handwritten_and_fashion_classfier.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "_-NQqZf4WW8o"
      },
      "outputs": [],
      "source": [
        "# Importing usefull libraries\n",
        "\n",
        "import torch\n",
        "import torch.nn.functional as F # For functional neural network\n",
        "import numpy as np # For numerical computations\n",
        "import matplotlib.pyplot as plt # For visualization\n",
        "\n",
        "# For loading and transforming the data\n",
        "\n",
        "from torchvision import datasets, transforms \n",
        "from torch import nn, optim"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#  A helper function for checking the prediction using graphs\n",
        "\n",
        "def view_classify(img, ps, version=\"MNIST\"):\n",
        "    ''' Function for viewing an image and it's predicted classes.\n",
        "    '''\n",
        "    ps = ps.data.numpy().squeeze()\n",
        "\n",
        "    fig, (ax1, ax2) = plt.subplots(figsize=(6,9), ncols=2)\n",
        "    ax1.imshow(img.resize_(1, 28, 28).numpy().squeeze())\n",
        "    ax1.axis('off')\n",
        "    ax2.barh(np.arange(10), ps)\n",
        "    ax2.set_aspect(0.1)\n",
        "    ax2.set_yticks(np.arange(10))\n",
        "    if version == \"MNIST\":\n",
        "        ax2.set_yticklabels(np.arange(10))\n",
        "    elif version == \"Fashion\":\n",
        "        ax2.set_yticklabels(['T-shirt/top',\n",
        "                            'Trouser',\n",
        "                            'Pullover',\n",
        "                            'Dress',\n",
        "                            'Coat',\n",
        "                            'Sandal',\n",
        "                            'Shirt',\n",
        "                            'Sneaker',\n",
        "                            'Bag',\n",
        "                            'Ankle Boot'], size='small');\n",
        "    ax2.set_title('Class Probability')\n",
        "    ax2.set_xlim(0, 1.1)\n",
        "\n",
        "    plt.tight_layout()"
      ],
      "metadata": {
        "id": "Q0-HYGMLy5RD"
      },
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Define a transform to normalize the hand written digits data\n",
        "\n",
        "transform = transforms.Compose([transforms.ToTensor(),\n",
        "                                transforms.Normalize((0.5,),(0.5,)),\n",
        "                                ])\n",
        "\n",
        "# Download and load the training data\n",
        "\n",
        "trainset = datasets.MNIST('MNIST_data/',\n",
        "                          download=True,\n",
        "                          train=True,\n",
        "                          transform = transform\n",
        "                          )\n",
        "trainloader = torch.utils.data.DataLoader(trainset,\n",
        "                                          batch_size = 64,\n",
        "                                          shuffle=True)\n"
      ],
      "metadata": {
        "id": "qjDeVhOhfOAB"
      },
      "execution_count": 28,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Download and load the training data for the fashion data\n",
        "\n",
        "trainset = datasets.FashionMNIST('Fasion_data/',\n",
        "                                 download=True,\n",
        "                                 train=True,\n",
        "                                 transform = transform\n",
        "                                 )\n",
        "trainloader = torch.utils.data.DataLoader(trainset,\n",
        "                                          batch_size = 64, \n",
        "                                          shuffle=True\n",
        "                                          )\n",
        "\n",
        "# Download and load the test data for the fashion data\n",
        "\n",
        "testset= datasets.FashionMNIST('Fasion_data/',\n",
        "                               download=True,\n",
        "                               train=False,\n",
        "                               transform = transform\n",
        "                               )\n",
        "testloader= torch.utils.data.DataLoader(testset,\n",
        "                                        batch_size = 64, \n",
        "                                        shuffle=True\n",
        "                                        )"
      ],
      "metadata": {
        "id": "X5ja1FoQ2aLU"
      },
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Lets check the size and type of the data we downloaded\n",
        "\n",
        "images,labels = next(iter(trainloader))\n",
        "print(type(images))\n",
        "print(images.shape)\n",
        "print(labels.shape)\n",
        "print(labels)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "tdkPgY7pgNdX",
        "outputId": "dc25137c-efba-4b5f-81ed-8215ad2a7734"
      },
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "<class 'torch.Tensor'>\n",
            "torch.Size([64, 1, 28, 28])\n",
            "torch.Size([64])\n",
            "tensor([2, 8, 7, 6, 1, 9, 7, 6, 0, 2, 8, 4, 5, 4, 7, 7, 5, 0, 5, 2, 4, 4, 0, 6,\n",
            "        9, 7, 8, 1, 6, 2, 3, 3, 3, 7, 0, 6, 4, 8, 3, 6, 1, 6, 0, 7, 5, 7, 8, 8,\n",
            "        6, 3, 1, 1, 6, 1, 5, 7, 1, 5, 7, 0, 0, 2, 1, 2])\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Lets check out a single image and its label\n",
        "\n",
        "plt.imshow(images[1].numpy().squeeze(),cmap='Greys_r')\n",
        "print(f'The label is {labels[1].numpy()}')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "4UpVowhyjK0I",
        "outputId": "5e2bebcf-67d0-4070-d76c-7c5979c95e30"
      },
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "The label is 8\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Training a model for classifiying hand written digits \n",
        "\n",
        "# Lets train a network of 2 hidden layers with each being Relu activated\n",
        "\n",
        "hand_written_model = nn.Sequential(nn.Linear(784,128),\n",
        "                      nn.ReLU(),\n",
        "                      nn.Linear(128,64),\n",
        "                      nn.ReLU(),\n",
        "                      nn.Linear(64,10),\n",
        "                      nn.LogSoftmax(dim=1))\n",
        "\n",
        "criterion = nn.NLLLoss() # Negative Log Likelihood loss\n",
        "\n",
        "# Lets use an optimizer with a Stochastic Gradient Decent of learning rate 0.003\n",
        "\n",
        "optimizer = optim.SGD(hand_written_model.parameters(),lr=0.003) \n",
        "\n",
        "epochs = 5\n",
        "\n",
        "for epoch in range(epochs):\n",
        "\n",
        "  # Initializing the running loss for later averaging\n",
        "  running_loss = 0.0\n",
        "\n",
        "  for images,labels in trainloader:\n",
        "\n",
        "    # Lets flatten the image to fit into the network layer\n",
        "    # .view will change our [64,1,28,28] to [64,784] i.e 28*28*1=784\n",
        "    \n",
        "    images = images.view(images.shape[0],-1)\n",
        "\n",
        "    optimizer.zero_grad()\n",
        "\n",
        "    output = hand_written_model.forward(images)\n",
        "\n",
        "    loss = criterion(output,labels)\n",
        "\n",
        "    loss.backward()\n",
        "\n",
        "    running_loss += loss.item()\n",
        "\n",
        "    optimizer.step()\n",
        "\n",
        "  print(f'Training loss: {running_loss/len(trainloader)}')\n"
      ],
      "metadata": {
        "id": "SZwVKXgOXYO_",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "40cb665b-b02d-43e8-8fe6-e2502d12a40b"
      },
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Training loss: 1.8479479027709473\n",
            "Training loss: 0.8185606790122701\n",
            "Training loss: 0.5224122261441847\n",
            "Training loss: 0.4295903445878771\n",
            "Training loss: 0.3849476414925254\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Testing the hand written classifier model\n",
        "\n",
        "images,labels = next(iter(trainloader))\n",
        "\n",
        "img = images[0].view(1,-1)\n",
        "\n",
        "with torch.no_grad():\n",
        "  logits = hand_written_model.forward(img)\n",
        "\n",
        "# Changing the probabilities from our nn output into positive number and \n",
        "# finding the displaying the result using our helper function\n",
        "\n",
        "ps = torch.exp(logits)\n",
        "\n",
        "view_classify(img,ps,version='MNIST')"
      ],
      "metadata": {
        "id": "a77zrW_CYuhn",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 253
        },
        "outputId": "7c702b75-c5ff-45e4-8eb9-d829f7055100"
      },
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x648 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Training a model for classifiying fasion-MNIST data set\n",
        "\n",
        "''' \n",
        "\n",
        "Here we have 3 hidden layers each having a drop out of 20% \n",
        " We added drop out for the purpose of decreasing overfitting\n",
        " We will eliminate all the dropouts for evaluation and get the back while \n",
        "training \n",
        "\n",
        "'''\n",
        "model = nn.Sequential(nn.Linear(784,256),\n",
        "                      nn.ReLU(),\n",
        "                      nn.Dropout(0.2),\n",
        "                      nn.Linear(256,128),\n",
        "                      nn.ReLU(),\n",
        "                      nn.Dropout(0.2),\n",
        "                      nn.Linear(128,64),\n",
        "                      nn.ReLU(),\n",
        "                      nn.Dropout(0.2),\n",
        "                      nn.Linear(64,10),\n",
        "                      nn.LogSoftmax(dim=1))\n",
        "\n",
        "criterion = nn.NLLLoss(reduction='sum')\n",
        "\n",
        "optimizer = optim.Adam(model.parameters(),lr=0.003)\n",
        "\n",
        "epochs = 30\n",
        "\n",
        "train_losses = []\n",
        "test_losses = []\n",
        "\n",
        "for epoch in range(epochs):\n",
        "\n",
        "  total_training_loss = 0.0\n",
        "\n",
        "  for images,labels in trainloader:\n",
        "\n",
        "    images = images.view(images.shape[0],-1)\n",
        "\n",
        "    optimizer.zero_grad()\n",
        "\n",
        "    output = model.forward(images)\n",
        "\n",
        "    loss = criterion(output,labels)\n",
        "\n",
        "    total_training_loss += loss.item()\n",
        "\n",
        "    loss.backward()\n",
        "\n",
        "    optimizer.step()\n",
        "\n",
        "   \n",
        "    \n",
        "  total_test_loss = 0.0\n",
        "  total_correct = 0.0\n",
        "\n",
        "  with torch.no_grad():\n",
        "\n",
        "    # Here we used model.eval() to remove the dropouts for testing our model\n",
        "    \n",
        "    model.eval()\n",
        "    for images,labels in testloader:\n",
        "\n",
        "      images = images.view(images.shape[0],-1)\n",
        "\n",
        "      logps = model(images)\n",
        "      test_loss = criterion(logps,labels)\n",
        "\n",
        "      total_test_loss += test_loss.item()\n",
        "\n",
        "      ps_fashion = torch.exp(logps)\n",
        "\n",
        "      top_p,top_class = ps_fashion.topk(1,dim=1)\n",
        "\n",
        "      equals = top_class == labels.view(*top_class.shape)\n",
        "    \n",
        "      total_correct += equals.sum().item()\n",
        "  \n",
        "  # Here we get the dropouts back to the model by using model.train()\n",
        "  \n",
        "  model.train()\n",
        "  train_loss = total_training_loss / len(trainloader.dataset)\n",
        "  test_loss = total_test_loss / len(testloader.dataset)\n",
        "  accuracy = total_correct / len(testloader.dataset)\n",
        "\n",
        "  train_losses.append(train_loss) \n",
        "  test_losses.append(test_loss)\n",
        "\n",
        "  print('==========================')\n",
        "  print(f'Epoch {epoch}')\n",
        "  print(f'Training loss: {train_loss}')\n",
        "  print(f'Test loss: {test_loss}')\n",
        "  print(f'Test accuracy: {accuracy*100}')\n",
        "  print('==========================')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "spr4XJkizyhA",
        "outputId": "1383d670-24d5-4935-9754-9defd947365f"
      },
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "==========================\n",
            "Epoch 0\n",
            "Training loss: 0.6069811887741089\n",
            "Test loss: 0.49761438035964967\n",
            "Test accuracy: 81.91000000000001\n",
            "==========================\n",
            "==========================\n",
            "Epoch 1\n",
            "Training loss: 0.4825502340157827\n",
            "Test loss: 0.4280258972167969\n",
            "Test accuracy: 84.92\n",
            "==========================\n",
            "==========================\n",
            "Epoch 2\n",
            "Training loss: 0.4501264285723368\n",
            "Test loss: 0.4220623287677765\n",
            "Test accuracy: 84.8\n",
            "==========================\n",
            "==========================\n",
            "Epoch 3\n",
            "Training loss: 0.43103395976225534\n",
            "Test loss: 0.42362824950218203\n",
            "Test accuracy: 85.09\n",
            "==========================\n",
            "==========================\n",
            "Epoch 4\n",
            "Training loss: 0.41853837067286176\n",
            "Test loss: 0.40511627292633057\n",
            "Test accuracy: 85.81\n",
            "==========================\n",
            "==========================\n",
            "Epoch 5\n",
            "Training loss: 0.4110726539770762\n",
            "Test loss: 0.38959638419151305\n",
            "Test accuracy: 86.21\n",
            "==========================\n",
            "==========================\n",
            "Epoch 6\n",
            "Training loss: 0.40291477995713554\n",
            "Test loss: 0.377946484041214\n",
            "Test accuracy: 86.50999999999999\n",
            "==========================\n",
            "==========================\n",
            "Epoch 7\n",
            "Training loss: 0.40233449681599937\n",
            "Test loss: 0.39297593989372254\n",
            "Test accuracy: 85.68\n",
            "==========================\n",
            "==========================\n",
            "Epoch 8\n",
            "Training loss: 0.39797556784947713\n",
            "Test loss: 0.3850397620201111\n",
            "Test accuracy: 86.06\n",
            "==========================\n",
            "==========================\n",
            "Epoch 9\n",
            "Training loss: 0.38909111011823017\n",
            "Test loss: 0.38659927418231965\n",
            "Test accuracy: 86.72999999999999\n",
            "==========================\n",
            "==========================\n",
            "Epoch 10\n",
            "Training loss: 0.3867277200539907\n",
            "Test loss: 0.37633272562026976\n",
            "Test accuracy: 87.29\n",
            "==========================\n",
            "==========================\n",
            "Epoch 11\n",
            "Training loss: 0.38025333762168884\n",
            "Test loss: 0.38063295249938967\n",
            "Test accuracy: 87.09\n",
            "==========================\n",
            "==========================\n",
            "Epoch 12\n",
            "Training loss: 0.3789516554514567\n",
            "Test loss: 0.3901952775001526\n",
            "Test accuracy: 86.61\n",
            "==========================\n",
            "==========================\n",
            "Epoch 13\n",
            "Training loss: 0.37504196561972303\n",
            "Test loss: 0.38179001512527466\n",
            "Test accuracy: 86.79\n",
            "==========================\n",
            "==========================\n",
            "Epoch 14\n",
            "Training loss: 0.3669297106663386\n",
            "Test loss: 0.37999982233047486\n",
            "Test accuracy: 87.03\n",
            "==========================\n",
            "==========================\n",
            "Epoch 15\n",
            "Training loss: 0.36433598573207854\n",
            "Test loss: 0.3917933277130127\n",
            "Test accuracy: 86.15\n",
            "==========================\n",
            "==========================\n",
            "Epoch 16\n",
            "Training loss: 0.36612445386250814\n",
            "Test loss: 0.3829483426094055\n",
            "Test accuracy: 86.83999999999999\n",
            "==========================\n",
            "==========================\n",
            "Epoch 17\n",
            "Training loss: 0.36557533179124196\n",
            "Test loss: 0.38245176162719724\n",
            "Test accuracy: 87.45\n",
            "==========================\n",
            "==========================\n",
            "Epoch 18\n",
            "Training loss: 0.364352214884758\n",
            "Test loss: 0.3916970725059509\n",
            "Test accuracy: 86.86\n",
            "==========================\n",
            "==========================\n",
            "Epoch 19\n",
            "Training loss: 0.3670731128136317\n",
            "Test loss: 0.3987360114455223\n",
            "Test accuracy: 86.48\n",
            "==========================\n",
            "==========================\n",
            "Epoch 20\n",
            "Training loss: 0.3575667373816172\n",
            "Test loss: 0.37595637629032136\n",
            "Test accuracy: 87.0\n",
            "==========================\n",
            "==========================\n",
            "Epoch 21\n",
            "Training loss: 0.35411430598100024\n",
            "Test loss: 0.37948103680610656\n",
            "Test accuracy: 87.48\n",
            "==========================\n",
            "==========================\n",
            "Epoch 22\n",
            "Training loss: 0.3508589632511139\n",
            "Test loss: 0.38479875135421754\n",
            "Test accuracy: 86.75\n",
            "==========================\n",
            "==========================\n",
            "Epoch 23\n",
            "Training loss: 0.35145374643007915\n",
            "Test loss: 0.37929495244026185\n",
            "Test accuracy: 86.95\n",
            "==========================\n",
            "==========================\n",
            "Epoch 24\n",
            "Training loss: 0.3522523019711177\n",
            "Test loss: 0.3746956869363785\n",
            "Test accuracy: 87.74\n",
            "==========================\n",
            "==========================\n",
            "Epoch 25\n",
            "Training loss: 0.3535412485122681\n",
            "Test loss: 0.3720868009567261\n",
            "Test accuracy: 87.22\n",
            "==========================\n",
            "==========================\n",
            "Epoch 26\n",
            "Training loss: 0.35165796043078107\n",
            "Test loss: 0.39614115834236147\n",
            "Test accuracy: 86.92\n",
            "==========================\n",
            "==========================\n",
            "Epoch 27\n",
            "Training loss: 0.3442408896764119\n",
            "Test loss: 0.37957577171325685\n",
            "Test accuracy: 87.48\n",
            "==========================\n",
            "==========================\n",
            "Epoch 28\n",
            "Training loss: 0.34338740928173067\n",
            "Test loss: 0.37214519572258\n",
            "Test accuracy: 87.69\n",
            "==========================\n",
            "==========================\n",
            "Epoch 29\n",
            "Training loss: 0.3405271060625712\n",
            "Test loss: 0.3877249622821808\n",
            "Test accuracy: 87.31\n",
            "==========================\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Plot the graph of training and validation loss to see the overfitting effect\n",
        "\n",
        "plt.plot(train_losses,label='Training losses')\n",
        "plt.plot(test_losses,label='Validation losses')\n",
        "plt.legend(frameon=False)\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        },
        "id": "DtVvhIxiJSnL",
        "outputId": "9c687148-3127-4ae6-8f89-a0725850bd72"
      },
      "execution_count": 41,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Testing the fasion classifier model\n",
        "\n",
        "fashion_images,labels = next(iter(testloader))\n",
        "\n",
        "img = fashion_images[0].view(1,-1)\n",
        "\n",
        "with torch.no_grad():\n",
        "  logits = model.forward(img)\n",
        "\n",
        "ps = F.softmax(logits,dim=1)\n",
        "\n",
        "view_classify(img,ps,version='Fashion')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 234
        },
        "id": "mpQx56Ij1AC1",
        "outputId": "79d7ce3e-b384-4229-9795-787a04f1cc76"
      },
      "execution_count": 42,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x648 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    }
  ]
}