: 0.5118\n",
            "Epoch 19/20\n",
            "1563/1563 [==============================] - 5s 3ms/step - loss: 1.3362 - accuracy: 0.5215\n",
            "Epoch 20/20\n",
            "1563/1563 [==============================] - 6s 4ms/step - loss: 1.3162 - accuracy: 0.5254\n",
            "Validation loss: 1.4244304895401\n",
            "Validation accuracy: 0.4925999939441681\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CEuYNmyHlVJd"
      },
      "source": [
        "### Problem Definition\n",
        "\n",
        "In this exercise, you are asked to test several CNN architectures in the code provided below. Do not modify the optimizer, loss used or parameters related to the training such as the learning rate, they will be investigated in future tutorials. You must focus on the architecture itself: number of convolutional layers, number of filters in every layer, activation functions, pooling operators, among others. Batch Normalization and Dropout layers, which are quite used in CNN architectures, will be also investigated in a future tutorial so you do not have to discuss them.\n",
        "\n",
        "\n",
        "**Report**:\n",
        "*   Present a bar figure with the training and validation accuracies for different design choices. Discuss only the parameters that have a significant influence on the network's performance. Explain any discrepancy between training and validation accuracies.\n",
        "*   Present a sketch that introduces your best architecture. See some examples on how to display networks in [cv-tricks' blog](https://cv-tricks.com/cnn/understand-resnet-alexnet-vgg-inception/).\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r8ATD38dDM1q"
      },
      "source": [
        "import tensorflow as tf\n",
        "tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Dense, Activation\n",
        "from keras.utils import np_utils\n",
        "from keras.datasets import cifar10\n",
        "from keras.optimizers import RMSprop\n",
        "from keras.layers import Conv2D, MaxPooling2D, Flatten\n",
        "\n",
        "# the data, shuffled and split between train and val sets\n",
        "# Here we are using the official test set as our validation set, in further\n",
        "# tutorials, test and validation splits will be explained properly.\n",
        "# Hence, even though the variables are `x_test` and `y_test`, they represent our validation set\n",
        "(x_train, y_train), (x_test, y_test) = cifar10.load_data()\n",
        "\n",
        "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 validation samples: {0}'.format(x_test.shape[0]))\n",
        "\n",
        "x_train = x_train.astype('float32')\n",
        "x_test = x_test.astype('float32')\n",
        "\n",
        "# Normalize the image\n",
        "x_train /= 255\n",
        "x_test /= 255\n",
        "\n",
        "y_train_class = np_utils.to_categorical(y_train, 10)\n",
        "y_test_class = np_utils.to_categorical(y_test, 10)\n",
        "\n",
        "\n",
        "# . . . \n",
        "\n",
        "\n",
        "# Define here your architecture\n",
        "\n",
        "\n",
        "# . . . \n",
        "\n",
        "\n",
        "# initiate RMSprop optimizer\n",
        "opt = RMSprop(lr=0.0001, decay=1e-6)\n",
        "\n",
        "model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy'])\n",
        "\n",
        "model.fit(x_train, y_train_class, batch_size=32, epochs=20)\n",
        "\n",
        "score = model.evaluate(x_test, y_test_class, verbose=0)\n",
        "print('Validation loss:', score[0])\n",
        "print('Validation accuracy:', score[1])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3w8eUV9RDPkL"
      },
      "source": [
        "\n",
        "---\n",
        "---\n",
        "\n",
        "## Task 2: Regression\n",
        "\n",
        "Now we face a regression task instead of a classification problem. Loss function, activations, and dataset will change in the following task. Thus, instead of having one vector with the probabilities of each class, in this regression problem, the output is a single scalar. \n",
        "\n",
        "For this second task, we chose the task of estimating house prices based on input images. To get the data run the following script, which clones Ahmed and Moustafa’s [repository](https://github.com/emanhamed/Houses-dataset) into colmap."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ibukle2ODMEp"
      },
      "source": [
        "!git clone https://github.com/emanhamed/Houses-dataset\n",
        "%cd /content/Houses-dataset/Houses\\ Dataset"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ye6xldkVEV-t"
      },
      "source": [
        "This dataset contains four images of the house (kitchen, frontal, bedroom and bathroom), and attributes (number of bedrooms, number of bathrooms, zip code...). For our exercise, we only use the images of the house. We start with front door images. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "71urud9ZHq9d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 286
        },
        "outputId": "9c7c3b61-b346-4eda-fb75-d7b8d7cff831"
      },
      "source": [
        "import cv2\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from numpy.random import seed\n",
        "\n",
        "house_section = 'frontal' # select between: kitchen, frontal, bedroom or bathroom\n",
        "print('We have selected {} images. You can switch to the kitchen, frontal, bedroom or bathroom images by changing house_section variable.'.format(house_section))\n",
        "images = []\n",
        "for i_im in range(1, 536):\n",
        "  image = cv2.imread(str(i_im)+'_'+house_section+'.jpg')\n",
        "  image = cv2.resize(image, (64, 64))\n",
        "  images.append(image)\n",
        "\n",
        "labels = []\n",
        "f = open('HousesInfo.txt', \"r\")\n",
        "for x in f:\n",
        "  label = (x).split(' ')[-1].split('\\n')[0]\n",
        "  labels.append(label)\n",
        "\n",
        "# Let's visualize some examples\n",
        "N=3\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",
        "for row in range(N):\n",
        "  for col in range(N):\n",
        "    idx = start_val+row+N*col    \n",
        "    tmp = cv2.cvtColor(images[idx],cv2.COLOR_BGR2RGB)\n",
        "    axes[row,col].imshow(tmp, cmap='gray')\n",
        "    fig.subplots_adjust(hspace=0.5)\n",
        "    target = int(labels[idx])\n",
        "    axes[row,col].set_title(str(target) + '$')\n",
        "    axes[row,col].set_xticks([])\n",
        "    axes[row,col].set_yticks([])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "We have selected frontal images. You can switch to the kitchen, frontal, bedroom or bathroom images by changing house_section variable.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 9 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "scNfczb73nGS"
      },
      "source": [
        "Prepare the dataset for training the model:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FORNJ1QU2L-t",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "833c61ef-51f2-407c-c03c-bdb26cee6a47"
      },
      "source": [
        "seed(1234)\n",
        "images = np.asarray(images).astype('float32')\n",
        "labels = np.asarray(labels).astype('float32')\n",
        "\n",
        "# Normalize the image\n",
        "max_price = labels.max()\n",
        "images /= 255.\n",
        "labels /= max_price\n",
        "\n",
        "random_idx = np.random.permutation(len(images))\n",
        "images = images[random_idx]\n",
        "labels = labels[random_idx]\n",
        "\n",
        "split_size_val = int(0.8*len(images))\n",
        "X_train, X_val = images[:split_size_val], images[split_size_val+1:]\n",
        "Y_train, Y_val = labels[:split_size_val], labels[split_size_val+1:]\n",
        "\n",
        "# Print shape of training and val images \n",
        "print('X_train image shape: {0}'.format(X_train.shape))\n",
        "print('X_val image shape: {0}'.format(X_val.shape))\n",
        "\n",
        "# Print shape of training and val labels \n",
        "print('Y_train labels shape: {0}'.format(Y_train.shape))\n",
        "print('Y_val labels shape: {0}'.format(Y_val.shape))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "X_train image shape: (428, 64, 64, 3)\n",
            "X_val image shape: (106, 64, 64, 3)\n",
            "Y_train labels shape: (428,)\n",
            "Y_val labels shape: (106,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JGC6dXqdM2eU"
      },
      "source": [
        "### Problem Definition\n",
        "\n",
        "Similar to the previous task, you are asked to design a CNN architecture able to perform the estimation of house prices based on the `frontal` house image. Design a new model by changing parameters such as the number of convolutional layers, activation functions, strides, or pooling operators, among others."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YE8g42x3M2Ko"
      },
      "source": [
        "import tensorflow as tf\n",
        "tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Dense, Activation\n",
        "from keras.utils import np_utils\n",
        "from keras.optimizers import Adam\n",
        "from keras.layers import Conv2D, MaxPooling2D, Flatten\n",
        "tf.random.set_seed(1234)\n",
        "\n",
        "chanDim = -1\n",
        "model = Sequential()\n",
        "\n",
        "# . . . \n",
        "\n",
        "\n",
        "# Define here your architecture\n",
        "\n",
        "\n",
        "# . . . \n",
        "\n",
        "model.summary()\n",
        "opt = Adam(lr=1e-3, decay=1e-3 / 200)\n",
        "model.compile(loss=\"mean_absolute_percentage_error\", optimizer=opt, metrics=['mean_absolute_percentage_error'])\n",
        "\n",
        "# train the model\n",
        "print(\"[INFO] training model...\")\n",
        "model.fit(X_train, Y_train, validation_data=(X_val, Y_val), epochs=100, batch_size=32)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JyXrXBFVQ-tL"
      },
      "source": [
        "The metric used in this problem to evaluate the performance is the same we used for training the model, the mean absolute percentage error. Mean absolute percentage error is defined as $\\frac{100}{n} \\sum_n \\frac{|\\hat{y} - y|}{|y|}$ where $y$ is the ground-truth, $\\hat{y}$ is the estimation of the model and `n` the number of elements in the set we are evaluating."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u2u2qQh90C3K"
      },
      "source": [
        "prices_predicted = model.predict(X_val).flatten()\n",
        "diff = ((prices_predicted - Y_val) / Y_val) * 100\n",
        "error_mean = np.mean(np.abs(diff))\n",
        "\n",
        "print(\"Predicting house prices - Estimation Error: {:.2f}%\".format(error_mean))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "n2uJs6-xG6lD"
      },
      "source": [
        "**Report**:\n",
        "\n",
        "\n",
        "*   Propose a CNN architecture that has an estimation error in the validation set below 75%. \n",
        "*   Present a figure showing the training and validation loss vs the number of training epochs for different architectural design choices. Discuss the gap between the training and validation loss depending on the proposed architecture.\n",
        "*   Report a table with results when using any of the other images from the house (kitchen, bedroom, and bathroom)."
      ]
    }
  ]
}