worked-solutions/03_mnist_classification_solutions.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Classifying handwritten digits with a CNN\n",
    "\n",
    "\n",
    "In the previous exercise, we worked with fully-connected neural networks, which are good at handling tabular data, where the inputs and targets are easily presented as vectors.\n",
    "\n",
    "However, in the case of images, or image-like objects, such models are less efficient for reasons we have discussed in the slides. When inputs are images, or image-like data, a more natural choice of model is a convolution neural network—in particularly, a model which uses 2D convolutional layers.\n",
    "\n",
    "Before we start worrying about choosing models, let's first acquaint ourselves with the MNIST data.\n",
    "\n",
    "The first step is to select a directory for the data to live."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_root = \"mnist-data/\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If we all set a path this way it will help to maintain consistency throughout this exercise."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 1: Access the MNIST dataset.\n",
    "\n",
    "PyTorch has a (sort of) sister Python library for dealing with images: [``Torchvision``](https://pytorch.org/vision/stable/index.html) (take a look at the website for a few minutes).\n",
    "\n",
    "In the previous exercise, we used a custom ``Dataset`` object created specifically for this event, but with ``Torchvision`` come several easy-to-use datasets, one of which is the MNIST digits.\n",
    "\n",
    "- Look at the arguments of the [MNIST datset](https://pytorch.org/vision/stable/generated/torchvision.datasets.MNIST.html): what options do we have?\n",
    "- Instantiate the (training) dataset.\n",
    "- Iterate over it: how are the inputs and targets presented to us?\n",
    "- Plot some images, and set their targets as the title, to make sure the data make sense.\n",
    "\n",
    "Note: this section might be challenging for many people, so we will go through it together first."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset MNIST\n",
      "    Number of datapoints: 60000\n",
      "    Root location: mnist-data/\n",
      "    Split: Train\n",
      "\n",
      "\n",
      "Input is: <PIL.Image.Image image mode=L size=28x28 at 0x156F5A290>\n",
      "\n",
      "Target is: 5\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 15 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from torchvision.datasets import MNIST\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "data_set = MNIST(root=mnist_root, download=True)\n",
    "\n",
    "print(data_set)\n",
    "\n",
    "img_i, target_i = data_set[0]\n",
    "print(\"\\n\")\n",
    "print(f\"Input is: {img_i}\\n\")\n",
    "print(f\"Target is: {target_i}\\n\")\n",
    "\n",
    "fig, axes = plt.subplots(3, 5)\n",
    "\n",
    "for i, ax in enumerate(axes.ravel()):\n",
    "    img_i, target_i = data_set[i]\n",
    "    ax.imshow(img_i, cmap=\"inferno\")\n",
    "    ax.set_title(f\"{target_i}\")\n",
    "    ax.set_axis_off()\n",
    "\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "\n",
    "### Task 2\n",
    "\n",
    "As before, we have a dataset which gives us inputs and targets, but we still need to convert the data to ``torch.Tensors``\n",
    "\n",
    "#### Part (a) — raw data to ``Tensor``s\n",
    "\n",
    "Let's do this basic steps first: supply transforms to map between the raw data and ``torch.Tensors``.\n",
    "\n",
    "\n",
    "\n",
    "#### Part (b) — fun with data augmentation\n",
    "\n",
    "Additionally, with image data there are some other consideration we might make:\n",
    "- Are CNNs rotationally invariant?\n",
    "  - If we want our model to work on images which are not of a regular orientation, we must use random rotations as a form of augmentation.\n",
    "- If we train a model on purely black-and-white images, how will it fare on more colourful data?\n",
    "- Go to ``Torchvision``'s [transforms](https://pytorch.org/vision/stable/transforms.html) and look at the available forms of images augmentation.\n",
    "  - Take a few minutes to pick ones you think might be relevant.\n",
    "  - Let's discuss and choose some:\n",
    "    - `RandomRotation` will prevent rotational invariance as discussed in the lectures.\n",
    "      - It will be interesting to see how well the net handles the difference between 6 and 9, however.\n",
    "    - `ColorJitter` applies random variation to the images in terms of hue, saturation, brightness, and contrast. This will enhance feature detection for this dataset.\n",
    "    - ..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1500x1500 with 25 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from torchvision.transforms import ToTensor, Compose, RandomRotation, ColorJitter\n",
    "from torch import Tensor, tensor, float32, eye\n",
    "\n",
    "\n",
    "def get_img_tfms(training: bool) -> Compose:\n",
    "    \"\"\"Return a composition of image transforms.\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    training : bool\n",
    "        Are we training? If ``False``, we are validating.\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    Compose\n",
    "        A composition of image transforms.\n",
    "\n",
    "    \"\"\"\n",
    "    # Convert the image from being a 2D tensor to have 3 values for each pixel (RGB).\n",
    "    # Since the three values are the same it remains the same grayscale image\n",
    "    tfm_list: List[Callable] = [ToTensor(), lambda x: x.repeat(3, 1, 1)]\n",
    "\n",
    "    if training is True:\n",
    "        # tfm_list.append(???)\n",
    "        tfm_list.append(RandomRotation(180))\n",
    "        tfm_list.append(\n",
    "            ColorJitter(brightness=0.1, hue=0.1, saturation=0.1, contrast=0.1)\n",
    "        )\n",
    "        pass\n",
    "\n",
    "    return Compose(tfm_list)\n",
    "\n",
    "\n",
    "def target_as_tensor(target_idx: int) -> Tensor:\n",
    "    \"\"\"Express the target as a ``torch.Tensor``.\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    target_idx : int\n",
    "        The ground truth (i.e. the number shown in the image).\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    Tensor\n",
    "        The target as a one-hot-encoded vector.\n",
    "\n",
    "    \"\"\"\n",
    "    return eye(10)[target_idx]\n",
    "\n",
    "\n",
    "train_set = MNIST(\n",
    "    mnist_root,\n",
    "    transform=get_img_tfms(training=True),\n",
    "    target_transform=target_as_tensor,\n",
    "    train=True,\n",
    ")\n",
    "\n",
    "\n",
    "valid_set = MNIST(\n",
    "    mnist_root,\n",
    "    transform=get_img_tfms(training=False),\n",
    "    target_transform=target_as_tensor,\n",
    "    train=False,\n",
    ")\n",
    "\n",
    "figure, axes = plt.subplots(5, 5, figsize=(15, 15))\n",
    "\n",
    "for idx, axis in enumerate(axes.ravel()):\n",
    "    image, target = train_set[idx]\n",
    "\n",
    "    axis.imshow(image.permute(1, 2, 0), cmap=\"inferno\")\n",
    "    axis.set_title(target.argmax())\n",
    "    axis.set_axis_off()\n",
    "\n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 3: ``Dataset`` $\\to$ ``DataLoader``\n",
    "\n",
    "As before, wrap the ``Dataset``s in ``DataLoader``s."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n",
      "torch.Size([64, 3, 28, 28]) torch.Size([64, 10])\n"
     ]
    }
   ],
   "source": [
    "from torch.utils.data import DataLoader\n",
    "\n",
    "batch_size = 64\n",
    "\n",
    "train_loader = DataLoader(\n",
    "    train_set,\n",
    "    batch_size=batch_size,\n",
    "    shuffle=True,\n",
    "    drop_last=True,\n",
    ")\n",
    "\n",
    "valid_loader = DataLoader(\n",
    "    valid_set,\n",
    "    batch_size=batch_size,\n",
    "    shuffle=False,\n",
    ")\n",
    "\n",
    "for _, (batch, targets) in zip(range(10), valid_loader):\n",
    "    print(batch.shape, targets.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 4: Choose a model architecture\n",
    "\n",
    "Torchvision provides a collection of models, [here](https://pytorch.org/vision/stable/models.html).\n",
    "\n",
    "Since we are all using laptops, many of which don't have CUDA-enabled GPUs, we will select a modest neural network than won't melt any of our hardware. Such a network, designed for mobile phones, is [``MOBILENET``](https://pytorch.org/vision/stable/models/mobilenetv3.html).\n",
    "\n",
    "#### Task 4 (a): instantiate the small version of ``MOBILENET``, and print it out.\n",
    "\n",
    "Note:\n",
    "- Torchvision's models can optionally be endowed with pretrained weights (from corresponding models pretrain on the ImageNet dataset).\n",
    "- We can (optionally) supply these weights.\n",
    "  - Using weights from a model, trained on one problem, as an initial condition in another problem, is called transfer learning.\n",
    "  - Why do you think this might be advantageous, even in disparate problems?\n",
    "    - It will already be trained to recognise some generic/abstract features.\n",
    "    - ..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MobileNetV3(\n",
      "  (features): Sequential(\n",
      "    (0): Conv2dNormActivation(\n",
      "      (0): Conv2d(3, 16, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
      "      (1): BatchNorm2d(16, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "      (2): Hardswish()\n",
      "    )\n",
      "    (1): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(16, 16, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=16, bias=False)\n",
      "          (1): BatchNorm2d(16, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): ReLU(inplace=True)\n",
      "        )\n",
      "        (1): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(16, 8, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(8, 16, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (2): Conv2dNormActivation(\n",
      "          (0): Conv2d(16, 16, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(16, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (2): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(16, 72, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(72, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): ReLU(inplace=True)\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(72, 72, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=72, bias=False)\n",
      "          (1): BatchNorm2d(72, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): ReLU(inplace=True)\n",
      "        )\n",
      "        (2): Conv2dNormActivation(\n",
      "          (0): Conv2d(72, 24, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(24, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (3): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(24, 88, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(88, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): ReLU(inplace=True)\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(88, 88, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=88, bias=False)\n",
      "          (1): BatchNorm2d(88, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): ReLU(inplace=True)\n",
      "        )\n",
      "        (2): Conv2dNormActivation(\n",
      "          (0): Conv2d(88, 24, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(24, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (4): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(24, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(96, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(96, 96, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=96, bias=False)\n",
      "          (1): BatchNorm2d(96, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (2): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(96, 24, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(24, 96, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (3): Conv2dNormActivation(\n",
      "          (0): Conv2d(96, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(40, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (5): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(240, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(240, 240, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=240, bias=False)\n",
      "          (1): BatchNorm2d(240, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (2): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(240, 64, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(64, 240, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (3): Conv2dNormActivation(\n",
      "          (0): Conv2d(240, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(40, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (6): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(240, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(240, 240, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=240, bias=False)\n",
      "          (1): BatchNorm2d(240, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (2): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(240, 64, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(64, 240, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (3): Conv2dNormActivation(\n",
      "          (0): Conv2d(240, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(40, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (7): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(40, 120, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(120, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(120, 120, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=120, bias=False)\n",
      "          (1): BatchNorm2d(120, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (2): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(120, 32, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(32, 120, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (3): Conv2dNormActivation(\n",
      "          (0): Conv2d(120, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(48, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (8): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(48, 144, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(144, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(144, 144, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=144, bias=False)\n",
      "          (1): BatchNorm2d(144, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (2): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(144, 40, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(40, 144, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (3): Conv2dNormActivation(\n",
      "          (0): Conv2d(144, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(48, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (9): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(48, 288, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(288, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(288, 288, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=288, bias=False)\n",
      "          (1): BatchNorm2d(288, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (2): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(288, 72, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(72, 288, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (3): Conv2dNormActivation(\n",
      "          (0): Conv2d(288, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(96, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (10): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(96, 576, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(576, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(576, 576, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=576, bias=False)\n",
      "          (1): BatchNorm2d(576, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (2): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(576, 144, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(144, 576, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (3): Conv2dNormActivation(\n",
      "          (0): Conv2d(576, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(96, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (11): InvertedResidual(\n",
      "      (block): Sequential(\n",
      "        (0): Conv2dNormActivation(\n",
      "          (0): Conv2d(96, 576, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(576, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (1): Conv2dNormActivation(\n",
      "          (0): Conv2d(576, 576, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=576, bias=False)\n",
      "          (1): BatchNorm2d(576, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "          (2): Hardswish()\n",
      "        )\n",
      "        (2): SqueezeExcitation(\n",
      "          (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "          (fc1): Conv2d(576, 144, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (fc2): Conv2d(144, 576, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (activation): ReLU()\n",
      "          (scale_activation): Hardsigmoid()\n",
      "        )\n",
      "        (3): Conv2dNormActivation(\n",
      "          (0): Conv2d(576, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(96, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "    (12): Conv2dNormActivation(\n",
      "      (0): Conv2d(96, 576, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (1): BatchNorm2d(576, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)\n",
      "      (2): Hardswish()\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AdaptiveAvgPool2d(output_size=1)\n",
      "  (classifier): Sequential(\n",
      "    (0): Linear(in_features=576, out_features=1024, bias=True)\n",
      "    (1): Hardswish()\n",
      "    (2): Dropout(p=0.2, inplace=True)\n",
      "    (3): Linear(in_features=1024, out_features=1000, bias=True)\n",
      "  )\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "from torchvision.models import mobilenet_v3_small\n",
    "from torchvision.models import MobileNet_V3_Small_Weights\n",
    "\n",
    "model = mobilenet_v3_small(weights=MobileNet_V3_Small_Weights.DEFAULT)\n",
    "print(model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Task 4 (b): overloading the final layer\n",
    "\n",
    "Uh oh, we've hit a problem.\n",
    "\n",
    "- Look at the final linear layer.\n",
    "  - How many output classes are there?\n",
    "  - How many do we need?\n",
    "- We need to \"overload\" the final layer to produce the correct number of output features for our problem. Fortunately this is easy.\n",
    "- Uncomment the code below, choose the correct number of output features, and print the model again."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.nn import Linear\n",
    "\n",
    "model.classifier[3] = Linear(model.classifier[3].in_features, 10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 5: Set up the remaining PyTorch bits and bobs\n",
    "\n",
    "- We need to choose a loss function appropriate for classification.\n",
    "  - Can you remember what we chose previously?\n",
    "- We need an optimiser, too.\n",
    "  - Remember our friend, Adam?\n",
    "- Instantiate the model and loss function."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.nn import BCELoss\n",
    "from torch.optim import Adam\n",
    "\n",
    "loss_func = BCELoss()\n",
    "optimiser = Adam(model.parameters(), lr=1e-4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 6: Set the device\n",
    "\n",
    "We could have done this when we created the model, but it's fine to do it here.\n",
    "- If [CUDA](https://pytorch.org/docs/stable/cuda.html) (NVIDIA GPU) is available we will elect to run on that.\\n\",\n",
    "- Alternatively, if we are on Apple Silicon we can use the [MPS backend](https://developer.apple.com/metal/pytorch/) for GPU acceleration.\n",
    "- If neither of these are available we will resort to using the (slower) CPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "mps\n"
     ]
    }
   ],
   "source": [
    "from torch.cuda import is_available\n",
    "from torch.backends import mps\n",
    "\n",
    "if is_available():\n",
    "    DEVICE = \"cuda\"\n",
    "elif mps.is_available():\n",
    "    DEVICE = \"mps\"\n",
    "else:\n",
    "    DEVICE = \"cpu\"\n",
    "\n",
    "print(DEVICE)\n",
    "\n",
    "# Note: here the return/assignment to ``_`` is just to suppress the print.\n",
    "# The model is moved onto the correct device in-place.\n",
    "_ = model.to(DEVICE)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 7: Writing our training and validation loops\n",
    "\n",
    "As before, we need to write our training and validation loops.\n",
    "\n",
    "- Complete the training loop\n",
    "- Complete the validation loop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from typing import Dict\n",
    "\n",
    "from torch import no_grad\n",
    "from torch.nn import Module\n",
    "\n",
    "from numpy import mean\n",
    "\n",
    "\n",
    "def train_one_epoch(\n",
    "    model: Module,\n",
    "    train_loader: DataLoader,\n",
    "    optimiser: Adam,\n",
    "    loss_func: BCELoss,\n",
    ") -> Dict[str, float]:\n",
    "    \"\"\"Train the model for a single epoch.\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    model : Module\n",
    "        A neural network.\n",
    "    train_loader : DataLoader\n",
    "        The ``DataLoader`` for the training set.\n",
    "    optimiser : Adam\n",
    "        The optimiser to update the model's parameters with.\n",
    "    loss_func : BCELoss\n",
    "        Binary-cross-entropy loss function.\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    Dict[str, float]\n",
    "        A dictionary of metrics.\n",
    "\n",
    "    \"\"\"\n",
    "    model.train()\n",
    "    metrics: Dict[str, float] = {\"loss\": [], \"accuracy\": []}\n",
    "\n",
    "    for batch, targets in train_loader:\n",
    "        batch, targets = batch.to(DEVICE), targets.to(DEVICE)\n",
    "\n",
    "        optimiser.zero_grad()\n",
    "\n",
    "        preds = model(batch).softmax(dim=1)\n",
    "\n",
    "        loss = loss_func(preds, targets)\n",
    "\n",
    "        loss.backward()\n",
    "\n",
    "        optimiser.step()\n",
    "\n",
    "        metrics[\"loss\"].append(loss.item())\n",
    "        metrics[\"accuracy\"].append(batch_level_accuracy(preds, targets))\n",
    "\n",
    "    return {key: mean(val) for key, val in metrics.items()}\n",
    "\n",
    "\n",
    "@no_grad()\n",
    "def validate_one_epoch(\n",
    "    model: Module,\n",
    "    valid_loader: DataLoader,\n",
    "    loss_func: BCELoss,\n",
    ") -> Dict[str, float]:\n",
    "    \"\"\"Train the model for a single epoch.\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    model : Module\n",
    "        A neural network.\n",
    "    valid_loader : DataLoader\n",
    "        The ``DataLoader`` for the validation set.\n",
    "    loss_func : BCELoss\n",
    "        Binary-cross-entropy loss function.\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    Dict[str, float]\n",
    "        A dictionary of metrics.\n",
    "\n",
    "    \"\"\"\n",
    "    model.eval()\n",
    "    metrics: Dict[str, float] = {\"loss\": [], \"accuracy\": []}\n",
    "\n",
    "    for batch, targets in valid_loader:\n",
    "        batch, targets = batch.to(DEVICE), targets.to(DEVICE)\n",
    "\n",
    "        preds = model(batch).softmax(dim=1)\n",
    "\n",
    "        loss = loss_func(preds, targets)\n",
    "\n",
    "        metrics[\"loss\"].append(loss.item())\n",
    "        metrics[\"accuracy\"].append(batch_level_accuracy(preds, targets))\n",
    "\n",
    "    return {key: mean(val) for key, val in metrics.items()}\n",
    "\n",
    "\n",
    "@no_grad()\n",
    "def batch_level_accuracy(preds: Tensor, targets: Tensor):\n",
    "    \"\"\"Compute the batch-level accuracy.\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    preds : Tensor\n",
    "        The model's predictions.\n",
    "    targets : Tensor\n",
    "        The corresponding labels.\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    float\n",
    "        The batch-level accuracy.\n",
    "\n",
    "    Notes\n",
    "    -----\n",
    "    - This function assumes the ``preds`` have had the softmax applied to them\n",
    "      along dimension 1, and that the predicted class is therefore\n",
    "      ``preds.argmax(dim=1)``.\n",
    "\n",
    "    \"\"\"\n",
    "    return (preds.argmax(dim=1) == targets.argmax(dim=1)).float().mean().item()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 8: Training and extracting metrics\n",
    "\n",
    "- Now we can train our model for a specified number of epochs.\n",
    "  - During each epoch the model \"sees\" each training item once.\n",
    "- Append the training and validation metrics to a list.\n",
    "- Turn them into a ``pandas.DataFrame``\n",
    "  - Note: You can turn a ``List[Dict[str, float]]``, say ``my_list`` into a ``DataFrame`` with ``DataFrame(my_list)``."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0 time: 139.890 seconds.\n",
      "Epoch 1 time: 137.898 seconds.\n",
      "Epoch 2 time: 135.639 seconds.\n",
      "Epoch 3 time: 136.681 seconds.\n",
      "Epoch 4 time: 136.099 seconds.\n",
      "\n",
      "\n",
      "   loss_train  accuracy_train  loss_valid  accuracy_valid\n",
      "0    0.265784        0.370114    0.180383        0.603006\n",
      "1    0.180741        0.600504    0.122150        0.751095\n",
      "2    0.125428        0.745081    0.079941        0.847731\n",
      "3    0.091905        0.820904    0.057376        0.894009\n",
      "4    0.073212        0.860559    0.048015        0.910828\n"
     ]
    }
   ],
   "source": [
    "from time import perf_counter\n",
    "\n",
    "from pandas import DataFrame\n",
    "\n",
    "epochs = 5\n",
    "\n",
    "train_metrics, valid_metrics = [], []\n",
    "\n",
    "for epoch in range(epochs):\n",
    "    start_time = perf_counter()\n",
    "\n",
    "    train_metrics.append(train_one_epoch(model, train_loader, optimiser, loss_func))\n",
    "\n",
    "    valid_metrics.append(validate_one_epoch(model, valid_loader, loss_func))\n",
    "\n",
    "    stop_time = perf_counter()\n",
    "\n",
    "    print(f\"Epoch {epoch} time: {stop_time - start_time:.3f} seconds.\")\n",
    "\n",
    "print(\"\\n\")\n",
    "\n",
    "train_metrics = DataFrame(train_metrics)\n",
    "valid_metrics = DataFrame(valid_metrics)\n",
    "\n",
    "metrics = train_metrics.join(valid_metrics, lsuffix=\"_train\", rsuffix=\"_valid\")\n",
    "\n",
    "print(metrics)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 9: Plotting metrics\n",
    "\n",
    " - Let plots the training and validation metrics together"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x400 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "from numpy import linspace\n",
    "\n",
    "\n",
    "quantities = [\"loss\", \"accuracy\"]\n",
    "splits = [\"train\", \"valid\"]\n",
    "\n",
    "fig, axes = plt.subplots(1, 2, figsize=(8, 4))\n",
    "\n",
    "for axis, quant in zip(axes.ravel(), quantities):\n",
    "    for split in splits:\n",
    "        key = f\"{quant}_{split}\"\n",
    "        axis.plot(\n",
    "            linspace(1, epochs, epochs),\n",
    "            metrics[key],\n",
    "            \"-o\",\n",
    "            label=split.capitalize(),\n",
    "        )\n",
    "    axis.set_ylabel(quant.capitalize(), fontsize=15)\n",
    "\n",
    "for axis in axes.ravel():\n",
    "    axis.legend(fontsize=15)\n",
    "    axis.set_ylim(bottom=0.0, top=1.0)\n",
    "    axis.set_xlim(left=1, right=epochs)\n",
    "    axis.set_xlabel(\"Epoch\", fontsize=15)\n",
    "\n",
    "fig.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Task 10: Visualising some predictions\n",
    "\n",
    "Let's pick some random validation items, predict on them, and visualise the results."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([3, 28, 28])\n",
      "torch.Size([3, 28, 28])\n",
      "torch.Size([3, 28, 28])\n",
      "torch.Size([3, 28, 28])\n",
      "torch.Size([3, 28, 28])\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x200 with 5 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "valid_inds = [0, 666, 1024, 2048, 5555]\n",
    "valid_inds = [8, 16, 24, 32, 40]\n",
    "\n",
    "# Stick the model on the cpu\n",
    "_ = model.to(\"cpu\")\n",
    "_ = model.eval()\n",
    "\n",
    "figure, axes = plt.subplots(1, len(valid_inds), figsize=(2 * len(valid_inds), 2))\n",
    "\n",
    "for idx, axis in zip(valid_inds, axes.ravel()):\n",
    "    img_tensor, target = valid_set[idx]\n",
    "\n",
    "    with no_grad():\n",
    "        pred = model(img_tensor.unsqueeze(0)).softmax(dim=1).argmax(dim=1).item()\n",
    "\n",
    "    axis.imshow(img_tensor.permute(1, 2, 0))\n",
    "    axis.set_axis_off()\n",
    "    axis.set_title(f\"pred: {pred}, act: {target.argmax()}\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}