cnn.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Convolutional Neural Networks with Pytorch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torchvision\n",
    "import torchvision.transforms as transforms\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import torch.nn as nn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Image data structure"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Transformation for image data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "transform = transforms.Compose(\n",
    "    [transforms.ToTensor(),\n",
    "     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Downloading some images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Files already downloaded and verified\n",
      "Files already downloaded and verified\n"
     ]
    }
   ],
   "source": [
    "trainset = torchvision.datasets.CIFAR10(root='./data', train=True,\n",
    "                                        download=True, transform=transform)\n",
    "trainloader = torch.utils.data.DataLoader(trainset, batch_size=4,\n",
    "                                          shuffle=True, num_workers=2)\n",
    "\n",
    "testset = torchvision.datasets.CIFAR10(root='./data', train=False,\n",
    "                                       download=True, transform=transform)\n",
    "testloader = torch.utils.data.DataLoader(testset, batch_size=4,\n",
    "                                         shuffle=False, num_workers=2)\n",
    "\n",
    "classes = ('plane', 'car', 'bird', 'cat',\n",
    "           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Tensor representation of images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4 images, 3 channels RGB, 32x32 image\n",
      "torch.Size([4, 3, 32, 32]) torch.Size([4])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  car  frog truck horse\n"
     ]
    }
   ],
   "source": [
    "def imshow(img):\n",
    "    img = img / 2 + 0.5     # unnormalize\n",
    "    npimg = img.numpy()\n",
    "    plt.imshow(np.transpose(npimg, (1, 2, 0)))\n",
    "    plt.show()\n",
    "\n",
    "\n",
    "# get some random training images\n",
    "dataiter = iter(trainloader)\n",
    "images, labels = dataiter.next()\n",
    "print(\"4 images, 3 channels RGB, 32x32 image\")\n",
    "print(images.shape, labels.shape)\n",
    "\n",
    "# show images\n",
    "imshow(torchvision.utils.make_grid(images))\n",
    "# print labels\n",
    "print(' '.join('%5s' % classes[labels[j]] for j in range(4)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CNN & Pooling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Create random 3 channel 64x64 image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "image = torch.rand(1, 3, 64, 64)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Create convolutional layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "conv_layer = nn.Conv2d(in_channels=3, out_channels=32, kernel_size=3, stride=1, padding=0, padding_mode='zeros')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Applying the convolutional layer changes the shape of our image representation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "shape BEFORE conv layer torch.Size([1, 3, 64, 64])\n",
      "shape AFTER conv layer torch.Size([1, 32, 62, 62])\n"
     ]
    }
   ],
   "source": [
    "print(\"shape BEFORE conv layer\", image.shape)\n",
    "print(\"shape AFTER conv layer\", conv_layer(image).shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Max Pooling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "pool_layer = nn.MaxPool2d(kernel_size=2, stride=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "shape BEFORE pool layer torch.Size([1, 3, 64, 64])\n",
      "shape AFTER pool layer torch.Size([1, 3, 32, 32])\n"
     ]
    }
   ],
   "source": [
    "print(\"shape BEFORE pool layer\", image.shape)\n",
    "print(\"shape AFTER pool layer\", pool_layer(image).shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Similarly can use Avg Pool instead of Max"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}