Lab_6_(part_2)_LearningRate_Regularization_HyperparameterSearch.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.7.3"
    },
    "colab": {
      "name": "Lab_6_(part_2)_LearningRate_Regularization_HyperparameterSearch.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9Mihd9hpHKVh"
      },
      "source": [
        "#Lab 6: Training neural networks (part 2/2)\n",
        "**Like always, remember to set Runtime environment to GPU**\n",
        "\n",
        "In this second part of lab 6 we will explore more tools that will help you train your own neural networks. This time we will be using convolutional neural networks (CNNs). \n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0tGBcfgIumcH"
      },
      "source": [
        "##1. Download the CIFAR 10 dataset\n",
        "We will be using the CIFAR 10 dataset."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u_awI9jy3V9P",
        "outputId": "da8b4db2-e57f-46be-e189-5b3c90fc7002",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 67
        }
      },
      "source": [
        "import torch\n",
        "import torchvision\n",
        "import torchvision.transforms as transforms\n",
        "\n",
        "mean = 0.5\n",
        "std = 0.5\n",
        "transform = transforms.Compose(\n",
        "    [transforms.ToTensor(),\n",
        "     transforms.Normalize((mean, mean, mean), (std, std, std))])\n",
        "\n",
        "# Batch size\n",
        "bs = 64\n",
        "\n",
        "trainset = torchvision.datasets.CIFAR10(root='./data', train=True,\n",
        "                                        download=True, transform=transform)\n",
        "trainloader = torch.utils.data.DataLoader(trainset, batch_size=bs,\n",
        "                                          shuffle=True, num_workers=2,drop_last=True)\n",
        "\n",
        "testset = torchvision.datasets.CIFAR10(root='./data', train=False,\n",
        "                                       download=True, transform=transform)\n",
        "testloader = torch.utils.data.DataLoader(testset, batch_size=bs,\n",
        "                                         shuffle=False, num_workers=2,drop_last=True)\n",
        "\n",
        "classes = ('plane', 'car', 'bird', 'cat',\n",
        "           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\r0it [00:00, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to ./data/cifar-10-python.tar.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            " 99%|█████████▊| 168173568/170498071 [00:12<00:00, 16844626.55it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Extracting ./data/cifar-10-python.tar.gz to ./data\n",
            "Files already downloaded and verified\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iorLYSEz8Xbs"
      },
      "source": [
        "Display some stats"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "U50pDb-_5uKq",
        "outputId": "6bd16257-ba8f-451d-8639-7f2b78ffb5f3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        }
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "# Function to show an image\n",
        "def imshow(img):\n",
        "    npimg = img.numpy() / 2 + 0.5 # Un-normalize\n",
        "    plt.imshow(np.transpose(npimg, (1, 2, 0)))\n",
        "    plt.show()\n",
        "\n",
        "# Number of samples\n",
        "num_train = bs * len(trainloader)\n",
        "num_test = bs * len(testloader)\n",
        "print('num_train',num_train)\n",
        "print('num_test',num_test)\n",
        "\n",
        "# Get a batch of some random training images\n",
        "dataiter = iter(trainloader)\n",
        "images, labels = dataiter.next()\n",
        "print('images.shape',images.shape)\n",
        "\n",
        "# show 16 images and print labels\n",
        "imshow(torchvision.utils.make_grid(images[0:16]))\n",
        "print(' '.join('%5s' % classes[labels[j]] for j in range(16)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "num_train 49984\n",
            "num_test 9984\n",
            "images.shape torch.Size([64, 3, 32, 32])\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXAAAAB4CAYAAADrPanmAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "  cat horse truck horse   cat horse   dog  bird  bird   car   car   car   car  deer  ship  deer\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "p410rTQvawvk"
      },
      "source": [
        "##2. Task 1 (optional): Setting up your CNN\n",
        "**You can skip this task and jump directly to Task 3, if you feel comfortable with PyTorch and the different layer types used in CNNs.**\n",
        "\n",
        "Here, we will set up our own CNN. For this purpose we will need the following layer types:\n",
        "\n",
        "- [Conv2d](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d)\n",
        "- [MaxPool2d](https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html#torch.nn.MaxPool2d)\n",
        "- [BatchNorm2d](https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm2d.html#torch.nn.BatchNorm2d)\n",
        "- [ReLU](https://pytorch.org/docs/stable/generated/torch.nn.ReLU.html#torch.nn.ReLU)\n",
        "- [Dropout](https://pytorch.org/docs/stable/generated/torch.nn.Dropout.html#torch.nn.Dropout)\n",
        "- [Linear](https://pytorch.org/docs/stable/generated/torch.nn.Linear.html#torch.nn.Linear])\n",
        "\n",
        "as well as the reshape operation (called `view` in PyTorch):\n",
        "\n",
        "- [view](https://pytorch.org/docs/stable/tensors.html#torch.Tensor.view)\n",
        "\n",
        "See complete overview of the basic building block for graphs [here](https://pytorch.org/docs/stable/nn.html).\n",
        "\n",
        "Let's see how each of these work:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PiLeU_-tYy6g"
      },
      "source": [
        "# First import what we need\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RRTKA44Wg-Pq"
      },
      "source": [
        "###2.1 Conv2d\n",
        "Reference to documentation: https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3hJym68ndltc"
      },
      "source": [
        "# Print shape of input\n",
        "print('Input dimensions are (batch_size, channels, height, width)')\n",
        "print('images.shape',images.shape)\n",
        "\n",
        "# Perform convolution\n",
        "conv = nn.Conv2d(3, 6, 5)\n",
        "x = conv(images)\n",
        "\n",
        "# Print shape of output\n",
        "print('x.shape',x.shape)\n",
        "\n",
        "# Print parameter shapes\n",
        "for name, param in conv.named_parameters(): print('parameter',name,param.shape)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HfUehTJBdvi9"
      },
      "source": [
        "####2.1.1 Questions\n",
        "1. What do each of the parameters (3, 6, 5) mean?\n",
        "2. Recall the formula to calculate the output size of a convolution: \n",
        "\n",
        " `output_size = (input_size - kernel_size) / stride + 1`\n",
        "\n",
        " or as defined in the slides (Lecture 5 slide 48):\n",
        "\n",
        " `output_size = (N - W) / stride + 1`\n",
        "\n",
        " Is this formula satisfied in the above example?\n",
        "3. How could you make the output (x) have the same width and height as the input (images)? (Think \"padding\"...)\n",
        "4. If the convolution operation performed above was a layer in a CNN, what would the number of parameters of that layer be?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2cjifwxHhELC"
      },
      "source": [
        "###2.2 MaxPool2d\n",
        "Reference to documentation: https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html#torch.nn.MaxPool2d"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "agvWu8aS_MvI"
      },
      "source": [
        "# Random input - dimensions are (batch_size, channels, height, width)\n",
        "x = torch.randn(64,6,28,28)\n",
        "\n",
        "# Perform max pool\n",
        "pool = nn.MaxPool2d(2, 2)\n",
        "y = pool(x)\n",
        "print('y.shape',y.shape)\n",
        "\n",
        "# Print some elements\n",
        "print('x[0,0,0:4,0:4]:\\n',x[0,0,0:4,0:4])\n",
        "print('y[0,0,0:2,0:2]:\\n',y[0,0,0:2,0:2])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ah3lVjAkhlSA"
      },
      "source": [
        "####2.2.1 Question\n",
        "1. What does MaxPool2d do? (Explain the output `y[0,0,0:2,0:2]` and compare to `x[0,0,0:2,0:2]`)\n",
        "2. What do each of the parameters (2, 2) mean?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Dr03cZUtiePH"
      },
      "source": [
        "###2.3 BatchNorm2d\n",
        "Reference to documentation: https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm2d.html#torch.nn.BatchNorm2d"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G8fNLPvkk97G"
      },
      "source": [
        "# Random input - dimensions are (batch_size, channels, height, width)\n",
        "x = torch.randn(64,6,28,28)*10 + 2\n",
        "print('x.mean',x.mean())\n",
        "print('x.std',x.std())\n",
        "\n",
        "# Perform batch normalization\n",
        "bn = nn.BatchNorm2d(6)\n",
        "y = bn(x)\n",
        "print('y.mean',y.mean())\n",
        "print('y.std',y.std())\n",
        "\n",
        "# Print parameter shapes\n",
        "for name, param in bn.named_parameters(): print('parameter',name,param.shape)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lTVNmKE1lodt"
      },
      "source": [
        "####2.3.1 Questions\n",
        "1. What does BacthNorm2D do?\n",
        "2. What does the input parameter (6) mean?\n",
        "3. Is the mean and standard deviation after batch normalization as expected?\n",
        "4. How many parameters does BatchNorm2D have? Why?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uotfgd9LmE71"
      },
      "source": [
        "###2.4 ReLU\n",
        "Reference to documentation: https://pytorch.org/docs/stable/generated/torch.nn.ReLU.html#torch.nn.ReLU"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qGdYW9i8mLke"
      },
      "source": [
        "# Random input - dimensions are (batch_size, channels, height, width)\n",
        "x = torch.randn(64,6,28,28)\n",
        "print('x.min',x.min())\n",
        "print('x.max',x.max())\n",
        "print('x[0,0,0:4,0:4]:\\n',x[0,0,0:4,0:4])\n",
        "\n",
        "# Apply ReLU\n",
        "y = F.relu(x)\n",
        "print('y.min',y.min())\n",
        "print('y.max',y.max())\n",
        "print('y[0,0,0:4,0:4]:\\n',y[0,0,0:4,0:4])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LIpw0OsNmh69"
      },
      "source": [
        "####2.4.1 Question\n",
        "1. What does ReLU do? (Explain the output `y[0,0,0:4,0:4]` and compare to `x[0,0,0:4,0:4]`)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZMdfxXxfnXNc"
      },
      "source": [
        "###2.5 Linear and view\n",
        "Reference to documentation:\n",
        "- https://pytorch.org/docs/stable/generated/torch.nn.Linear.html#torch.nn.Linear\n",
        "- https://pytorch.org/docs/stable/tensors.html#torch.Tensor.view\n",
        "\n",
        "Recall that CNNs for classification are divided into two blocks:\n",
        "\n",
        "1. **Convolutional base (encoder):** A sequence of convolutional layers resulting in a feature map that you may think of as a multi-dimensional image with a certain width, height, and depth (number of channels).\n",
        "2. **Fully connected layers (decoder):** One or more fully connected layers ending with a classifier.\n",
        "\n",
        "The input to the decoder has to be a vector with the correct dimensions. Thus, our task is to convert the feature map into a vector. In the code below we wish to take the feature map `x`, vectorize it, and feed it through a fully connected layer to produce 10-dimensional output.\n",
        "\n",
        "Your task is to figure out, what value `N` should have (see helper questions below):"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J8QTkgQsnkWQ"
      },
      "source": [
        "# Random input - dimensions are (batch_size, channels, height, width) \n",
        "x = torch.randn(64,6,28,28)\n",
        "\n",
        "N = # Your code goes here\n",
        "\n",
        "# Vectorize\n",
        "y = x.view(-1, N)\n",
        "print('y.shape',y.shape)\n",
        "\n",
        "# Fully connected layer\n",
        "fc = nn.Linear(N,10)\n",
        "z = fc(y)\n",
        "print('z.shape',z.shape)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_VFPz_h-puRE"
      },
      "source": [
        "The correct output shape (of z) is [64, 10], where 64 is the batch size and 10 is the dimensionality we want in the output side of the fully connected layer."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "J1b1TV5toah0"
      },
      "source": [
        "####2.5.1 Helper questions\n",
        "1. What does `view` do?\n",
        "2. What does `Linear` do?\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EykCYUemtgEa"
      },
      "source": [
        "##2.6 Dropout\n",
        "Reference to documentation: https://pytorch.org/docs/stable/generated/torch.nn.Dropout.html#torch.nn.Dropout\n",
        "\n",
        "Dropout is normally used only in relation to fully connected layers (i.e., not convolutional layers). The example below demonstrates the effect of the Dropout layer: "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rdvUyIHytmFM"
      },
      "source": [
        "# Random input - dimensions are (batch_size, num_in_features) \n",
        "x = torch.randn(64,10)\n",
        "print('x[0,:]',x[0,:])\n",
        "\n",
        "drop = nn.Dropout(p=0.5)\n",
        "y = drop(x)\n",
        "print('y[0,:]',y[0,:])\n",
        "\n",
        "print('ratios',y[0,:]/x[0,:])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "q2IUvi7uulBs"
      },
      "source": [
        "**IMPORTANT NOTE:** If necessary, re-run the code block until the printed part of `y[0,:]` contains both zeros and non-zeroes.\n",
        "\n",
        "####2.6.1 Questions\n",
        "1. Explain why there are zeroes in `y`.\n",
        "2. Ratios: Why are the values of the non-zero entries of `y` twice as high as the corresponding values of the input `x`?\n",
        "3. What happens to the ratios if you set `p=0.75`?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3JF7Iuo4-Ktj"
      },
      "source": [
        "##2.7 Setting up the model\n",
        "Your task is to fill in the blanks (???) below.\n",
        "\n",
        "Hint: You can insert print statements in the `forward` function, which is executed when calling `scores = model(images)`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Gj73e2T2-NXM"
      },
      "source": [
        "class Net(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(Net, self).__init__()\n",
        "        self.conv1 = nn.Conv2d(in_channels=3, out_channels=6, kernel_size=5, stride=1, padding=0)\n",
        "        self.bn1 = nn.BatchNorm2d(num_features=6)\n",
        "        self.pool = nn.MaxPool2d(kernel_size=2, stride=2)\n",
        "        self.conv2 = nn.Conv2d(in_channels=6, out_channels=16, kernel_size=5)\n",
        "        self.bn2 = nn.BatchNorm2d(num_features=???)\n",
        "        self.drop = nn.Dropout(p=0.5)\n",
        "        self.fc1 = nn.Linear(in_features=???, out_features=120)\n",
        "        self.fc2 = nn.Linear(in_features=???, out_features=10)\n",
        "\n",
        "    def forward(self, x):\n",
        "        x = self.conv1(x)\n",
        "        x = self.bn1(x)\n",
        "        x = F.relu(x)\n",
        "        x = self.pool(x)\n",
        "        x = self.conv2(x)\n",
        "        x = self.bn2(x)\n",
        "        x = F.relu(x)\n",
        "        x = self.pool(x)\n",
        "        x = x.view(-1, ???)\n",
        "        x = self.drop(x)\n",
        "        x = self.fc1(x)\n",
        "        x = F.relu(x)\n",
        "        x = self.fc2(x)\n",
        "        return x\n",
        "\n",
        "model = Net()\n",
        "\n",
        "print('Network architecture:\\n')\n",
        "print(model)\n",
        "\n",
        "# Print parameter shapes\n",
        "print('Network parameters:\\n')\n",
        "for name, param in model.named_parameters(): print('parameter',name,param.shape)\n",
        "\n",
        "# Test model\n",
        "scores = model(images)\n",
        "print(scores.shape)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qjLJIHy_rI76"
      },
      "source": [
        "##3. Task 2: Training the model\n",
        "In this task, just run the code, observe what happens and do your best to understand it."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Yi7JRPN09Oft"
      },
      "source": [
        "# First import what we need\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BZAtdZzzOTV0"
      },
      "source": [
        "###3.1 Setting up the CNN model\n",
        "First, lets set up a simple CNN with 2 convolutional layers, followed by two fully connected layers. The network is equivalent to the network defined above in Task 2. Each convolution operation is followed by batch normalization, ReLU, and max pooling (Note that instead of max pooling we could also have applied convolution with a stride of 2 to achieve the same level of down-sizing). The self-made Lambda layer helps us reshape the output if the last conv layer to a vector, because PyTorch has no built-in reshape layer. We perform dropout on this vector with p = 0.5, before applying the two fully connected layers."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dxomv8SqMeZX"
      },
      "source": [
        "class Lambda(nn.Module):\n",
        "    def __init__(self, func):\n",
        "        super().__init__()\n",
        "        self.func = func\n",
        "\n",
        "    def forward(self, x):\n",
        "        return self.func(x)\n",
        "\n",
        "def Net():\n",
        "  net = nn.Sequential(\n",
        "      nn.Conv2d(in_channels=3, out_channels=6, kernel_size=5, stride=1, padding=0),\n",
        "      nn.BatchNorm2d(num_features=6),\n",
        "      nn.ReLU(),\n",
        "      nn.MaxPool2d(kernel_size=2, stride=2),\n",
        "      nn.Conv2d(in_channels=6, out_channels=16, kernel_size=5),\n",
        "      nn.BatchNorm2d(num_features=16),\n",
        "      nn.ReLU(),\n",
        "      nn.MaxPool2d(kernel_size=2, stride=2),\n",
        "      Lambda(lambda x: x.view(x.size(0), -1)), # Vectorize\n",
        "      nn.Dropout(p=0.5),\n",
        "      nn.Linear(in_features=16*5*5,out_features=120),\n",
        "      nn.ReLU(),\n",
        "      nn.Linear(in_features=120, out_features=10)\n",
        "  )\n",
        "  return net\n",
        "\n",
        "model = Net()\n",
        "\n",
        "print('Network parameters:\\n')\n",
        "print(model)\n",
        "\n",
        "# Print parameter shapes\n",
        "for name, param in model.named_parameters(): print('parameter',name,param.shape)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "URqB3o81VhDt"
      },
      "source": [
        "Here is a neat function that you can call to re-initialize the weights/parameters of the model:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bulDYEyuUVWN"
      },
      "source": [
        "def reset_parameters(net):\n",
        "    '''Init layer parameters.'''\n",
        "    for m in net.modules():\n",
        "        if isinstance(m, nn.Conv2d):\n",
        "            torch.nn.init.kaiming_normal_(m.weight)\n",
        "            if m.bias is not None:\n",
        "                torch.nn.init.constant_(m.bias, 0)\n",
        "        elif isinstance(m, nn.BatchNorm2d):\n",
        "            torch.nn.init.constant_(m.weight, 1) # Why 1?\n",
        "            torch.nn.init.constant_(m.bias, 0) # Why 0?\n",
        "        elif isinstance(m, nn.Linear):\n",
        "            torch.nn.init.kaiming_normal_(m.weight)\n",
        "            if m.bias is not None:\n",
        "                torch.nn.init.constant_(m.bias, 0)\n",
        "\n",
        "reset_parameters(model)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TezEOgS6Qd0b"
      },
      "source": [
        "###3.2 Testing the model\n",
        "Let's test the model:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BZL4vNm1ylf9",
        "outputId": "f212e46b-ea22-4e08-bd97-369d9fbed3fe",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "# Move data to GPU\n",
        "images = images.cuda()\n",
        "labels = labels.cuda()\n",
        "\n",
        "# Calculate scores\n",
        "model = Net().cuda()\n",
        "scores = model(images)  # predictions\n",
        "\n",
        "print(scores.shape)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "torch.Size([64, 10])\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cCBQB7MkQrqo"
      },
      "source": [
        "###3.3 Calculating the accuracy\n",
        "The scores (logits) don't have any meaningful interpretation, but we can convert them into class probabilities using softmax. Since we are only going to be interested in the model's accuracy, we will wrap the softmax inside function accuracy that calculates the accuracy on a batch:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k7omGq3PxDL2",
        "outputId": "345f8357-ddb2-4bd2-d802-d1fee0e583c3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "def accuracy(scores, yb):\n",
        "    score2prob = nn.Softmax(dim=1)\n",
        "    preds = torch.argmax(score2prob(scores), dim=1)\n",
        "    return (preds == yb).float().mean()\n",
        "\n",
        "print('Accuracy', accuracy(scores,labels))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracy tensor(0.1094, device='cuda:0')\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5PW0hiPEQliS"
      },
      "source": [
        "###3.4 Calculating the loss\n",
        "I order to train your model, we also need a loss function. We will use the cross entropy loss already provided in PyTorch:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ckkpUHBYyi45",
        "outputId": "154d7b56-108d-4065-be7d-2e9a88ea0b0b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "loss_func = F.cross_entropy\n",
        "loss = loss_func(scores, labels)\n",
        "print('Loss', loss)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loss tensor(2.3281, device='cuda:0', grad_fn=<NllLossBackward>)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aXmfKk9tQ8He"
      },
      "source": [
        "###3.5 Training function (fit)\n",
        "The code below can be used to train a model and monitor important stats as training progresses. The training is carried out by calling the `fit` function, which takes any model as input.\n",
        "\n",
        "You can change the optimizer by replacing `base_optimizer` with your own function handle.\n",
        "\n",
        "The function `base_lr_scheduler` is a learning rate scheduler that updates the learning rate of the optimizer during training. Here it does does nothing, but you can modify it to implement your own learning rate scheduler. The inputs are\n",
        "- `T` : Total number of batches\n",
        "- `t` : Current batch index (max(t) = T)\n",
        "- `lr` : Current learning rate\n",
        "\n",
        "Other parameters to `fit` are:\n",
        "- `bs` the batch size\n",
        "- `epochs` the number of epochs\n",
        "- `batches_per_epoch` the number of batches per epoch. If set to `None`, all images in the dataset are used.\n",
        "\n",
        "Note that you can use `batches_per_epoch` to reduce the size of the training set. The effective size of the training set is `bs*batches_per_epoch`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PdmJibJ8xfvn"
      },
      "source": [
        "from torch import optim\n",
        "\n",
        "# Function handle that returns an optimizer\n",
        "def base_optimizer(model,lr=0.001, momentum=0.9):\n",
        "    return optim.SGD(model.parameters(), lr=lr,momentum=momentum)\n",
        "\n",
        "# Function handle that updates the learning rate\n",
        "# (note this is a dummy implementation that does nothing)\n",
        "def base_lr_scheduler(t,T,lr):\n",
        "  return lr\n",
        "\n",
        "# Function to fit a model\n",
        "def fit(model,\n",
        "        opt_func=base_optimizer,\n",
        "        lr_scheduler=base_lr_scheduler,\n",
        "        bs=256,\n",
        "        epochs=1,\n",
        "        batches_per_epoch=None, # Default: Use entire training set\n",
        "        show_summary=True):\n",
        "\n",
        "  # Set up data loaders\n",
        "  if batches_per_epoch == None:\n",
        "    # Use all images\n",
        "    train_dl = torch.utils.data.DataLoader(trainset, batch_size=bs,\n",
        "                                          shuffle=True, num_workers=2)\n",
        "    valid_dl = torch.utils.data.DataLoader(testset, batch_size=bs,\n",
        "                                         shuffle=False, num_workers=2)\n",
        "    batches_per_epoch = len(train_dl)\n",
        "  else:\n",
        "    # Only use a subset of the data\n",
        "    subset_indices = list(range(batches_per_epoch*bs))\n",
        "    train_dl = torch.utils.data.DataLoader(trainset, batch_size=bs, sampler=torch.utils.data.sampler.SubsetRandomSampler(subset_indices), num_workers=2)\n",
        "\n",
        "    # Use one fourth for validation\n",
        "    subset_indices = list(range(int(np.ceil(batches_per_epoch/4))*bs))\n",
        "    valid_dl = torch.utils.data.DataLoader(testset, batch_size=bs, sampler=torch.utils.data.sampler.SubsetRandomSampler(subset_indices), num_workers=2)\n",
        "\n",
        "  # Initialize optimizer\n",
        "  opt = opt_func(model)\n",
        "\n",
        "  # For book keeping\n",
        "  train_loss_history = []\n",
        "  valid_loss_history = []\n",
        "  plot_time_train = []\n",
        "  plot_time_valid = []\n",
        "\n",
        "  # Index of current batch\n",
        "  t = 1\n",
        "\n",
        "  # Total number of batches\n",
        "  T = batches_per_epoch * epochs\n",
        "  \n",
        "  print('Epochs:',epochs,'Batches per epoch:',batches_per_epoch,'Total number of batches',T)\n",
        "  \n",
        "  # Get initial validation loss and accuracy\n",
        "  model.eval()\n",
        "  with torch.no_grad():\n",
        "    valid_acc = sum(accuracy(model(xb.cuda()), yb.cuda()) for xb, yb in valid_dl) / len(valid_dl)\n",
        "    valid_loss = sum(loss_func(model(xb.cuda()), yb.cuda()) for xb, yb in valid_dl) / len(valid_dl)\n",
        "    valid_loss_history.append(valid_loss)\n",
        "    plot_time_valid.append(t)\n",
        "\n",
        "  # Train\n",
        "  for epoch in range(epochs):\n",
        "    model.train() # Train mode\n",
        "    for xb, yb in train_dl:\n",
        "\n",
        "      # Update learning rate\n",
        "      opt.param_groups[0]['lr'] = lr_scheduler(t,T,lr=opt.param_groups[0]['lr'])\n",
        "\n",
        "      # Forward prop\n",
        "      pred = model(xb.cuda())\n",
        "      loss = loss_func(pred, yb.cuda())\n",
        "\n",
        "      # Book keeping\n",
        "      train_loss_history.append(loss.detach().cpu().numpy())\n",
        "      plot_time_train.append(t)\n",
        "      t += 1\n",
        "\n",
        "      # Backward prop (calculate gradient)\n",
        "      loss.backward()\n",
        "\n",
        "      # Update model parameters\n",
        "      opt.step()\n",
        "      opt.zero_grad()\n",
        "\n",
        "      # Validation loss and accuracy\n",
        "      if t % 10 == 0:    # print every 10 mini-batches\n",
        "        model.eval() # Test mode\n",
        "        with torch.no_grad():\n",
        "            valid_acc = sum(accuracy(model(xb.cuda()), yb.cuda()) for xb, yb in valid_dl) / len(valid_dl)\n",
        "            valid_loss = sum(loss_func(model(xb.cuda()), yb.cuda()) for xb, yb in valid_dl) / len(valid_dl)\n",
        "            valid_loss_history.append(valid_loss)\n",
        "            plot_time_valid.append(t-1)\n",
        "            print('t',t,'lr',opt.param_groups[0]['lr'],'train loss',loss.detach().cpu().numpy(), 'val loss',valid_loss.detach().cpu().numpy(),'val accuracy', valid_acc.detach().cpu().numpy())\n",
        "        model.train() # Back to train mode\n",
        "\n",
        "  # Summary\n",
        "  if show_summary:\n",
        "    plt.figure()\n",
        "    lines = []\n",
        "    labels = []\n",
        "    l, = plt.plot(plot_time_train,train_loss_history)\n",
        "    lines.append(l)\n",
        "    labels.append('Training')\n",
        "    l, = plt.plot(plot_time_valid,valid_loss_history)\n",
        "    lines.append(l)\n",
        "    labels.append('Validation')  \n",
        "    plt.title('Loss')\n",
        "    plt.legend(lines, labels, loc=(1, 0), prop=dict(size=14))\n",
        "    plt.show()\n",
        "\n",
        "  return train_loss_history"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "H0brbs2QTF4o"
      },
      "source": [
        "###3.6 Train the model\n",
        "Train the model and inspect the loss curves."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OmWk0LUvy18m"
      },
      "source": [
        "# Re-initialize weights\n",
        "reset_parameters(model) \n",
        "\n",
        "# Train with defaul settings.\n",
        "train_loss_history = fit(model)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dvABXtUhGSad"
      },
      "source": [
        "**Note:** The validation loss (on the unseen data) is overall *lower* than the training loss. This might initially seem like an unexpected behaviour, but it results from the fact that we use 50% Dropout during training, meaning that half of the weights of the fully connected layers are not used during training. When evaluating the model on the validation set, Dropout is disabled, and all weights are used. This leads to a better prediction (i.e., a lower loss) on the validation set."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "u1N4oj_tRABK"
      },
      "source": [
        "###3.7 Performing quick experiments on reduced dataset\n",
        "When searching for hyperparameters it is often useful to perform quick experiments on only a small subset of the data. You can do this by decreasing the batch size and setting `batches_per_epoch` to say 20. The effective size of the training set is then `bs*batches_per_epoch` (say `128*20 = 2,560`), and the total number of batches being processed during training is `bs*batches_per_epoch*epochs` (say `128*20*5 = 51,200`).\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zYEcxzoHQxoT"
      },
      "source": [
        "# Re-initialize weights\n",
        "reset_parameters(model)\n",
        "\n",
        "# Example: Train on small subset\n",
        "train_loss_history = fit(model,bs=128,epochs=5,batches_per_epoch=20)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TnsvkrbOWl3r"
      },
      "source": [
        "###3.7 Train with custom optimizer\n",
        "Fell free to experiment with these optimizers."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5BXxefozVuWo"
      },
      "source": [
        "#SGD + Momentum\n",
        "def momentum_optimizer(model):\n",
        "    return optim.SGD(model.parameters(), lr=0.1, momentum=0.9)\n",
        "\n",
        "# Adagrad\n",
        "def adagrad_optimizer(model):\n",
        "  return optim.Adagrad(model.parameters(), lr=0.1, lr_decay=0.0, weight_decay=0.0, initial_accumulator_value=0)\n",
        "\n",
        "# RMSProp\n",
        "def rmsprop_optimizer(model):\n",
        "  return optim.RMSprop(model.parameters(), lr=0.1, alpha=0.99, eps=1e-08, weight_decay=0, momentum=0, centered=False)\n",
        "\n",
        "# Adam\n",
        "def adam_optimizer(model):\n",
        "  return optim.Adam(model.parameters(), lr=0.001, betas=(0.9, 0.999), eps=1e-08, weight_decay=0, amsgrad=False)\n",
        "\n",
        "# Train with custom optimizer (on small subset)\n",
        "reset_parameters(model) # Reset network weights\n",
        "train_loss_history = fit(model,opt_func=momentum_optimizer,bs=128,epochs=5,batches_per_epoch=20)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UbnzkuxTX4wV"
      },
      "source": [
        "##4. Tasks: You are free to choose any of the tasks below\n",
        "The overall goal is to make the model perform is good as possible on the CIFAR 10 dataset. Solving the tasks below might help you, but feel free to try other options. For inspiration, here is a list of other things, you could do:\n",
        "\n",
        "- **Increase the capacity of the network** (add more conv layers, add more kernels in each layer, add another fully connected layer, add more connections in the fully connected layer, etc.)\n",
        "- **Experiment with weight decay** (weight decay is an option for, say, the [SGD optimizer](https://pytorch.org/docs/stable/optim.html#torch.optim.SGD))\n",
        "\n",
        "General recommendations for hyperparameter search (see Lecture 6 slides):\n",
        "- Stage your search from course to fine\n",
        "- Start with a small sample (subset) of the dataset to tune in on hyperparameters such as learning rate, and start with 1 epoch or less\n",
        "- The second stage could then perform a narrower search with more data and/or more epochs\n",
        "- Search for learning rates in powers of 10. Good learning rates to try: 1e-1, 1e-2, 1e-3, 1e-4, etc.)\n",
        "- **Look at loss curves!!!**\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hwjZvWPzN-LE"
      },
      "source": [
        "###4.1 Implement a learning rate scheduler\n",
        "**Your task** is to replace the function `base_lr_scheduler` with your own `step_decay` function (or another type of learning rate scheduler if you prefer). In step decay the lerning rate is updated for every `step` batches according to the rule `lr = lr * decay_rate`. The input `t` is the current batch index, ranging from 1 to `T`, where `T = batches_per_epoch * epochs` is the total number of batches."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qJbbqKF6a5AI"
      },
      "source": [
        "def step_decay(t,T,lr,step=10,decay_rate=0.9):\n",
        "  # Your code goes here\n",
        "\n",
        "# Train\n",
        "model = Net().cuda()\n",
        "reset_parameters(model)\n",
        "train_loss_history = fit(model,lr_scheduler=step_decay,bs=128,epochs=5,batches_per_epoch=20)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "79vJl3fENgXF"
      },
      "source": [
        "**Sub-tasks:**\n",
        "- What new hyperparameters are added with learning rate decay?\n",
        "- See if you can make the model perform better with learning rate decay."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eWy7obmYYaWY"
      },
      "source": [
        "###4.2 Implement a learning rate finder\n",
        "Finding a good initial learning rate can be done efficiently with a learning rate finder. Here is the basic idea:\n",
        "\n",
        "You can modify the learning rate scheduler `base_lr_scheduler` to use a hardcoded list of learning rates. This can be uses to implement a learning rate finder according to this principle:\n",
        "- Start training and increase learning rate linearly after each batch (!!!), and calculate the loss\n",
        "- Then display the loss as a function of learning rate.\n",
        "- By manual inspection, select a suitable range of learning rates by locating the strongest downward slope.\n",
        "\n",
        "Say the hardcoded learning rates are:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "n2b6Vl1bZkzn"
      },
      "source": [
        "# Random un-ordered list of learning rates\n",
        "lr_list = [0.1,0.5,0.01]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "f9c3CL-xO0YD"
      },
      "source": [
        "Then simply implement a learning rate scheduler that returns the t´th entry of the learning rate list:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "z0GyYWN2Lb8Q"
      },
      "source": [
        "def lr_finder(t,T,lr):\n",
        "  return lr_list[t-1]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "75cMHVsJLc3K"
      },
      "source": [
        "Note, the learning rate scheduler must have this interface: `(t,T,lr)`\n",
        "\n",
        "Then train the model for as many batches as there are entries in the learning rate list:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NN1oLDpqZwNB",
        "outputId": "2ff61067-b07a-4ba9-d6fe-3881f0bae983",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "# Make sure that we process as many batches as there are learning rates in our list:\n",
        "batches_per_epoch = len(lr_list)\n",
        "epochs = 1\n",
        "\n",
        "# Train\n",
        "model = Net().cuda()\n",
        "reset_parameters(model)\n",
        "train_loss_history = fit(model,lr_scheduler=lr_finder,epochs=epochs,batches_per_epoch=batches_per_epoch,show_summary=False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epochs: 1 Batches per epoch: 3 Total number of batches 3\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_6ZOzLWjLvrl"
      },
      "source": [
        "Plot the loss versus learning rate and inspect:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kgPokRI_L2Tx",
        "outputId": "f0c5daf5-b326-49bb-b7ef-ef89306fb301",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        }
      },
      "source": [
        "plt.figure()\n",
        "plt.plot(train_loss_history)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[<matplotlib.lines.Line2D at 0x7f9790b3b4a8>]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 36
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAD4CAYAAAD8Zh1EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "o1Zzl2zcQESM"
      },
      "source": [
        "**Your task** is to implement a learning rate finder that searches through a range of learning rates, like above.\n",
        "\n",
        "The range of learning rates should be logarithmic (say from 10^-6 to 10^1). You can define the list of learning rates using `np.logspace`.\n",
        "\n",
        "Then plot the loss versus the learning rate. It is a good idea to use a logarithmic scale on the x-axis (see `plt.xscale('log')`)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3ESD1B4JQc1S"
      },
      "source": [
        "# Your code goes here"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JMkJAqzXU8q7"
      },
      "source": [
        "**Optional sub-task:** Can you come up with a simple way to implement Cyclic Learning Rate (CLR) based on the above?\n",
        "\n",
        "Hint: Run training multiple times using a carefully selected list of learning rates each time."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6bd9PAX1VIJu"
      },
      "source": [
        "# Your code goes here"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "h5zgyxJjXNYv"
      },
      "source": [
        "###4.3 Experiment with dropout\n",
        "Below is a slight modification of our model that allows changing the dropout fraction (p).\n",
        "\n",
        "**Your task** is to experiment with different values of p.\n",
        "\n",
        "- What does setting p = 0 mean?\n",
        "- Try different values of p (say 0, 0.1, 0.5 and 0.9). What is the optimal value of p?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YF_mWFkCXhw1"
      },
      "source": [
        "def Net(p=0.5):\n",
        "  net = nn.Sequential(\n",
        "      nn.Conv2d(in_channels=3, out_channels=6, kernel_size=5, stride=1, padding=0),\n",
        "      nn.BatchNorm2d(num_features=6),\n",
        "      nn.ReLU(),\n",
        "      nn.MaxPool2d(kernel_size=2, stride=2),\n",
        "      nn.Conv2d(in_channels=6, out_channels=16, kernel_size=5),\n",
        "      nn.BatchNorm2d(num_features=16),\n",
        "      nn.ReLU(),\n",
        "      nn.MaxPool2d(kernel_size=2, stride=2),\n",
        "      Lambda(lambda x: x.view(x.size(0), -1)), # Vectorize\n",
        "      nn.Dropout(p=p),\n",
        "      nn.Linear(in_features=16*5*5,out_features=120),\n",
        "      nn.ReLU(),\n",
        "      nn.Linear(in_features=120, out_features=10)\n",
        "  )\n",
        "  return net"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dLbTJvsvYHIl"
      },
      "source": [
        "# Model with p=0.25\n",
        "model = Net(p=0.25).cuda()\n",
        "\n",
        "reset_parameters(model)\n",
        "train_loss_history = fit(model,bs=128,epochs=5,batches_per_epoch=20)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vZG7wXrgbIRh"
      },
      "source": [
        "###4.4 Data augmentation\n",
        "Adding more training data is the best way to improve your models abiliy to generalize to unseen data. Data augmentations is basically about generating additional *fake data\" to expand the size of your training set.\n",
        "\n",
        "**Your task** is to experiment with different types of data augmentation. All you have to do is change the transformer below and add some of the available transforms: https://pytorch.org/docs/stable/torchvision/transforms.html\n",
        "\n",
        "You can find inspiration in Lab 4: https://github.com/aivclab/dlcourse/blob/master/Lab_4_(part3)_PyTorch_Transfer_Learning.ipynb\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2Rurvqei1ZF8"
      },
      "source": [
        "mean = 0.5\n",
        "std = 0.5\n",
        "transform = transforms.Compose(\n",
        "    [ # Add your transforms here (before calling ToTensor),\n",
        "      transforms.ToTensor(),\n",
        "      transforms.Normalize((mean, mean, mean), (std, std, std))\n",
        "     ])\n",
        "\n",
        "# Batch size\n",
        "bs = 64\n",
        "\n",
        "trainset = torchvision.datasets.CIFAR10(root='./data', train=True,\n",
        "                                        download=True, transform=transform)\n",
        "trainloader = torch.utils.data.DataLoader(trainset, batch_size=bs,\n",
        "                                          shuffle=True, num_workers=2,drop_last=True)\n",
        "\n",
        "testset = torchvision.datasets.CIFAR10(root='./data', train=False,\n",
        "                                       download=True, transform=transform)\n",
        "testloader = torch.utils.data.DataLoader(testset, batch_size=bs,\n",
        "                                         shuffle=False, num_workers=2,drop_last=True)\n",
        "\n",
        "# Get a batch of some random training images\n",
        "dataiter = iter(trainloader)\n",
        "images, labels = dataiter.next()\n",
        "print('images.shape',images.shape)\n",
        "\n",
        "# show 10 images and print labels\n",
        "imshow(torchvision.utils.make_grid(images[0:16]))\n",
        "print(' '.join('%5s' % classes[labels[j]] for j in range(16)))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u8KrOsoCNoeZ"
      },
      "source": [
        "# Train\n",
        "model = Net().cuda()\n",
        "reset_parameters(model)\n",
        "train_loss_history = fit(model,bs=128,epochs=5,batches_per_epoch=20)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Vnl8gRJqP8nl"
      },
      "source": [
        "###4.5 Multi-task learning (MTL)\n",
        "By sharing representations between related tasks, we can enable our model to generalize better on our original task. This is the basic idea behind MTL.\n",
        "\n",
        "There is one additional task that we can always add, and that is an **autoencoder**. Recall that an autoencoder is basically an identity function that maps input x to itself. It does so by first encoding x into a more compact representation E(x), and then decodes this representation to get a reconstruction of x = D(E(x)), where E() is the encoder, and D() is the decoder.\n",
        "\n",
        "The network below uses the **same encoder** to solve both a **classification task** and an **autoencoder task**."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5rmVdxMMgCy9"
      },
      "source": [
        "# define the NN architecture\n",
        "class MTLNet(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(MTLNet, self).__init__()\n",
        "\n",
        "        ## shared encoder layers ##\n",
        "        self.conv1 = nn.Conv2d(3, 16, 5, stride=2, padding=2)\n",
        "        self.conv2 = nn.Conv2d(16, 4, 5, stride=2, padding=2)\n",
        "        \n",
        "        ## decoder layers ##\n",
        "        # Classifier\n",
        "        self.fc1 = nn.Linear(in_features=4*8*8, out_features=120)\n",
        "        self.fc2 = nn.Linear(in_features=120, out_features=10)\n",
        "\n",
        "        ## Autoencoder\n",
        "        self.t_conv1 = nn.ConvTranspose2d(4, 16, 2, stride=2)\n",
        "        self.t_conv2 = nn.ConvTranspose2d(16, 3, 2, stride=2)\n",
        "\n",
        "    def forward(self, x):\n",
        "        ## encode ##\n",
        "        x = F.relu(self.conv1(x))\n",
        "        x = F.relu(self.conv2(x))\n",
        "        \n",
        "        # decoder for task 1 - outputs scores for classifier\n",
        "        scores = x.view(-1, 4*8*8)\n",
        "        scores = F.relu(self.fc1(scores))\n",
        "        scores = self.fc2(scores)\n",
        "\n",
        "        ## decoder for task 2 - outputs reconstruction of input x\n",
        "        recon = F.relu(self.t_conv1(x))\n",
        "        recon = self.t_conv2(recon)\n",
        "                \n",
        "        return scores, recon\n",
        "\n",
        "# initialize the NN\n",
        "model = MTLNet()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "V-AO2RNBQ2S6"
      },
      "source": [
        "Let's look at the output of this model:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0XRN0F3ws4Y4",
        "outputId": "5aa66e44-3a8e-43cd-caf0-db172104acca",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 374
        }
      },
      "source": [
        "# Get a batch of some random training images\n",
        "dataiter = iter(trainloader)\n",
        "images, labels = dataiter.next()\n",
        "\n",
        "scores,recon = model(images)\n",
        "print('scores.shape',scores.shape)\n",
        "print('recon.shape',recon.shape)\n",
        "\n",
        "# show 16 images and print labels\n",
        "print('\\nInput images:\\n')\n",
        "imshow(torchvision.utils.make_grid(images[0:16]))\n",
        "print('Reconstructed images:\\n')\n",
        "imshow(torchvision.utils.make_grid(recon[0:16].detach().cpu()))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "scores.shape torch.Size([64, 10])\n",
            "recon.shape torch.Size([64, 3, 32, 32])\n",
            "\n",
            "Input images:\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXAAAAB4CAYAAADrPanmAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Reconstructed images:\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXAAAAB4CAYAAADrPanmAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n8QAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Z_imIbERRY21"
      },
      "source": [
        "In order to optimize our MTL model we need a loss function. However, since we are solving two tasks, we also need two terms in our loss function - one for each task.\n",
        "\n",
        "The loss for the classification task is the familiar cross entropy:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-0-KmzXxjbIe",
        "outputId": "155bd8be-b2f5-47fe-c23e-caf3d8104e8b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "# Classification loss\n",
        "loss_func_1 = F.cross_entropy\n",
        "loss1 = loss_func_1(scores, labels)\n",
        "print('Loss 1', loss1)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loss 1 tensor(2.3157, grad_fn=<NllLossBackward>)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YEPtN5-3Rw0v"
      },
      "source": [
        "A suitable loss for the autoencoder task would be the Mean Squared Error (MSE) between the input images (say `images`) and the networks reonstruction (say `recon`): "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iz02YBIBlBH9",
        "outputId": "da85d2ce-f64e-42be-8167-194eada71ded",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "# MSE loss for reconstruction\n",
        "loss_func_2 = F.mse_loss\n",
        "loss2 = loss_func_2(images,recon)\n",
        "print('Loss 2', loss2)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loss 2 tensor(0.2794, grad_fn=<MeanBackward0>)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cBXZ074LSRdB"
      },
      "source": [
        "Below is a modified function to fit the MTL model.\n",
        "\n",
        "**Your task** is simply to fill in the missing code (marked with ???). This is not as trivial as you might think (hint: do the two losses have the same scale?)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QrOnBz0sgt1b"
      },
      "source": [
        "# Function to fit a model\n",
        "def fit(model,\n",
        "        opt_func=base_optimizer,\n",
        "        lr_scheduler=base_lr_scheduler,\n",
        "        bs=256,\n",
        "        epochs=1,\n",
        "        batches_per_epoch=None,\n",
        "        show_summary=True):\n",
        "\n",
        "  # Set up data loaders\n",
        "  if batches_per_epoch == None:\n",
        "    # Use all images\n",
        "    train_dl = torch.utils.data.DataLoader(trainset, batch_size=bs,\n",
        "                                          shuffle=True, num_workers=2)\n",
        "    valid_dl = torch.utils.data.DataLoader(testset, batch_size=bs,\n",
        "                                         shuffle=False, num_workers=2)\n",
        "    batches_per_epoch = len(train_dl)\n",
        "  else:\n",
        "    # Only use a subset of the data\n",
        "    subset_indices = list(range(batches_per_epoch*bs))\n",
        "    train_dl = torch.utils.data.DataLoader(trainset, batch_size=bs, sampler=torch.utils.data.sampler.SubsetRandomSampler(subset_indices), num_workers=2)\n",
        "\n",
        "    subset_indices = list(range(int(np.ceil(batches_per_epoch/4))*bs))\n",
        "    valid_dl = torch.utils.data.DataLoader(testset, batch_size=bs, sampler=torch.utils.data.sampler.SubsetRandomSampler(subset_indices), num_workers=2)\n",
        "  \n",
        "  loss1_history = []\n",
        "  loss2_history = []\n",
        "  plot_time_train = []\n",
        "  \n",
        "  # Initialize optimizer\n",
        "  opt = opt_func(model)\n",
        "\n",
        "  t = 1 # Index of current batch\n",
        "  T = batches_per_epoch * epochs # Total number of batches\n",
        "  \n",
        "  print('Epochs:',epochs,'Batches per epoch:',batches_per_epoch,'Total number of batches',T)\n",
        "  \n",
        "  # Train\n",
        "  for epoch in range(epochs):\n",
        "    model.train() # Train mode\n",
        "    for xb, yb in train_dl:\n",
        "\n",
        "      # Update learning rate\n",
        "      opt.param_groups[0]['lr'] = lr_scheduler(t,T,lr=opt.param_groups[0]['lr'])\n",
        "\n",
        "      # Forward prop\n",
        "      scores,recon = model(xb.cuda())\n",
        "      loss1 = loss_func_1(scores, yb.cuda())\n",
        "      loss2 = loss_func_2(recon, xb.cuda())\n",
        "      \n",
        "      loss = ??? # Your code goes here\n",
        "\n",
        "      if t % 10 == 0:    # print every 10 mini-batches\n",
        "        print('loss1',loss1.detach().cpu().numpy(),'loss2',loss2.detach().cpu().numpy())\n",
        "\n",
        "      loss1_history.append(loss1.detach().cpu().numpy())\n",
        "      loss2_history.append(loss2.detach().cpu().numpy())\n",
        "      plot_time_train.append(t)\n",
        "      t += 1\n",
        "\n",
        "      # Backward prop (calculate gradient)\n",
        "      loss.backward()\n",
        "\n",
        "      # Update model parameters\n",
        "      opt.step()\n",
        "      opt.zero_grad()\n",
        "\n",
        "  # Summary\n",
        "  if show_summary:\n",
        "    plt.figure()\n",
        "    lines = []\n",
        "    labels = []\n",
        "    l, = plt.plot(plot_time_train,loss1_history)\n",
        "    lines.append(l)\n",
        "    labels.append('Loss1 (classification)')\n",
        "    l, = plt.plot(plot_time_train,loss2_history)\n",
        "    lines.append(l)\n",
        "    labels.append('Loss2 (autoencoder)')\n",
        "    plt.title('Loss')\n",
        "    plt.legend(lines, labels, loc=(1, 0), prop=dict(size=14))\n",
        "    plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5toVL8DfmHK9"
      },
      "source": [
        "model = MTLNet().cuda()\n",
        "\n",
        "# Re-initialize weights\n",
        "reset_parameters(model)\n",
        "\n",
        "# Example (in practise you need more epochs than 5)\n",
        "train_loss_history = fit(model,opt_func=momentum_optimizer,bs=128,epochs=5,batches_per_epoch=20)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2TWEuL3lVlWo"
      },
      "source": [
        "Look at some results:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3GjuU2vUmPID"
      },
      "source": [
        "scores,recon = model(images.cuda())\n",
        "\n",
        "# validation accuracy\n",
        "acc = 0\n",
        "for xb, yb in testloader:\n",
        "  scores,recon = model(xb.cuda())\n",
        "  acc += accuracy(scores,yb.cuda())\n",
        "print('accuracy',acc/len(testloader))\n",
        "\n",
        "# show 16 images\n",
        "dataiter = iter(trainloader)\n",
        "images, labels = dataiter.next()\n",
        "\n",
        "print('\\nInput images:\\n')\n",
        "imshow(torchvision.utils.make_grid(images[0:16]))\n",
        "print('Reconstructed images:\\n')\n",
        "imshow(torchvision.utils.make_grid(recon[0:16].detach().cpu()))"
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}