EMNIST_Classification (2).ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "EMNIST-Classification.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "k1f8C6IivG3z",
        "V65UGX3ewGPR",
        "4sBskJLVxQ5I",
        "r0O0-MKAxyY3",
        "LYiumYJs3NAk"
      ]
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "a6dae65ba5e548f498f9457c12325636": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_3bc057e08fbf45c3ae0a7eeec559b4dc",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_e21d2a1442f64967bb843d0df7d61137",
              "IPY_MODEL_a40166bb96564a629b7b6c21c1f2c99a"
            ]
          }
        },
        "3bc057e08fbf45c3ae0a7eeec559b4dc": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "e21d2a1442f64967bb843d0df7d61137": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_77b805033d7e43c7b522c79e42cadbce",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "info",
            "max": 1,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 1,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_b4813452c204414e99f5e986d02684e3"
          }
        },
        "a40166bb96564a629b7b6c21c1f2c99a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_8e6d1b6623d845788a4e093b2eeff752",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 561758208/? [02:00&lt;00:00, 3652755.79it/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_e41554e79023416bb95b92ea78fb3e72"
          }
        },
        "77b805033d7e43c7b522c79e42cadbce": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "b4813452c204414e99f5e986d02684e3": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "8e6d1b6623d845788a4e093b2eeff752": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "e41554e79023416bb95b92ea78fb3e72": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        }
      }
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qKJ49Pdgptnm",
        "colab_type": "text"
      },
      "source": [
        "# Using Convolutional Neural Network to classify the EMNIST Dataset\n",
        "\n",
        "PyTorch provides a wide range of built-in datasets to work on like, MNIST, EMNIST, Fashion MNIST, CIFAR10 etc. In this notebook, we are using the EMNIST dataset which is split into 6 sub-datasets. We will be using the 'letters' sub-dataset, which consists of 26 balanced classes, each having 28x28 pixel images of hand-written letters from the english alphabet. The objective of the model is Hand-written Alphabet Recognition.\n",
        "\n",
        "We'll be training the model on 124800 images from the training dataset, out of which 1% of the images will be used for the validation set, and the model will be tested on 20800 images from the test dataset.\n",
        "\n",
        "In the model, we will use CNNs to build our model."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XNoQa1JOrAh0",
        "colab_type": "text"
      },
      "source": [
        "#### Importing libraries and the EMNIST dataset "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7X7zr1jQornk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import torch\n",
        "import torchvision\n",
        "from torchvision.datasets import EMNIST"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cd2A99Nfrlxt",
        "colab_type": "text"
      },
      "source": [
        "#### Downloading the EMNIST dataset to the notebook\n",
        "The root directory where we want the dataset to be downloaded, the sub-dataset and the download permissions are to be passed as arguments to get the in-built dataset to work on."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FyRiDRYNpDG2",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 256,
          "referenced_widgets": [
            "a6dae65ba5e548f498f9457c12325636",
            "3bc057e08fbf45c3ae0a7eeec559b4dc",
            "e21d2a1442f64967bb843d0df7d61137",
            "a40166bb96564a629b7b6c21c1f2c99a",
            "77b805033d7e43c7b522c79e42cadbce",
            "b4813452c204414e99f5e986d02684e3",
            "8e6d1b6623d845788a4e093b2eeff752",
            "e41554e79023416bb95b92ea78fb3e72"
          ]
        },
        "outputId": "ee1f637a-9852-4deb-b90b-417caa7abf5e"
      },
      "source": [
        "dataset = EMNIST(root='data/', split='letters', download=True)"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading and extracting zip archive\n",
            "Downloading http://www.itl.nist.gov/iaui/vip/cs_links/EMNIST/gzip.zip to data/EMNIST/raw/emnist.zip\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "a6dae65ba5e548f498f9457c12325636",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Extracting data/EMNIST/raw/emnist.zip to data/EMNIST/raw\n",
            "Processing byclass\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/pytorch/torch/csrc/utils/tensor_numpy.cpp:141: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Processing bymerge\n",
            "Processing balanced\n",
            "Processing letters\n",
            "Processing digits\n",
            "Processing mnist\n",
            "Done!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "O2jVHzjatgKw",
        "colab_type": "text"
      },
      "source": [
        "#### Exploring the Dataset\n",
        "Now, to get familiar with the dataset, we'll just look at the length of the dataset, some of the nodes of the dataset and print their image and label for some visual familiarity, and see the classes the dataset is divided into."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0b9OHhg-pab0",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "afa89e2e-7e61-46f1-b760-86da322fbbaa"
      },
      "source": [
        "len(dataset)"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "124800"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 3
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "v83_HRy78jFz",
        "colab_type": "text"
      },
      "source": [
        "The training dataset contains 124800 data points."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XMY4sCUgqHsM",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "613b275a-49b1-494b-e18f-64bc3662a66f"
      },
      "source": [
        "dataset[0]"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(<PIL.Image.Image image mode=L size=28x28 at 0x7EFC1598DC88>, 23)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ugs7VGYLqIHn",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "9b232746-f7eb-400a-cebb-91d4e8cb84b7"
      },
      "source": [
        "dataset[1]"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(<PIL.Image.Image image mode=L size=28x28 at 0x7EFC14C220F0>, 7)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aVhhAvZH82m_",
        "colab_type": "text"
      },
      "source": [
        "The first two datapoints of the training dataset are the letters 'w' (23rd alphabet) and 'g' (7th alphabet)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5gcN8YI9sfLf",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dhCvRxuooGkq",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 459
        },
        "outputId": "2c82f6e1-7b7d-4bc1-e5f9-3fa5a693659f"
      },
      "source": [
        "classes = dataset.classes\n",
        "classes"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['a',\n",
              " 'b',\n",
              " 'c',\n",
              " 'd',\n",
              " 'e',\n",
              " 'f',\n",
              " 'g',\n",
              " 'h',\n",
              " 'i',\n",
              " 'j',\n",
              " 'k',\n",
              " 'l',\n",
              " 'm',\n",
              " 'n',\n",
              " 'o',\n",
              " 'p',\n",
              " 'q',\n",
              " 'r',\n",
              " 's',\n",
              " 't',\n",
              " 'u',\n",
              " 'v',\n",
              " 'w',\n",
              " 'x',\n",
              " 'y',\n",
              " 'z']"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fFiPIp_boUym",
        "colab_type": "text"
      },
      "source": [
        "The 26 classes are all the letters of the english alphabet."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sdADeXWVqZ8i",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "ce0a2953-0684-4142-e32d-9aee7bf39190"
      },
      "source": [
        "image, label = dataset[2]\n",
        "label -= 1\n",
        "plt.imshow(image, cmap='gray')\n",
        "print(classes[label])"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "p\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sP1AFQTiqaYQ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "d59eb964-52e9-4e07-9ade-e78077e98384"
      },
      "source": [
        "image, label = dataset[3]\n",
        "label -= 1\n",
        "plt.imshow(image, cmap='gray')\n",
        "print(classes[label])"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "o\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yfrEC3OxuCXp",
        "colab_type": "text"
      },
      "source": [
        "#### Transforming the images in the dataset to tensors \n",
        "To work with data points and perform operations on them, we need to convert them to tensors."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4O43ev7Utb7C",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import torchvision.transforms as transforms"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YZS2M8g3uA_0",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "b375e886-bcff-4575-85c7-985d561cf94f"
      },
      "source": [
        "dataset = EMNIST(root='data/', train=True, transform=transforms.ToTensor(), split='letters')\n",
        "len(dataset)"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "124800"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 11
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "j1YShb6zuYna",
        "colab_type": "text"
      },
      "source": [
        "#### Exploring the image tensors\n",
        "Each data point has two parts: the image tensor and it's label.\n",
        "The shape of the image tensor is [1, 28, 28], signifying that it only consists of one channel(lightness channel, L, since it's a grayscale image), conveyed by the first dimension; and the size of the image that 28x28 pixels, as conveyed by the subsequent 2 dimensions "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "asptGdV7qskZ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "26ca1225-a9de-49fa-a7c4-aea751851745"
      },
      "source": [
        "img_tensor, labels = dataset[0]\n",
        "print(img_tensor.shape, labels-1)"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "torch.Size([1, 28, 28]) 22\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3HH5bh8Aqs72",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "67c5523c-faa6-4095-b5f4-f9ad5284c6ee"
      },
      "source": [
        "img_tensor, labels = dataset[1]\n",
        "print(img_tensor.shape, labels-1)"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "torch.Size([1, 28, 28]) 6\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AbOKguJEq2Dn",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "5ff95311-d9b2-4e07-9fe9-ed846ccb4e97"
      },
      "source": [
        "img_tensor, labels = dataset[2]\n",
        "print(img_tensor, labels-1)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tensor([[[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0078, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.1255, 0.6588, 0.3529,\n",
            "          0.0392, 0.0000, 0.0000, 0.0078, 0.0000, 0.0000, 0.0039, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0157, 0.0157, 0.0196, 0.4235, 0.9922, 0.9686,\n",
            "          0.8157, 0.5059, 0.5490, 0.7137, 0.5451, 0.5059, 0.6235, 0.4980,\n",
            "          0.4471, 0.1294, 0.0196, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0824, 0.1961, 0.4902, 0.4980, 0.5059, 0.8275, 0.9961, 0.9961,\n",
            "          0.9961, 0.9804, 0.9804, 0.9922, 0.9804, 0.9804, 0.9843, 0.9804,\n",
            "          0.9608, 0.7255, 0.1255, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0157, 0.1804, 0.4902, 0.5490,\n",
            "          0.9098, 0.9804, 0.9961, 0.9961, 0.9961, 0.9961, 0.9961, 0.9961,\n",
            "          0.9961, 0.9961, 0.9961, 0.9961, 0.9961, 0.9961, 1.0000, 0.9961,\n",
            "          0.9961, 0.8510, 0.1451, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0314, 0.4510, 0.8000, 0.9608, 0.9922, 0.9608,\n",
            "          0.6745, 0.4980, 0.4980, 0.4980, 0.4980, 0.5059, 0.9255, 0.9961,\n",
            "          0.9255, 0.5059, 0.4980, 0.5059, 0.7961, 0.8510, 0.8510, 0.8510,\n",
            "          0.7216, 0.4706, 0.0706, 0.0000],\n",
            "         [0.0000, 0.0784, 0.6196, 0.9922, 0.9922, 0.6941, 0.4471, 0.1804,\n",
            "          0.0118, 0.0000, 0.0000, 0.0000, 0.0000, 0.0157, 0.8510, 0.9961,\n",
            "          0.8510, 0.0157, 0.0000, 0.0000, 0.0157, 0.0157, 0.0157, 0.0157,\n",
            "          0.0118, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.1451, 0.8431, 0.9961, 0.8667, 0.1412, 0.0157, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0353, 0.8706, 0.9961,\n",
            "          0.8431, 0.0157, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.1255, 0.7961, 0.9961, 0.5765, 0.0196, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.1961, 0.9804, 0.9647,\n",
            "          0.4980, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0157, 0.4510, 0.9961, 0.9804, 0.3922, 0.0824, 0.0118,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0314, 0.6863, 0.9961, 0.8000,\n",
            "          0.1333, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0863, 0.7922, 0.9765, 0.9882, 0.9098, 0.6706,\n",
            "          0.2000, 0.1451, 0.1451, 0.1608, 0.5176, 0.9922, 0.9647, 0.3098,\n",
            "          0.0118, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0039, 0.1412, 0.7647, 0.9804, 0.9961, 0.9843,\n",
            "          0.8706, 0.8510, 0.8510, 0.8510, 0.9255, 0.9961, 0.6863, 0.0314,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0314, 0.1961, 0.6706, 0.9098,\n",
            "          0.9804, 0.9804, 0.9804, 0.9804, 0.9608, 0.6235, 0.0784, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0118, 0.0824,\n",
            "          0.1451, 0.1451, 0.1451, 0.1451, 0.1255, 0.0118, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000],\n",
            "         [0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,\n",
            "          0.0000, 0.0000, 0.0000, 0.0000]]]) 15\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XpURZB2muwvT",
        "colab_type": "text"
      },
      "source": [
        "#### Why do we need a validation set?\n",
        "We need a validation set to check our model before finally testing it on the test dataset. It helps us to change hyperparameters accordingly to get the best results, and then pick the best version of the model before reporting the final results.\n",
        "\n",
        "#### Create validation set from the training set\n",
        "We take the 1% of the training set as our validation set in this notebook, though any small percentage can be taken. The validation set is created by randomly splitting the training dataset into two parts of suitable size. The random split plays a major role here as, if we split it in order, the training dataset might not have some classes at all, and since the model was never trained for those classes which got divided into the validation set, the model will never be able to guess them. Which means, even after training our model, our model's accuracy on the validation set will always be zero."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EdO1T0osupKG",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "7c03c35c-8e5d-4965-9ad2-ffcfa08e0ddc"
      },
      "source": [
        "from torch.utils.data import random_split\n",
        "\n",
        "val_per = 0.1\n",
        "val_len = int(val_per*len(dataset))\n",
        "train_ds, val_ds = random_split(dataset, [len(dataset)-val_len, val_len])\n",
        "print(len(train_ds), len(val_ds))"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "112320 12480\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "k1f8C6IivG3z",
        "colab_type": "text"
      },
      "source": [
        "#### Load the training and validation data in batches\n",
        "We use the DataLoader to load the training and validation dataset with a batch size of 128 data points. We shuffle the training dataset because the most randomized inputs produce the most general model, which is favourable. However, we do not shuffle the validation dataset as, we cannot compare different evaluations if the dataset on which we are evaluating is not the same."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YkYLpztUu-sX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from torch.utils.data import DataLoader\n",
        "\n",
        "batch_size = 128\n",
        "\n",
        "train_load = DataLoader(train_ds, batch_size, shuffle=True, num_workers=4, pin_memory=True)\n",
        "val_load = DataLoader(val_ds, batch_size, num_workers=4, pin_memory=True)"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PlMXMNyuTdH-",
        "colab_type": "text"
      },
      "source": [
        "#### Using a GPU\n",
        "If a GPU(cuda) is available, get_default_device will return it, else cpu. The function to_device moves the data to the specified device in the argument. The class DeviceDataLoader has functions to load the data to the device."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lin4u5j0TqbS",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def get_default_device():\n",
        "    \"\"\"Pick GPU if available, else CPU\"\"\"\n",
        "    if torch.cuda.is_available():\n",
        "        return torch.device('cuda')\n",
        "    else:\n",
        "        return torch.device('cpu')\n",
        "    \n",
        "def to_device(data, device):\n",
        "    \"\"\"Move tensor(s) to chosen device\"\"\"\n",
        "    if isinstance(data, (list,tuple)):\n",
        "        return [to_device(x, device) for x in data]\n",
        "    return data.to(device, non_blocking=True)\n",
        "\n",
        "class DeviceDataLoader():\n",
        "    \"\"\"Wrap a dataloader to move data to a device\"\"\"\n",
        "    def __init__(self, dl, device):\n",
        "        self.dl = dl\n",
        "        self.device = device\n",
        "        \n",
        "    def __iter__(self):\n",
        "        \"\"\"Yield a batch of data after moving it to device\"\"\"\n",
        "        for b in self.dl: \n",
        "            yield to_device(b, self.device)\n",
        "\n",
        "    def __len__(self):\n",
        "        \"\"\"Number of batches\"\"\"\n",
        "        return len(self.dl)"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RKQIPqF6U_6f",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "247506ee-0508-4390-d46c-1c9837a613b1"
      },
      "source": [
        "device = get_default_device()\n",
        "device"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "device(type='cuda')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Al6E1TboVFW_",
        "colab_type": "text"
      },
      "source": [
        "The default device is cuda, since the Accelerator is set to GPU."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VUqcx7lURmFI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "train_load = DeviceDataLoader(train_load, device)\n",
        "val_load = DeviceDataLoader(val_load, device)"
      ],
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ooZeazNbvgm0",
        "colab_type": "text"
      },
      "source": [
        "#### Creating a convolutional neural network"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MdcDcl61Zjag",
        "colab_type": "text"
      },
      "source": [
        "Since the input and output sizes should be scalar integers for us to implement regression on them, input size can't be in the form of [28, 28], therefore, it is 28*28 = 784 (number of pixels in the image if we flatten the image). And output size is clearly 26, because the image will be classified into one of the 26 classes."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hmiMQaokvZH7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from torch import nn\n",
        "\n",
        "input_size = 28*28\n",
        "output_size = 26"
      ],
      "execution_count": 20,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "X3rfJjGua40P",
        "colab_type": "text"
      },
      "source": [
        "We extend the nn.Module class to define our EmnistModel. It is a Convolutional Neural Network. We use nn.Sequential to chain all the layers and activation functions into one network. The network has 7 layers. The first layer, increases the number of channels from 1 to 32, taking kernel_size as 3 and padding of 1. The activation function, relu, is applied. The second layer increases channels to 64, activation function is applied again. Then max-pooling is done, in which the height and width of the image is halved. Similar operations are repeated again. The resulting tensor of 256x7x7 is flattened, and it goes through nn.Linear and activation functions, till we get the required number of outputs. \n",
        "\n",
        "Then we have the training step where we use the cross_entropy loss function on the predictions and targets to calculate the loss for the training dataset. In the validation step, we also calculate the accuracy of the model along with the loss on the validation set.\n",
        "\n",
        "In the last two functions, we combine the validation losses and accuracies after each epoch, so that we can track the progress of the training."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7T02H5h_vpKO",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class EmnistModel(nn.Module):\n",
        "    def __init__(self):\n",
        "        super().__init__()\n",
        "        self.network = nn.Sequential(\n",
        "            nn.Conv2d(1, 32, kernel_size = 3, padding = 1),\n",
        "            nn.ReLU(),\n",
        "            nn.Conv2d(32, 64, kernel_size = 3, padding = 1),\n",
        "            nn.ReLU(),\n",
        "            nn.MaxPool2d(2, 2), #64*14*14\n",
        "\n",
        "            nn.Conv2d(64, 128, kernel_size = 3, padding = 1),\n",
        "            nn.ReLU(),\n",
        "            nn.Conv2d(128, 256, kernel_size = 3, padding = 1),\n",
        "            nn.ReLU(),\n",
        "            nn.MaxPool2d(2, 2), #256*7*7\n",
        "\n",
        "            nn.Flatten(),\n",
        "            nn.Linear(256*7*7, 1024),\n",
        "            nn.ReLU(),\n",
        "            nn.Linear(1024, 256),\n",
        "            nn.ReLU(),\n",
        "            nn.Linear(256, 26)\n",
        "        )\n",
        "        \n",
        "    def forward(self, xb):\n",
        "        return self.network(xb)\n",
        "    \n",
        "    def training_step(self, batch):\n",
        "        images, label = batch\n",
        "        labels = to_device(torch.tensor([x-1 for x in label]), device)\n",
        "        out = self(images)                  # Generate predictions\n",
        "        loss = F.cross_entropy(out, labels) # Calculate loss\n",
        "        return loss\n",
        "    \n",
        "    def validation_step(self, batch):\n",
        "        images, label = batch\n",
        "        labels = to_device(torch.tensor([x-1 for x in label]), device)\n",
        "        out = self(images)                    # Generate predictions\n",
        "        loss = F.cross_entropy(out, labels)   # Calculate loss\n",
        "        acc = accuracy(out, labels)           # Calculate accuracy\n",
        "        return {'val_loss': loss, 'val_acc': acc}\n",
        "        \n",
        "    def validation_epoch_end(self, outputs):\n",
        "        batch_losses = [x['val_loss'] for x in outputs]\n",
        "        epoch_loss = torch.stack(batch_losses).mean()   # Combine losses\n",
        "        batch_accs = [x['val_acc'] for x in outputs]\n",
        "        epoch_acc = torch.stack(batch_accs).mean()      # Combine accuracies\n",
        "        return {'val_loss': epoch_loss.item(), 'val_acc': epoch_acc.item()}\n",
        "    \n",
        "    def epoch_end(self, epoch, result):\n",
        "        print(\"Epoch [{}], val_loss: {:.4f}, val_acc: {:.4f}\".format(epoch, result['val_loss'], result['val_acc']))"
      ],
      "execution_count": 21,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ew17xcKLv1V7",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 357
        },
        "outputId": "d827f7fd-e037-4325-f853-39b2cef18d69"
      },
      "source": [
        "model = EmnistModel()\n",
        "to_device(model, device)"
      ],
      "execution_count": 86,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "EmnistModel(\n",
              "  (network): Sequential(\n",
              "    (0): Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
              "    (1): ReLU()\n",
              "    (2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
              "    (3): ReLU()\n",
              "    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
              "    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
              "    (6): ReLU()\n",
              "    (7): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
              "    (8): ReLU()\n",
              "    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
              "    (10): Flatten()\n",
              "    (11): Linear(in_features=12544, out_features=1024, bias=True)\n",
              "    (12): ReLU()\n",
              "    (13): Linear(in_features=1024, out_features=256, bias=True)\n",
              "    (14): ReLU()\n",
              "    (15): Linear(in_features=256, out_features=26, bias=True)\n",
              "  )\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 86
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "V65UGX3ewGPR",
        "colab_type": "text"
      },
      "source": [
        "#### Predictions with the initial hyperparameters"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UmRLTgjRv6eO",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import torch.nn.functional as F\n",
        "\n",
        "for images, label in train_load:\n",
        "    labels = to_device(torch.tensor([x-1 for x in label]), device)\n",
        "    outputs = to_device(model(images), device)\n",
        "    break"
      ],
      "execution_count": 87,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Swj1IxwxeROk",
        "colab_type": "text"
      },
      "source": [
        "We pass the outputs through the softmax function so that the outputs are in the range 0 to 1 and the sum of the outputs for each data point in the batch is 1, and the outputs can represent probabilities."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "laQkhunSreFK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "probs = F.softmax(outputs, dim=1)"
      ],
      "execution_count": 88,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6cjpJioKglOX",
        "colab_type": "text"
      },
      "source": [
        "We can take the class with the highest probability as the predicted class. The 'max_probs' represents the highest probability and 'preds' represents it's index, which can be called the predicted label."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WR-E-7Cyrkkg",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 119
        },
        "outputId": "6f6a2914-3a72-436a-d8f3-cd2ca98c09b9"
      },
      "source": [
        "max_probs, preds = torch.max(probs, dim=1)\n",
        "preds"
      ],
      "execution_count": 89,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
              "        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
              "        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
              "        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
              "        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
              "        1, 1, 1, 1, 1, 1, 1, 1], device='cuda:0')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 89
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2H4HfQGbhah2",
        "colab_type": "text"
      },
      "source": [
        "These are the labels predicted with the initial set of weights and biases. Since it's totally random, they are very likely to be wrong."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4sBskJLVxQ5I",
        "colab_type": "text"
      },
      "source": [
        "#### Defining accuracy of the model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nafHzdeuiDVH",
        "colab_type": "text"
      },
      "source": [
        "We calculate the accuracy by checking the number of right predictions."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K-deem1Vw_HZ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def accuracy(outputs, labels):\n",
        "    _, preds = torch.max(outputs, dim=1)\n",
        "    return torch.tensor(torch.sum(preds == labels).item() / len(preds))"
      ],
      "execution_count": 90,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5pIwkgn4xaB-",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "c0d3e31e-2e95-4219-914d-3d206e4afca6"
      },
      "source": [
        "accuracy(outputs, labels)"
      ],
      "execution_count": 91,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor(0.0469)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 91
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0clWKEIViNj_",
        "colab_type": "text"
      },
      "source": [
        "The accuracy is quite low, as expected."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "r0O0-MKAxyY3",
        "colab_type": "text"
      },
      "source": [
        "#### Defining the loss function"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HbG1Mug7xpma",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "fecd7693-903c-42d7-e069-41bfc179f5bb"
      },
      "source": [
        "loss_fn = F.cross_entropy\n",
        "\n",
        "loss = loss_fn(outputs, labels)\n",
        "print(loss)"
      ],
      "execution_count": 92,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tensor(3.2587, device='cuda:0', grad_fn=<NllLossBackward>)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "28JPo7I2x8Sd",
        "colab_type": "text"
      },
      "source": [
        "#### Training the model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KPOApHh1i2y2",
        "colab_type": "text"
      },
      "source": [
        "We evaluate the model on the basis of the loss and accuracy.\n",
        "\n",
        "We fit the model to the training and validation datasets. The optimizer we use is SGD to change the weights and biases. For each epoch, in each batch, the loss is calculated on the training data and the model parameters are updated using the optimizer, which is then initialized again, so that the gradients do not keep adding up after each epoch. We evaluate the model with the updated model parameters on the validation dataset, and add each result to history. Number of epochs and learning rate are the hyperparameters whose best value, we have to decide in order to get the best version of the model. Epochs are the iterations, and learning rate is the rate by which the model parameters are changed during gradient descent."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nbg5yRgvx2GC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "@torch.no_grad()\n",
        "def evaluate(model, val_loader):\n",
        "    model.eval()\n",
        "    outputs = [model.validation_step(batch) for batch in val_loader]\n",
        "    return model.validation_epoch_end(outputs)\n",
        "\n",
        "def fit(epochs, lr, model, train_loader, val_loader, opt_func=torch.optim.SGD):\n",
        "    torch.cuda.empty_cache()\n",
        "    history = []\n",
        "    optimizer = opt_func(model.parameters(), lr)\n",
        "    train_losses = []\n",
        "    for epoch in range(epochs):\n",
        "        # Training Phase \n",
        "        model.train()\n",
        "        for batch in train_loader:\n",
        "            loss = model.training_step(batch)\n",
        "            train_losses.append(loss)\n",
        "            loss.backward()\n",
        "            optimizer.step()\n",
        "            optimizer.zero_grad()\n",
        "        # Validation phase\n",
        "        result = evaluate(model, val_loader)\n",
        "        result['train_loss'] = torch.stack(train_losses).mean().item()\n",
        "        model.epoch_end(epoch, result)\n",
        "        history.append(result)\n",
        "    return history"
      ],
      "execution_count": 93,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xqXSCAbbyD3H",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "cca99da2-0409-4223-b1d8-b82fc1e36513"
      },
      "source": [
        "result0 = evaluate(model, val_load)\n",
        "result0"
      ],
      "execution_count": 94,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'val_acc': 0.04113520309329033, 'val_loss': 3.2586867809295654}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 94
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GCBgFBAIyjU9",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "8ba1b2f9-8283-4a19-9c9f-93ba96e1a979"
      },
      "source": [
        "history1 = fit(5, 0.1, model, train_load, val_load)"
      ],
      "execution_count": 95,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch [0], val_loss: 0.2963, val_acc: 0.9026\n",
            "Epoch [1], val_loss: 0.2207, val_acc: 0.9278\n",
            "Epoch [2], val_loss: 0.2315, val_acc: 0.9197\n",
            "Epoch [3], val_loss: 0.2372, val_acc: 0.9184\n",
            "Epoch [4], val_loss: 0.1814, val_acc: 0.9397\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9hF7sSNvyvEi",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "add698c4-85a0-4fdd-c856-08e112296d71"
      },
      "source": [
        "history2 = fit(5, 0.01, model, train_load, val_load)"
      ],
      "execution_count": 96,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch [0], val_loss: 0.1644, val_acc: 0.9463\n",
            "Epoch [1], val_loss: 0.1661, val_acc: 0.9465\n",
            "Epoch [2], val_loss: 0.1683, val_acc: 0.9472\n",
            "Epoch [3], val_loss: 0.1718, val_acc: 0.9467\n",
            "Epoch [4], val_loss: 0.1725, val_acc: 0.9468\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X9gSKp-vze2-",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "220e4b3a-18c9-41e9-ef25-67eba47de1df"
      },
      "source": [
        "history3 = fit(5, 0.001, model, train_load, val_load)"
      ],
      "execution_count": 97,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch [0], val_loss: 0.1735, val_acc: 0.9466\n",
            "Epoch [1], val_loss: 0.1743, val_acc: 0.9467\n",
            "Epoch [2], val_loss: 0.1751, val_acc: 0.9463\n",
            "Epoch [3], val_loss: 0.1759, val_acc: 0.9465\n",
            "Epoch [4], val_loss: 0.1762, val_acc: 0.9470\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FybgSbkiqORr",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        },
        "outputId": "8a477d97-1a96-48d6-999d-35ec098b7b70"
      },
      "source": [
        "history = [result0] + history1 + history2 + history3\n",
        "accuracies = [result['val_acc'] for result in history]\n",
        "plt.plot(accuracies, '-x')\n",
        "plt.xlabel('epoch')\n",
        "plt.ylabel('accuracy')\n",
        "plt.title('Accuracy vs. No. of epochs');"
      ],
      "execution_count": 98,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEWCAYAAABrDZDcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3de5xcdX3/8dd7d3NPSEJ2cyFXSCIaEMhFoVy8oLUoClptlRRrW5Xan1Ss+Cuo/SFFbb38fvYGFqlWlDta1JQiFxFRiSAL4Y5AAglJ2N3s5jKb22Sz2c/vj3M2md3sZidxJzM75/18POaxcy5zzmfOzM7nfL/f8/0eRQRmZpZdNeUOwMzMysuJwMws45wIzMwyzonAzCzjnAjMzDLOicDMLOOcCMyqhKQvSmqT1FzuWAAkXS7p+nLHYQNzIrA+Sfq5pM2SRpQ7lqFC0hxJIemOXvOvl3R5ifc9C7gYWBARU0u5L6s+TgS2H0lzgDOAAM45zPuuO5z7K5GTJZ16mPc5C9gYERsO836tCjgRWF/+FHgQuBb4UOECSTMl3SapVdJGSVcWLPuopGclbZX0jKRF6fyQNK9gvWslfTF9/iZJ6yRdklZpfEfSREm3p/vYnD6fUfD6IyV9R9Ir6fIfpfOfkvSugvWGpVUlC3u/wTTOdxZM16X7WyRpZHoWv1HSFkkPS5pyEMfvq8CX+luYHqeVkjZJWibpqGI2Kmm8pO+lca6R9HeSaiS9FbgHOErSNknX9vP6d0p6LH1PyyWdULBstaTPpJ/b5vT4jiwmZknHSbonXdYi6bMFux2exrxV0tOSlhS87hJJ69Nlz0l6SzHHwUogIvzwo8cDWAn8L2AxsBuYks6vBR4H/gkYA4wETk+X/RGwHngdIGAeMDtdFsC8gu1fC3wxff4moBP4CjACGAVMAt4LjAbGAd8HflTw+v8BbgEmAsOAN6bz/xa4pWC9c4En+3mPlwE3FEyfDTybPv9L4L/T/demx+GIIo7bnPS9jkuPxVvT+dcDl6fPzwTagEXp+/034BdFfi7fA36cbn8O8Dzw4YLjuO4Ar10IbABOTt/Th4DVwIh0+WrgKWAmcCTwQMFn1G/MaSxNJNVSI9Ppk9NllwN54B3pPv8ReDBddiywFjiq4NjNLfd3P6uPsgfgR2U9gNNJfvzr0+nfAn+TPv89oBWo6+N1dwEX9bPNgRJBBzDyADGdBGxOn08DuoCJfax3FLC1+0cb+AHwt/1sc1667uh0+gbgsvT5XwDLgRMO8th1J4I6kkTa/aNXmAi+DXy14DVj0+M9Z4Bt16bHaUHBvL8Efl5wHA+UCP4d+EKvec+xL4muBj5WsOwdwKqBYgbOA1b0s8/LgZ8WTC8AdhYc/w3AW4Fh5f7eZ/3hqiHr7UPA3RHRlk7fyL7qoZnAmojo7ON1M4FVh7jP1ojId09IGi3pm2n1RzvwC2CCpNp0P5siYnPvjUTEKyRnsu+VNAF4O8kP/H4iYiXwLPAuSaNJ2kJuTBdfR5LYbk6rn74qadhBvqdvAVMKq6pSRwFrCuLYBmwEpg+wvXqS0s+agnlrinhdt9nAxWm10BZJW0iOZWG11Npe2+5edqCYB/rcC69g2gGMlFSXHv9PkiSLDZJuLraKzAafE4HtJWkU8MfAGyU1p3X2fwOcKOlEkh+KWf006K4F5vaz6R0k1Szdel/V0nsI3ItJqg5OjogjgDd0h5ju58j0h74v3wXOJ6mq+nVErO9nPYCbSM5ozwWeSX+ciIjdEfH3EbEAOBV4J0m7SdEiogP4e+ALadzdXiH5UU7ekDSGpCrsQHFCUjWzu/C1JA3EA72u21rgSxExoeAxOiJuKlhnZq9tv1JEzGuBY4qMoYeIuDEiTk+3HSTVg1YGTgRW6N3AHpIi/Enp4zXAL0l+CH9DUh/8ZUlj0kbV09LXfgv4tKTFSsyT1P3j8RiwVFKtpLOANw4QxzhgJ7BF0pHA57sXREQT8BPgG2mj8jBJbyh47Y9I6rIvIqlTP5CbgbcBf8W+0gCS3izptWkJpJ3kB7hrgG315TqSevOzCubdBPy5pJOUXJr7D8BDEbH6QBuKiD3ArcCXJI1Lj+2nSKqdivEfwMcknZx+PmMknS1pXME6H5c0Iz3mnyNphxko5tuBaZI+KWlEGtvJAwUj6VhJZ6bby5N83odyjG0wlLtuyo/KeQB3Av+vj/l/TFLEryM5U/wRSdVAG/CvBet9jKTeeRtJw+PCdP4S4GmSOvnrSH5YCtsI1vXa31HAz9PtPE9SFx6kbRMkjZnfBVqAzcBtvV7/LWA7MLaI93wvSWP11IJ556XvY3u6j38t2PfVwNX9bGtOYZwFxy5I2wgKjtMqYBPJD+mMdP6s9D3P6mf7E0l++FtJzsQvA2r6O459vP4s4GFgC0lC/z4wLl22GvgM8Ey6/Luk7ScHijlddnx6HDen35NL0/mXA9f3dXyAE0hOLLYWbPOocv8PZPWh9AMyqxqSLgNeFRHnlzuWoULSauAjEfHTcsdih181dN4x2yut1vgw8MFyx2I2VLiNwKqGpI+SVJn8JCJ+Ue54zIYKVw2ZmWWcSwRmZhk35NoI6uvrY86cOeUOw8xsSHnkkUfaIqKhr2VDLhHMmTOHxsbGcodhZjakSFrT3zJXDZmZZZwTgZlZxjkRmJllnBOBmVnGORGYmWWcE4FVhavvX8XyVW095i1f1cbV9x/aLRIGe3tDIcZK395QiHEovOe+OBFYVThhxnguvHHF3n+Y5avauPDGFZwwY3xFbG8oxFjp2xsKMQ6F99yXITfExJIlS8L9CIa+q+9fxQkzxnPq3Pq985avauOJdTk+9sb+7m+zv66uYGu+ky07O/jF86189c7nOONV9fzi+TY+cvrRHDt13N673hR+1SOd2z1v3zr7VnqhZSvfWb6a0+bW88CqNj54yhyOaRhDRNAV0JX+jQi6uvbNi0i2XzjdvXzNxu3c8VQTJ0yfwJPrc5z92mnMmjSaGoEkJKiREOnfdH6Nkrvb1NQky7rXfaltOzc99DKnHDOJB1/cyNKTZzG3Yex+7zkK7v3T+z13zwjgxdbt3Nq4ltfNOZKHV2/i/UtmckzDGEhjAlD6RKjgec/53Suv2rCN6x9cw+/NrefXq9o4/5TZzJs8ttf++/5Men4u+9Z5sXUbN/1mLScffSQPvbSJpSfPYl7D2ORYpfvuPkbJce0Vq7Q3XiGeb9nKt3/1Em8o0ffmT0+Zw9zJY5Mhn+ke+j95bVfBcwq+T8myfdt9qW07tz26nnMXHsXdT7dw5dKFPf53iiHpkYhY0ucyJwIrh+6zmu4v9PKVbXz8xke5/F3HcUzDWDbv6GDLzt1s2dHBlh272byjg1z6N5mfLMvt3E3X0PoKm/1OPnHmPD71tmMP+nUHSgRDrmdxNRqss+OhICJ4edMOmnN5Fs6cwAe//RuG14qdu5ObU110y2N9vm7siDomjB7GhNHDmDh6ONMnjGLi6OFMGD2M8aOSec25nXzzFy9y9gnT+J8nm/js21/DSbOSO1p2n892nxUm8+g1r+c6K17ezBduf5ZzTpzGssebuOKc41g8ZyI1UvrYd9bZPU817D2bryk4w+9e58EXN3LhTSs4/+RZXP/Qy/zbBxZy8jFH9jw77C5J9JhXUPpISxwR0Lh6E5/74VO8Z+F0frhiPV949/EsmTOxz/dceL9MCs/eC9ZpXL2JS297kvctmsEPHl3Hl//wtSyefeS+EkXBn8Kz9Ogxf99Z86NrNnPZsqd590lH8aPHXuGKc45j0eyJ+8IoOEvv+zPZP8ZHVm/isz98ij9cNJ3bHl3PF99zPItmTexxtt29/+6S2b6z8TTi2PceHlu7mS/d8Sxnv7Y035vLz1nA4llHpqW7fSWSvSW+dBM1Us+STHfpRsnncvGtj7M0/d6cMnfSQZcIDqjcd8Y52MfixYuj2jywsjUWXnF3PLCytc/poSy/uzMaV2+Kb96/Mi743sOx+Av3xOxLbo/Zl9wex192Z5z+lXtj9iW3x3u/8UBcc/+quOXhl+Pup5vjNy9tjBda2mNDez46OvcMuJ/BPoal+EwqPcZK395QiLGS3zPQGNVyh7JqrRp64IU2/uqGR/iD46dy99Mt/PP7T+TNr55ySNsqZwmjbdsuHlmzmUfXbKZxzWaeXJejY09ytj970mgWz5rIotkTWTJnIm1bO/jEzfvOjg+l3rPbYL/nUhzDSo+x0rc3FGKs5PfsNoIK9cqWnfzqhTZ+ubKNB1a2sWl7R4/lo4bVUj9uOJPGjKB+bPJ30tjh1I/d/+/E0cOprUnKpvvVv/eaPlj9fREfX7uFM189hcY1m/b++K/euAOA4bU1HD/9CJbMOZJFsyayaPYEJo8b2eP1gxmjmR2YE0GF2Lark4de3MgvX2jjly+0sqp1OwD1Y0fw6qnjWPHyZs58zWR+9uwG3nXiUYwdUUfbtl1s3N5B27YONqbP9/TROirBkaOHMylNGCipnz1xRnJ1yjknHsXRDWOolait6fXoa173/FrxXPNWrrpvJZ9863zmTR7Hfz/+Cj9+bD11Nfvq9ieNGc7i2RNZnJ7tH3fUeEYOq+33WGSpXcSsEjgRlMmeruCJdVv2nvU/umYznV3ByGE1vP7oSZwxr54zXlXPpm0dXHhTcWfHXV1Be343bds6kiSxrYON23ftSxQF0+s376BjT2k+35kTR3H6/AaWpD/+syeN3tvgZ2aVx4lgkB3obPYdx0/jlytb+dULbSxftZHczt0AHD/9CE6f18AZ8+tZPHtij7PlUpwddyeTPzl5Fjc8+DL//IGTeN2cI9kTwZ49wZ4IOru66Oqix989XemyPUFXBJ1dyTXynV3BLQ+v5Ycr1nPBG47hs+94zSEePTMrB18+Osi6e/pduXQhxx01nmsfeIlv/HwVE0YP48s/+S0A08aP5A+Om8Lp8xs4be4kJo0d0e/2+vqxP3Vu/SHXlfcuUfze3Em/c/378lVt3P98K584cx7XP/Qybzq2wXX5ZlXCJYJD9LPftnDB9x6hM62vH1lXw2nz6jl9fj1nzG9gbsOYslWVDHYJww27ZkOfSwQlMH7UsL1J4L2LpvOPf3gCw+sqY+imwS5hPLEu1+NH/9S59Vy5dCFPrMs5EZhVASeCQ3T/860ALD15Fnc+1Uzjmk1V+6M42InFzCpLZZzCDjHLV7XxrV++BMAlZ72aK5cu7DE6oJnZUOJEcAieWJfjTa9qYNSwWo4YWdejqsTMbKhxIjgEH3vjXFQjpo0fubdB+NS59e4IZWZDkhPBIWrJ5ZlyxMiBVzQzq3BOBIeoKZdn2ngnAjMb+pwIDkFXV9DSnmeqE4GZVQEngkOwcXsHnV3hRGBmVcGJ4BA05/IATHUbgZlVASeCQ9CU2wnAtPGjyhyJmdnvzongELS0JyWCKeP7H0jOzGyocCI4BE25PHU1on6ME4GZDX1OBIegOe1DUFPjG7GY2dDnRHAImn3pqJlVESeCQ9CccyIws+rhRHCQIoLm9jzTfOmomVUJJ4KD1J7vZEfHHpcIzKxqOBEcpL2dyZwIzKxKlDQRSDpL0nOSVkq6tI/lsyTdJ2mFpCckvaOU8QyG5nb3Kjaz6lKyRCCpFrgKeDuwADhP0oJeq/0dcGtELAQ+AHyjVPEMlua0V7FLBGZWLUpZIng9sDIiXoyIDuBm4Nxe6wRwRPp8PPBKCeMZFE25PBJMHudEYGbVoZSJYDqwtmB6XTqv0OXA+ZLWAXcAf93XhiRdIKlRUmNra2spYi1aS3ueSWNGMLzOzStmVh3K/Wt2HnBtRMwA3gFcJ2m/mCLimohYEhFLGhoaDnuQhXxDGjOrNqVMBOuBmQXTM9J5hT4M3AoQEb8GRgL1JYzpd+bOZGZWbUqZCB4G5ks6WtJwksbgZb3WeRl4C4Ck15AkgvLW/QyguT3vK4bMrKqULBFERCdwIXAX8CzJ1UFPS7pC0jnpahcDH5X0OHAT8GcREaWK6XeV372HLTt2u0RgZlWlrpQbj4g7SBqBC+ddVvD8GeC0UsYwmLo7k7mNwMyqSbkbi4eUJt+i0syqkBPBQWhud2cyM6s+TgQHoTm3C3AiMLPq4kRwEJpzOzliZB2jh5e0acXM7LByIjgISWeyUeUOw8xsUDkRHISW9jxTXC1kZlXGieAgNOV8ZzIzqz5OBEXavaeL1m273FBsZlXHiaBIrVt3EeErhsys+jgRFKnJt6g0syrlRFCklnYPL2Fm1cmJoEgeXsLMqpUTQZGaczsZOayG8aOGlTsUM7NB5URQpOb2XUwbPwpJ5Q7FzGxQOREUqTm3kylHjCh3GGZmg86JoEgeXsLMqpUTQRG6uoIN7buY4oZiM6tCTgRF2LSjg449Xb501MyqkhNBEZrdmczMqpgTQRGa3YfAzKqYE0ERmtyr2MyqmBNBEZpzO6mrEZPG+vJRM6s+TgRFaM7tYvK4EdTWuDOZmVUfJ4IiNLfvdEOxmVUtJ4IiNLszmZlVMSeCAUQETbm8O5OZWdVyIhjA1l2d7OjY4yuGzKxqOREMoCXtQzDFicDMqpQTwQC6b0jjEoGZVSsnggG4V7GZVTsnggE0p72K3VhsZtXKiWAATbk89WOHM7zOh8rMqpN/3QbQnHNnMjOrbk4EA2hu3+X2ATOrak4EA3CJwMyqXUkTgaSzJD0naaWkS/tZ548lPSPpaUk3ljKeg5XfvYfNO3Z7eAkzq2p1pdqwpFrgKuD3gXXAw5KWRcQzBevMBz4DnBYRmyVNLlU8h6LFVwyZWQaUskTwemBlRLwYER3AzcC5vdb5KHBVRGwGiIgNJYznoLkzmZllQSkTwXRgbcH0unReoVcBr5L0gKQHJZ3V14YkXSCpUVJja2tricLdX3eJwG0EZlbNyt1YXAfMB94EnAf8h6QJvVeKiGsiYklELGloaDhswTW5V7GZZUApE8F6YGbB9Ix0XqF1wLKI2B0RLwHPkySGitCcyzNuZB1jRpSsKcXMrOyKSgSSbpN0tqSDSRwPA/MlHS1pOPABYFmvdX5EUhpAUj1JVdGLB7GPkmrO5V0aMLOqV+wP+zeApcALkr4s6diBXhARncCFwF3As8CtEfG0pCsknZOudhewUdIzwH3A/46IjQf9LkqkqT3v9gEzq3pF1XlExE+Bn0oaT1KX/1NJa4H/AK6PiN39vO4O4I5e8y4reB7Ap9JHxWnO7eTYKYevTcLMrByKruqRNAn4M+AjwArgX4BFwD0liazMOvd00brVw0uYWfUrqkQg6YfAscB1wLsioilddIukxlIFV06t23bRFTDVvYrNrMoVeznMv0bEfX0tiIglgxhPxXBnMjPLimKrhhYUXt8vaaKk/1WimCrC3nsVu2rIzKpcsYngoxGxpXsiHRLio6UJqTK4RGBmWVFsIqiVpO6JdEC54aUJqTK0tOcZUVfDhNHDyh2KmVlJFdtGcCdJw/A30+m/TOdVraZc0oegIP+ZmVWlYhPBJSQ//n+VTt8DfKskEVUI9yo2s6wotkNZF/Dv6SMTmtvzLJy13/h3ZmZVp9h+BPOBfwQWAHtPkyPimBLFVVYRkZQI3FBsZhlQbGPxd0hKA53Am4HvAdeXKqhy27S9g449XUxz1ZCZZUCxiWBURNwLKCLWRMTlwNmlC6u8mn1DGjPLkGIbi3elQ1C/IOlCkvsKjC1dWOXV3H1DGg8vYWYZUGyJ4CJgNPAJYDFwPvChUgVVbu5MZmZZMmCJIO089v6I+DSwDfjzkkdVZi3teWprRP3YEeUOxcys5AYsEUTEHuD0wxBLxWjK5Zk8bgS1Ne5MZmbVr9g2ghWSlgHfB7Z3z4yI20oSVZm1+M5kZpYhxSaCkcBG4MyCeQFUZSJoyuWZP7lq28LNzHootmdx1bcLFGrO5Tljfn25wzAzOyyK7Vn8HZISQA8R8ReDHlGZbc3vZtuuTl8xZGaZUWzV0O0Fz0cC7wFeGfxwyq+l3TekMbNsKbZq6L8KpyXdBPyqJBGV2b4+BO5MZmbZUGyHst7mA5MHM5BKsbdXsUsEZpYRxbYRbKVnG0EzyT0Kqk53Iph8hDuTmVk2FFs1NK7UgVSKpvY8k8YMZ+Sw2nKHYmZ2WBRVNSTpPZLGF0xPkPTu0oVVPi25vBuKzSxTim0j+HxE5LonImIL8PnShFReTbm8Lx01s0wpNhH0tV6xl54OKc0eXsLMMqbYRNAo6euS5qaPrwOPlDKwcsjv3sOm7R2+YsjMMqXYRPDXQAdwC3AzkAc+XqqgymVD+y7AdyYzs2wp9qqh7cClJY6l7LpvUenOZGaWJcVeNXSPpAkF0xMl3VW6sMqjKbcTgKnj3YfAzLKj2Kqh+vRKIQAiYjNV2LPY9yo2sywqNhF0SZrVPSFpDn2MRjrUNbfnGTuijrEjqvKCKDOzPhWbCD4H/ErSdZKuB+4HPjPQiySdJek5SSsl9dvGIOm9kkLSkiLjKYnmnC8dNbPsKSoRRMSdwBLgOeAm4GJg54Fek970/irg7cAC4DxJC/pYbxxwEfDQQUVeAu5MZmZZVGxj8UeAe0kSwKeB64DLB3jZ64GVEfFiRHSQXHZ6bh/rfQH4CsklqWXV0u7hJcwse4qtGroIeB2wJiLeDCwEthz4JUwH1hZMr0vn7SVpETAzIv6nyDhKpnNPFxu27nKJwMwyp9hEkI+IPICkERHxW+DY32XHkmqAr5OUMgZa9wJJjZIaW1tbf5fd9qttWwd7usJtBGaWOcUmgnVpP4IfAfdI+jGwZoDXrAdmFkzPSOd1GwccD/xc0mrgFGBZXw3GEXFNRCyJiCUNDQ1FhnxwujuTeXgJM8uaYnsWvyd9ermk+4DxwJ0DvOxhYL6ko0kSwAeApQXbzAH13dOSfg58OiIai45+EDXv7UzmRGBm2XLQF8xHxP1Frtcp6ULgLqAW+M+IeFrSFUBjRCw72H2XUrPvVWxmGVXSnlMRcQdwR695l/Wz7ptKGctAmtrzDK+rYeLoYeUMw8zssDvUm9dXneZcnqlHjERSuUMxMzusnAhS7lVsZlnlRJBqbs/7iiEzyyQnAiAiPLyEmWWWEwGwZcduOjq7PLyEmWWSEwHJYHOASwRmlklOBEBzuzuTmVl2OREAzTnftN7MssuJgGR4iRpBw1jfq9jMsseJgOTS0cnjRlJX68NhZtnjXz6SxuIprhYys4xyIiDpVTzNl46aWUY5EZD2KnaJwMwyKvOJYNuuTrbmO50IzCyzMp8Imt2ZzMwyLvOJoCW9RaWHlzCzrMp8IvDwEmaWdZlPBN33KnaJwMyyyomgPc/E0cMYOay23KGYmZWFE0Euz1TfsN7MMizzicA3pDGzrMt8Imhpz7t9wMwyLdOJYFfnHtq2dbhEYGaZlulEsKHd9yEwM8t0ImhOO5NNddWQmWVYphOBO5OZmWU8EbSkicBVQ2aWZZlOBE25PGOG1zJu5LByh2JmVjaZTgTN7TtdGjCzzMt2Isj5hjRmZk4ER3h4CTPLtswmgj1dQcvWXb5iyMwyL7OJYOO2XezpCqY4EZhZxmU2EeztQ+DOZGaWcZlNBHt7FbtEYGYZV9JEIOksSc9JWinp0j6Wf0rSM5KekHSvpNmljKdQszuTmZkBJUwEkmqBq4C3AwuA8yQt6LXaCmBJRJwA/AD4aqni6a0pl2d4bQ1Hjh5+uHZpZlaRSlkieD2wMiJejIgO4Gbg3MIVIuK+iNiRTj4IzChhPD20tOeZMn4ENTU6XLs0M6tIpUwE04G1BdPr0nn9+TDwk74WSLpAUqOkxtbW1kEJrim306OOmplRIY3Fks4HlgBf62t5RFwTEUsiYklDQ8Og7NP3KjYzS5QyEawHZhZMz0jn9SDprcDngHMiYlcJ49krImhuzzP1iBGHY3dmZhWtlIngYWC+pKMlDQc+ACwrXEHSQuCbJElgQwlj6SG3czf53V0uEZiZUcJEEBGdwIXAXcCzwK0R8bSkKySdk672NWAs8H1Jj0la1s/mBpVvSGNmtk9dKTceEXcAd/Sad1nB87eWcv/96e5MNsWNxWZmldFYfLg1u0RgZrZXJhNBUy5PjaBhnBuLzcwymQhacnnqx45gWG0m376ZWQ+Z/CVsas+7WsjMLJXJRNDiW1Same2VyUTg4SXMzPbJXCLY0dFJe77TncnMzFKZSwS+dNTMrKfMJgJ3JjMzS2QuEXh4CTOznjKXCHyvYjOznrKXCHJ5JowexshhteUOxcysImQuETTl8r501MysQOYSQUu7O5OZmRXKXCJoynl4CTOzQplKBB2dXWzcvoupR7gzmZlZt0wlgg1b80TA1PEeftrMrFumEkF3ZzIPL2Fmtk+2EkG7O5OZmfWWrUTg4SXMzPaTqUTQlMszengtR4ysK3coZmYVI1OJoDntQyCp3KGYmVWMbCUC9yo2M9tP9hKBG4rNzHrITCLo6opkeAmXCMzMeshMImjbvovOrvClo2ZmvVR9Irj6/lUsX9VGS24XkHQmW76qjavvX1XmyMzMKkPVJ4ITZoznwhtXcN9zGwDY0J7nwhtXcMKM8WWOzMysMlR9Ijh1bj1XLl24twTwtbuf48qlCzl1bn2ZIzMzqwxVnwggSQZvPnYyAB88ebaTgJlZgUwkguWr2vj1ixv5xJnzuOE3L7N8VVu5QzIzqxhVnwiWr2rjwhtXcOXShXzqbcdy5dKFXHjjCicDM7NU1SeCJ9blerQJdLcZPLEuV+bIzMwqgyKi3DEclCVLlkRjY2O5wzAzG1IkPRIRS/paVvUlAjMzO7CSJgJJZ0l6TtJKSZf2sXyEpFvS5Q9JmlPKeMzMbH8lSwSSaoGrgLcDC4DzJC3otdqHgc0RMQ/4J+ArpYrHzMz6VsoSweuBlRHxYkR0ADcD5/Za51zgu+nzHwBvkW8WYGZ2WJUyEUwH1hZMr0vn9blORHQCOWBS7w1JukBSo6TG1tbWEoVrZpZNQ+KejRFxDXANgKRWSWsOcVP1QKV3IKj0GCs9PnCMg6HS44PKj7HS4pvd34JSJoL1wMyC6RnpvL7WWSepDhgPbDzQRiOi4VADktTY3+VTlaLSY6z0+MAxDoZKjw8qP8ZKj69QKauGHmcl9/kAAAV7SURBVAbmSzpa0nDgA8CyXussAz6UPn8f8LMYah0bzMyGuJKVCCKiU9KFwF1ALfCfEfG0pCuAxohYBnwbuE7SSmATSbIwM7PDqKRtBBFxB3BHr3mXFTzPA39Uyhh6ueYw7utQVXqMlR4fOMbBUOnxQeXHWOnx7TXkhpgwM7PB5SEmzMwyzonAzCzjMpMIBhr3qJwkzZR0n6RnJD0t6aJyx9QfSbWSVki6vdyx9EXSBEk/kPRbSc9K+r1yx1RI0t+kn/FTkm6SNLICYvpPSRskPVUw70hJ90h6If07sQJj/Fr6OT8h6YeSJlRSfAXLLpYUkir21oiZSARFjntUTp3AxRGxADgF+HiFxVfoIuDZcgdxAP8C3BkRrwZOpIJilTQd+ASwJCKOJ7marhKulLsWOKvXvEuBeyNiPnBvOl1O17J/jPcAx0fECcDzwGcOd1AFrmX/+JA0E3gb8PLhDuhgZCIRUNy4R2UTEU0R8Wj6fCvJj1fv4TjKTtIM4GzgW+WOpS+SxgNvILksmYjoiIgt5Y1qP3XAqLQD5WjglTLHQ0T8guTy7UKF44B9F3j3YQ2ql75ijIi706FpAB4k6bRaFv0cQ0gG0/xboKKvyslKIihm3KOKkA7FvRB4qLyR9OmfSb7UXeUOpB9HA63Ad9Lqq29JGlPuoLpFxHrg/5KcHTYBuYi4u7xR9WtKRDSlz5uBKeUMpgh/Afyk3EEUknQusD4iHi93LAPJSiIYEiSNBf4L+GREtJc7nkKS3glsiIhHyh3LAdQBi4B/j4iFwHbKX6WxV1rPfi5JwjoKGCPp/PJGNbC0t3/FntFK+hxJ9eoN5Y6lm6TRwGeBywZatxJkJREUM+5RWUkaRpIEboiI28odTx9OA86RtJqkau1MSdeXN6T9rAPWRUR3aeoHJImhUrwVeCkiWiNiN3AbcGqZY+pPi6RpAOnfDWWOp0+S/gx4J/AnFTY8zVyShP94+j8zA3hU0tSyRtWPrCSCYsY9Kpv0HgzfBp6NiK+XO56+RMRnImJGRMwhOX4/i4iKOpuNiGZgraRj01lvAZ4pY0i9vQycIml0+pm/hQpqzO6lcBywDwE/LmMsfZJ0FklV5TkRsaPc8RSKiCcjYnJEzEn/Z9YBi9LvaMXJRCJIG5S6xz16Frg1Ip4ub1Q9nAZ8kOQs+7H08Y5yBzVE/TVwg6QngJOAfyhzPHulJZUfAI8CT5L8/5V9GAJJNwG/Bo6VtE7Sh4EvA78v6QWSksyXKzDGK4FxwD3p/8zVFRbfkOEhJszMMi4TJQIzM+ufE4GZWcY5EZiZZZwTgZlZxjkRmJllnBOB2WEk6U2VOnKrZZcTgZlZxjkRmPVB0vmSfpN2VPpmeh+GbZL+Kb2fwL2SGtJ1T5L0YMG4+BPT+fMk/VTS45IelTQ33fzYgnsm3JD2MjYrGycCs14kvQZ4P3BaRJwE7AH+BBgDNEbEccD9wOfTl3wPuCQdF//Jgvk3AFdFxIkkYwp1j+a5EPgkyb0xjiHpWW5WNnXlDsCsAr0FWAw8nJ6sjyIZdK0LuCVd53rgtvQeCBMi4v50/neB70saB0yPiB8CREQeIN3ebyJiXTr9GDAH+FXp35ZZ35wIzPYn4LsR0eOOV5L+T6/1DnV8ll0Fz/fg/0MrM1cNme3vXuB9kibD3vv3zib5f3lfus5S4FcRkQM2Szojnf9B4P70TnPrJL073caIdIx6s4rjMxGzXiLiGUl/B9wtqQbYDXyc5EY3r0+XbSBpR4BkmOar0x/6F4E/T+d/EPimpCvSbfzRYXwbZkXz6KNmRZK0LSLGljsOs8HmqiEzs4xzicDMLONcIjAzyzgnAjOzjHMiMDPLOCcCM7OMcyIwM8u4/w/JkoAwSAuUjAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8kMcRJaQqX8B",
        "colab_type": "text"
      },
      "source": [
        "As we can see, the accuracy becomes constant towards the end. It means that we have reached the linear regression limit, and to get better accuracies, we will have to use to more advanced models."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hxJBzRJNZ8Sa",
        "colab_type": "text"
      },
      "source": [
        "Training loss and validation loss is plotted to check if there is a case of overfitting."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ylxeO5lIsLJD",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def plot_losses(history):\n",
        "    train_losses = [x.get('train_loss') for x in history]\n",
        "    val_losses = [x['val_loss'] for x in history]\n",
        "    plt.plot(train_losses, '-bx')\n",
        "    plt.plot(val_losses, '-rx')\n",
        "    plt.xlabel('epoch')\n",
        "    plt.ylabel('loss')\n",
        "    plt.legend(['Training', 'Validation'])\n",
        "    plt.title('Loss vs. No. of epochs');"
      ],
      "execution_count": 99,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rkJsPsKQsTEm",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        },
        "outputId": "e8b3f7d0-f2e0-4640-8f73-e910982ad7f5"
      },
      "source": [
        "plot_losses(history)"
      ],
      "execution_count": 100,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wSdP1JinnNBW",
        "colab_type": "text"
      },
      "source": [
        "Our model is not overfitting."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "U4MOtieL0iOM",
        "colab_type": "text"
      },
      "source": [
        "#### Downloading and Transforming the test dataset\n",
        "The test dataset is downloaded with train permissions as False, since we are not going to use it for training, and converted to tensors."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PMs3LTfq0QsH",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "test_dataset = EMNIST(root='data/', \n",
        "                     train=False,\n",
        "                     transform=transforms.ToTensor(), split='letters')"
      ],
      "execution_count": 101,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fTwQURA85-Z1",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "5cea6b0b-da00-4a99-977b-8580ba723be9"
      },
      "source": [
        "len(test_dataset)"
      ],
      "execution_count": 102,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "20800"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 102
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x_AT6VkT0tY1",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 299
        },
        "outputId": "adfe194c-0b8c-4280-ea60-6d59c657e185"
      },
      "source": [
        "img, label = test_dataset[0]\n",
        "plt.imshow(img[0], cmap='gray')\n",
        "print('Shape:', img.shape)\n",
        "print('Label:', label-1)"
      ],
      "execution_count": 104,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Shape: torch.Size([1, 28, 28])\n",
            "Label: 0\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAPHUlEQVR4nO3db4xV9Z3H8c8XGDAZ0DAMO04AhSU8gTVSHVCjMawF4vrAoSYqPNjQVJc+KElr+mCNG1PjRmPWbck+qplGUzBdm0ZEUOtuAauIDxrRsPzTVpYwlnFgBJTShD878N0HczCDzvmd4f47F77vVzKZued7f/d+c+XjOff+7jk/c3cBuPKNKbsBAI1B2IEgCDsQBGEHgiDsQBDjGvlkZsZH/0CdubuNtL2qPbuZ3W1mfzSz/Wb2aDWPBaC+rNJ5djMbK+lPkpZIOiTpfUkr3H1fYgx7dqDO6rFnXyhpv7sfcPezkn4tqbuKxwNQR9WEfZqkPw+7fSjbdhEzW2VmO8xsRxXPBaBKdf+Azt17JPVIHMYDZapmz94nacaw29OzbQCaUDVhf1/SHDObZWbjJS2XtKk2bQGotYoP49190MxWS/pvSWMlveDue2vWGYCaqnjqraIn4z07UHd1+VINgMsHYQeCIOxAEIQdCIKwA0EQdiCIhp7PjsYbO3Zssj558uRkfcKECcn6sWPHkvXTp08n62gc9uxAEIQdCIKwA0EQdiAIwg4EQdiBIJh6uwy0tLQk69OmfeNqYF/p6upKjr3//vuT9alTpybrb7/9drK+bt263Fpvb29yLIuO1hZ7diAIwg4EQdiBIAg7EARhB4Ig7EAQhB0Ignn2JlB0GurcuXOT9TVr1uTW5s2blxzb1taWrI8Zk94fLFiwIFmfM2dObu3xxx9Pjj148GCyjkvDnh0IgrADQRB2IAjCDgRB2IEgCDsQBGEHgmCevQHa29uT9RtvvDFZX716dbJ+22235daKLgVdrdbW1mR92bJlubX9+/cnxz777LPJ+qlTp5J1zoe/WFVhN7ODkk5KOidp0N3TV0oAUJpa7Nn/3t2P1uBxANQR79mBIKoNu0v6nZl9YGarRrqDma0ysx1mtqPK5wJQhWoP4+9w9z4z+xtJm83sY3ffNvwO7t4jqUeSzIxPTICSVLVnd/e+7PeApA2SFtaiKQC1V3HYzazVzCZd+FvSUkl7atUYgNqq5jC+Q9IGM7vwOP/p7v9Vk66uMNdee22yvnBh+oDo1ltvTdbHjx9/yT1dkP33q1jRXPZVV12VW5s/f35ybNHr1tfXl6yfOXMmWY+m4rC7+wFJ6W+DAGgaTL0BQRB2IAjCDgRB2IEgCDsQBKe41sC4cemXccmSJcn6fffdl6x3dHRcck+jdf78+WS92qm51GWyFy9enBzb3d2drG/YsCFZ51LUF2PPDgRB2IEgCDsQBGEHgiDsQBCEHQiCsANBMM9eA0XLHs+cOTNZnzJlSlXPnzrN9LPPPkuOLbocc9FpphMnTkzW0TzYswNBEHYgCMIOBEHYgSAIOxAEYQeCIOxAEMyzj1LqnPXly5cnxz744IPJ+uTJk5P1oss1HzhwILf24osvJscW9VZ0rn6R1Pnyb731VnLsxo0bk/X+/v6KeoqKPTsQBGEHgiDsQBCEHQiCsANBEHYgCMIOBME8+yil5sLvvPPO5Nj29vZkfcyY9P9zBwYGkvWnn346t3b27Nnk2KLeqlkOWkovm7xt27bk2E8//TRZHxwcrKinqAr37Gb2gpkNmNmeYdvazGyzmX2S/U5/KwRA6UZzGP9LSXd/bdujkra6+xxJW7PbAJpYYdjdfZuk41/b3C1pbfb3WknLatwXgBqr9D17h7tf+GLyYUm5i5GZ2SpJqyp8HgA1UvUHdO7uZpZ7poa790jqkaTU/QDUV6VTb0fMrFOSst/pj4sBlK7SsG+StDL7e6Wk9LmIAEpXeBhvZi9JWiSp3cwOSfqJpGck/cbMHpLUK+mBejbZDFpbW3Nr119/fXJs0Tx6kSNHjiTre/fuza0tWrQoObbouu/V9n7y5MncWuo8fKl47XhcmsKwu/uKnNK3a9wLgDri67JAEIQdCIKwA0EQdiAIwg4EwSmuo5S6pPLVV19d1WMXnaq5devWZL2jI/fbylq6dGly7IQJE5L1IkXTY9u3b8+tvffee1U9Ni4Ne3YgCMIOBEHYgSAIOxAEYQeCIOxAEIQdCIJ59ho4d+5csl50KeiipYd7e3uT9Ycffji3VnSZ62pPYS2aC9+3b19u7YsvvqjquXFp2LMDQRB2IAjCDgRB2IEgCDsQBGEHgiDsQBDMs4/Sl19+mVt7/fXXk2N3796drKfmoiVp8eLFyfpdd92VW0udhy9JZpasFymap583b15uLbUMtlT8/QRcGvbsQBCEHQiCsANBEHYgCMIOBEHYgSAIOxAE8+yjdOzYsdzaU089lRx76tSpZL3onPC5c+cm62fOnMmtpZaaliR3T9aLFM2z33777bm1e++9Nzl2y5YtyfqJEyeSdc6Xv1jhnt3MXjCzATPbM2zbE2bWZ2Y7s5976tsmgGqN5jD+l5LuHmH7Gnefn/38trZtAai1wrC7+zZJxxvQC4A6quYDutVmtis7zM/9krOZrTKzHWa2o4rnAlClSsP+c0mzJc2X1C/pp3l3dPced+9y964KnwtADVQUdnc/4u7n3P28pF9IWljbtgDUWkVhN7POYTe/I2lP3n0BNAcrmmc1s5ckLZLULumIpJ9kt+dLckkHJX3f3dMXPx96rOomda9QReeU33zzzcn6I488klu76aabkmOL1mcvmqdva2tL1lPz8J9//nly7Kuvvpqsv/baa8n6m2++mVu7ktd+d/cR/0EVfqnG3VeMsPn5qjsC0FB8XRYIgrADQRB2IAjCDgRB2IEgOMW1CRRNfx44cCBZf/nll3NrH3/8cXLsNddck6zPmjUrWU+dwipJU6dOza0VTdsV1cePH5+s42Ls2YEgCDsQBGEHgiDsQBCEHQiCsANBEHYgCObZLwPHj6cvAbhx48bcWtFpoC0tLcn6pEmTkvWiy0E/+eSTubX29vbk2KJ59mnTpiXrqdN3iy7vfSVizw4EQdiBIAg7EARhB4Ig7EAQhB0IgrADQTDPfgVIXRa56JLJg4ODyXrRfHTR5Z5Tl6Lu7u5Ojr3llluS9euuuy5Z7+3tza298cYbybFX4qWm2bMDQRB2IAjCDgRB2IEgCDsQBGEHgiDsQBDMs6MqR48eTdafe+653Nq4cel/fl1dXcl6Z2dnsj579uzcWtEy2Veiwj27mc0ws9+b2T4z22tmP8y2t5nZZjP7JPs9uf7tAqjUaA7jByX92N3nSrpV0g/MbK6kRyVtdfc5krZmtwE0qcKwu3u/u3+Y/X1S0keSpknqlrQ2u9taScvq1SSA6l3Se3YzmynpW5L+IKnD3fuz0mFJHTljVklaVXmLAGph1J/Gm9lESesl/cjd/zK85kMrE464OqG797h7l7unP20BUFejCruZtWgo6L9y91eyzUfMrDOrd0oaqE+LAGqh8DDehuYonpf0kbv/bFhpk6SVkp7JfudfzxhhnTlzJre2fv365NgbbrghWZ8+fXqy/u677+bWzp07lxx7JRrNe/bbJf2jpN1mtjPb9piGQv4bM3tIUq+kB+rTIoBaKAy7u2+XlPcNhG/Xth0A9cLXZYEgCDsQBGEHgiDsQBCEHQiCU1xRmsOHDyfrW7ZsSdYXLFiQrA99sRMXsGcHgiDsQBCEHQiCsANBEHYgCMIOBEHYgSCYZ0dpTp8+naxv3749WZ80aVKynlryedeuXcmxRUtZX47YswNBEHYgCMIOBEHYgSAIOxAEYQeCIOxAEMyzo2kVne/+zjvvJOtLly7NrbW2tibHnjhxIlm/HLFnB4Ig7EAQhB0IgrADQRB2IAjCDgRB2IEgrOja2mY2Q9I6SR2SXFKPu/+HmT0h6Z8kfZ7d9TF3/23BY3Ehb9RMS0tLsj5lypTc2tGjR5NjL+fz2d19xFWXR/OlmkFJP3b3D81skqQPzGxzVlvj7v9eqyYB1M9o1mfvl9Sf/X3SzD6SNK3ejQGorUt6z25mMyV9S9Ifsk2rzWyXmb1gZpNzxqwysx1mtqOqTgFUpfA9+1d3NJso6R1JT7n7K2bWIemoht7H/6ukTnf/XsFj8J4dNcN79pHlvWcf1Z7dzFokrZf0K3d/JXvAI+5+zt3PS/qFpIW1ahZA7RWG3cxM0vOSPnL3nw3b3jnsbt+RtKf27QGoldFMvd0h6V1JuyWdzzY/JmmFpPkaOow/KOn72Yd5qcfiMB6os7zD+FG/Z68Fwg7UX1Xv2QFc/gg7EARhB4Ig7EAQhB0IgrADQRB2IAjCDgRB2IEgCDsQBGEHgiDsQBCEHQiCsANBNHrJ5qOSeofdbs+2NaNm7a1Z+5LorVK17O36vEJDz2f/xpOb7XD3rtIaSGjW3pq1L4neKtWo3jiMB4Ig7EAQZYe9p+TnT2nW3pq1L4neKtWQ3kp9zw6gccreswNoEMIOBFFK2M3sbjP7o5ntN7NHy+ghj5kdNLPdZraz7PXpsjX0Bsxsz7BtbWa22cw+yX6PuMZeSb09YWZ92Wu308zuKam3GWb2ezPbZ2Z7zeyH2fZSX7tEXw153Rr+nt3Mxkr6k6Qlkg5Jel/SCnff19BGcpjZQUld7l76FzDM7E5Jf5W0zt3/Ltv2b5KOu/sz2f8oJ7v7PzdJb09I+mvZy3hnqxV1Dl9mXNIySd9Via9doq8H1IDXrYw9+0JJ+939gLuflfRrSd0l9NH03H2bpONf29wtaW3291oN/WNpuJzemoK797v7h9nfJyVdWGa81Ncu0VdDlBH2aZL+POz2ITXXeu8u6Xdm9oGZrSq7mRF0DFtm67CkjjKbGUHhMt6N9LVlxpvmtatk+fNq8QHdN93h7jdJ+gdJP8gOV5uSD70Ha6a5059Lmq2hNQD7Jf20zGayZcbXS/qRu/9leK3M126EvhryupUR9j5JM4bdnp5tawru3pf9HpC0Qc23FPWRCyvoZr8HSu7nK820jPdIy4yrCV67Mpc/LyPs70uaY2azzGy8pOWSNpXQxzeYWWv2wYnMrFXSUjXfUtSbJK3M/l4paWOJvVykWZbxzltmXCW/dqUvf+7uDf+RdI+GPpH/X0n/UkYPOX39raT/yX72lt2bpJc0dFj3fxr6bOMhSVMkbZX0iaQtktqaqLcXNbS09y4NBauzpN7u0NAh+i5JO7Ofe8p+7RJ9NeR14+uyQBB8QAcEQdiBIAg7EARhB4Ig7EAQhB0IgrADQfw/yvSbk7MFlSAAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2wpSOeX-0y_y",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 299
        },
        "outputId": "87cd64d0-f646-443a-e19f-17a7999339b1"
      },
      "source": [
        "img, label = test_dataset[1]\n",
        "plt.imshow(img[0], cmap='gray')\n",
        "print('Shape:', img.shape)\n",
        "print('Label:', label-1)"
      ],
      "execution_count": 105,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Shape: torch.Size([1, 28, 28])\n",
            "Label: 0\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vd2KPSmf2C2E",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 299
        },
        "outputId": "128ec10b-2e6f-467d-bb6f-794166046b7d"
      },
      "source": [
        "img, label = test_dataset[3000]\n",
        "plt.imshow(img[0], cmap='gray')\n",
        "print('Shape:', img.shape)\n",
        "print('Label:', label-1)"
      ],
      "execution_count": 106,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Shape: torch.Size([1, 28, 28])\n",
            "Label: 3\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "y-_KQ5UTmQhF",
        "colab_type": "text"
      },
      "source": [
        "We are adding an extra dimension, with the value 1, which the model considers as a batch with 1 image."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "146fZB2t2Gp1",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "2b4e21da-2671-443c-cddd-8e9d2799b8ed"
      },
      "source": [
        "img.unsqueeze(0).shape"
      ],
      "execution_count": 107,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "torch.Size([1, 1, 28, 28])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 107
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tN4h4Au02RK2",
        "colab_type": "text"
      },
      "source": [
        "#### Test the model on the test dataset\n",
        "We predict the output using the method described above, by considering the outputs as probabilities and the class with the highest probability as the prediction."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xPmc0Q9z2QPj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def predict_image(img, model):\n",
        "    xb = to_device(img.unsqueeze(0), device)\n",
        "    yb = model(xb)\n",
        "    _, preds  = torch.max(yb, dim=1)\n",
        "    return preds[0].item()"
      ],
      "execution_count": 108,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NGqWA26g2cqX",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "dd076aa5-cd51-41f7-fda9-8da89decf08c"
      },
      "source": [
        "img, label = test_dataset[0]\n",
        "plt.imshow(img[0], cmap='gray')\n",
        "print('Label:', label, ', Predicted Label:', predict_image(img, model) + 1, ', Predicted Alphabet:', classes[predict_image(img, model)])"
      ],
      "execution_count": 109,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Label: 1 , Predicted Label: 1 , Predicted Alphabet: a\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Wl5FutNG2hQ1",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "5f33c449-eb34-435c-e77d-de85e2e5a595"
      },
      "source": [
        "img, label = test_dataset[30]\n",
        "plt.imshow(img[0], cmap='gray')\n",
        "print('Label:', label, ', Predicted Label:', predict_image(img, model) + 1, ', Predicted Alphabet:', classes[predict_image(img, model)])"
      ],
      "execution_count": 110,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Label: 1 , Predicted Label: 1 , Predicted Alphabet: a\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAOL0lEQVR4nO3db4hddX7H8c/HREUTibHBGLMa/xAQLTarg1QqxboY/z0wgqgRxaI24j9c2QcVfbDiIynuLn20MItitmyzru5GjUprKkpahMVRbP52VitJTEicrGLMaqJO5tsHc7Kd6Nzfmdz/zvf9guHee773zP1ymM+cc8/vnvtzRAjA9HdUrxsA0B2EHUiCsANJEHYgCcIOJDGzmy9mm1P/QIdFhCdb3tKe3faVtodtv2/7oVZ+F4DOcrPj7LZnSPqDpMsl7ZD0lqTlEbG5sA57dqDDOrFnv0jS+xHxQUR8JenXkq5t4fcB6KBWwr5Q0ocTHu+olh3G9grbQ7aHWngtAC3q+Am6iBiUNChxGA/0Uit79p2STpvw+HvVMgB9qJWwvyVpse0zbR8j6SZJL7anLQDt1vRhfESM2r5P0r9LmiHpqYjY1LbO8J0wc2bz7wTHxsZaquPIND301tSL8Z592iHs/acjH6oB8N1B2IEkCDuQBGEHkiDsQBKEHUiiq9ezo/uOPfbYYv28884r1i+77LJi/aabbirWZ8+e3bD25ptvFte95557ivUDBw4U6zgce3YgCcIOJEHYgSQIO5AEYQeSIOxAElz1Ng0cf/zxDWvXXXddcd3HHnusWD/llFOK9aOPPrpYtye9AEuS9NVXXxXXvfnmm4v1l156qVg/ePBgsT5dcdUbkBxhB5Ig7EAShB1IgrADSRB2IAnCDiTBJa7fAfPmzSvW77rrroa122+/vbjuwoXfmrHrMMPDw8X6888/X6zPnTu3Ya1uHP2WW24p1usukd2zZ0+xng17diAJwg4kQdiBJAg7kARhB5Ig7EAShB1IgnH2PjBjxoxi/YknnijWb7zxxoa13bt3N72uJK1du7ZY/+KLL4r1U089tWHtnHPOKa57/vnnF+tz5swp1hlnP1xLYbe9VdI+SQcljUbEQDuaAtB+7diz/11E/LENvwdAB/GeHUii1bCHpFdtv217xWRPsL3C9pDtoRZfC0ALWj2MvyQidto+WdJa2/8TEesmPiEiBiUNSnzhJNBLLe3ZI2JndTsiabWki9rRFID2azrstmfZPuHQfUlLJW1sV2MA2quVw/j5klZX3ws+U9K/RsS/taWrZBYtWlSsX3XVVcX66Ohow9rTTz9dXLfT371+zDHHNKzVXaffzTkNMmg67BHxgaS/amMvADqIoTcgCcIOJEHYgSQIO5AEYQeS4BLXLjjuuOOK9bvvvrtYL017LEn3339/w9ozzzxTXLfVobW63krDayeeeGJx3ZUrVxbrO3bsKNZxOPbsQBKEHUiCsANJEHYgCcIOJEHYgSQIO5AE4+xdsHjx4mJ96dKlxfqHH35YrK9bt65hbf/+/cV1W3X22WcX6w888EDDWt1XQa9evbpYP3DgQLGOw7FnB5Ig7EAShB1IgrADSRB2IAnCDiRB2IEkGGfvgmXLlhXrdVMXP/LII8X69u3bj7inQ+qutb/ggguK9eeee65Ynz17dsNa3VTUGzZsKNZxZNizA0kQdiAJwg4kQdiBJAg7kARhB5Ig7EASjLN3Qd13q4+NjRXrb7zxRtOvfdZZZxXrdd9Zf8UVVxTrJ598crFeuta+7jvtmbK5vWr37Lafsj1ie+OEZSfZXmv7vep2bmfbBNCqqRzGPy3pym8se0jSaxGxWNJr1WMAfaw27BGxTtIn31h8raRDc/OslFT+PCiAnmv2Pfv8iNhV3d8taX6jJ9peIWlFk68DoE1aPkEXEWG74ZmUiBiUNChJpecB6Kxmh94+sr1Akqrbkfa1BKATmg37i5Juq+7fJumF9rQDoFNqD+Ntr5J0qaR5tndI+rGkxyX9xvYdkrZJuqGTTX7XffbZZ8X66OhosX7mmWcW66V5zu+9997iutdcc02xPnNm+U+k7jMEZ5xxRsPawMBAcd3h4eFinXH4I1Mb9ohY3qD0gzb3AqCD+LgskARhB5Ig7EAShB1IgrADSbibwxdZP0FXd5nps88+W6y/8sorxfonn3zz0oX/d+eddxbX3bJlS7G+adOmYn3JkiXF+uWXX96wVve3d/HFFxfr69evL9aziohJx0PZswNJEHYgCcIOJEHYgSQIO5AEYQeSIOxAEnyVdBfs2rWrWH/99deL9brx6FWrVjWsrVmzprju3r17i/VPP/20WF+0aFGxXpqWue7y2qVLlxbrmzdvLtbrLh3Ohj07kARhB5Ig7EAShB1IgrADSRB2IAnCDiTB9exdcNRR5f+pdePNt956a7H+4IMPNqzt3LmzuG6nXXjhhQ1rddfxl67Tl6Trr7++WN+6dWuxPl1xPTuQHGEHkiDsQBKEHUiCsANJEHYgCcIOJMH17F0wNjZWrL/66qvFet33o4+MjBxxT93y8ccfN6zVjaOfcMIJxXrddNI4XO2e3fZTtkdsb5yw7FHbO22/W/1c3dk2AbRqKofxT0u6cpLlP4uIJdVPecoSAD1XG/aIWCepfLwFoO+1coLuPtvrq8P8uY2eZHuF7SHbQy28FoAWNRv2n0s6W9ISSbsk/aTREyNiMCIGImKgydcC0AZNhT0iPoqIgxExJukXki5qb1sA2q2psNteMOHhdZI2NnougP5QO1Bpe5WkSyXNs71D0o8lXWp7iaSQtFXSXR3scdr78ssvi/Vt27Z1qZP2K11P//LLLxfXXb58ebE+Y8aMpnrKqjbsETHZFn+yA70A6CA+LgskQdiBJAg7kARhB5Ig7EASXCOIjip9VXndpb+zZs0q1k8//fRifXh4uFjPhj07kARhB5Ig7EAShB1IgrADSRB2IAnCDiTBODv6Vt2lv3v27OlSJ9MDe3YgCcIOJEHYgSQIO5AEYQeSIOxAEoQdSIJxdvSt0dHRYn3//v1d6mR6YM8OJEHYgSQIO5AEYQeSIOxAEoQdSIKwA0kwzo6+NWfOnGL93HPPLdb53vjD1e7ZbZ9m+3Xbm21vsv1Atfwk22ttv1fdzu18uwCaNZXD+FFJP4qIcyX9taR7bZ8r6SFJr0XEYkmvVY8B9KnasEfEroh4p7q/T9IWSQslXStpZfW0lZKWdapJAK07ovfsts+Q9H1Jv5c0PyJ2VaXdkuY3WGeFpBXNtwigHaZ8Nt72bEm/lfTDiPhsYi3GZ++bdAa/iBiMiIGIGGipUwAtmVLYbR+t8aD/KiJ+Vy3+yPaCqr5A0khnWgTQDrWH8bYt6UlJWyLipxNKL0q6TdLj1e0LHekQaX3++efF+vbt27vUyfQwlffsfyPpVkkbbL9bLXtY4yH/je07JG2TdENnWgTQDrVhj4j/kuQG5R+0tx0AncLHZYEkCDuQBGEHkiDsQBKEHUiCS1zRt77++utife/evV3qZHpgzw4kQdiBJAg7kARhB5Ig7EAShB1IgrADSTDOjr61b9++Yr1uSmccjj07kARhB5Ig7EAShB1IgrADSRB2IAnCDiTBODt6ZmxsrFhfs2ZNsb579+52tjPtsWcHkiDsQBKEHUiCsANJEHYgCcIOJEHYgSSmMj/7aZJ+KWm+pJA0GBH/bPtRSf8gaU/11Icj4pVONYrvptJY+tDQUHHduvnXDxw40FRPWU3lQzWjkn4UEe/YPkHS27bXVrWfRcQTnWsPQLtMZX72XZJ2Vff32d4iaWGnGwPQXkf0nt32GZK+L+n31aL7bK+3/ZTtuQ3WWWF7yHb5mA1AR0057LZnS/qtpB9GxGeSfi7pbElLNL7n/8lk60XEYEQMRMRAG/oF0KQphd320RoP+q8i4neSFBEfRcTBiBiT9AtJF3WuTQCtqg27bUt6UtKWiPjphOULJjztOkkb298egHZxRJSfYF8i6T8lbZB0aBzlYUnLNX4IH5K2SrqrOplX+l3lFwPQsojwZMtrw95OhB3ovEZh5xN0QBKEHUiCsANJEHYgCcIOJEHYgSQIO5AEYQeSIOxAEoQdSIKwA0kQdiAJwg4kQdiBJLo9ZfMfJW2b8Hhetawf9Wtv/dqXRG/NamdvixoVuno9+7de3B7q1++m69fe+rUvid6a1a3eOIwHkiDsQBK9Dvtgj1+/pF9769e+JHprVld66+l7dgDd0+s9O4AuIexAEj0Ju+0rbQ/bft/2Q73ooRHbW21vsP1ur+enq+bQG7G9ccKyk2yvtf1edTvpHHs96u1R2zurbfeu7at71Ntptl+3vdn2JtsPVMt7uu0KfXVlu3X9PbvtGZL+IOlySTskvSVpeURs7mojDdjeKmkgInr+AQzbfyvpT5J+GRF/WS37J0mfRMTj1T/KuRHxj33S26OS/tTrabyr2YoWTJxmXNIySX+vHm67Ql83qAvbrRd79oskvR8RH0TEV5J+LenaHvTR9yJinaRPvrH4Wkkrq/srNf7H0nUNeusLEbErIt6p7u+TdGia8Z5uu0JfXdGLsC+U9OGExzvUX/O9h6RXbb9te0Wvm5nE/AnTbO2WNL+XzUyidhrvbvrGNON9s+2amf68VZyg+7ZLIuICSVdJurc6XO1LMf4erJ/GTqc0jXe3TDLN+J/1cts1O/15q3oR9p2STpvw+HvVsr4QETur2xFJq9V/U1F/dGgG3ep2pMf9/Fk/TeM92TTj6oNt18vpz3sR9rckLbZ9pu1jJN0k6cUe9PEttmdVJ05ke5akpeq/qahflHRbdf82SS/0sJfD9Ms03o2mGVePt13Ppz+PiK7/SLpa42fk/1fSI73ooUFfZ0n67+pnU697k7RK44d1X2v83MYdkv5C0muS3pP0H5JO6qPe/kXjU3uv13iwFvSot0s0foi+XtK71c/Vvd52hb66st34uCyQBCfogCQIO5AEYQeSIOxAEoQdSIKwA0kQdiCJ/wNG03/z4PZJzwAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GLF7bLD-2lKR",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "c39a48af-6420-4c95-bd90-d2dadd8eb759"
      },
      "source": [
        "img, label = test_dataset[3000]\n",
        "plt.imshow(img[0], cmap='gray')\n",
        "print('Label:', label, ', Predicted Label:', predict_image(img, model) + 1, ', Predicted Alphabet:', classes[predict_image(img, model)])"
      ],
      "execution_count": 119,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Label: 4 , Predicted Label: 4 , Predicted Alphabet: d\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lPaEpjRE23bf",
        "colab_type": "text"
      },
      "source": [
        "#### Evaluate the model on the test dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4VzlRk252uM9",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "74ae902c-27be-49f3-ab31-2de471e9aaf7"
      },
      "source": [
        "test_loader = DeviceDataLoader(DataLoader(test_dataset, batch_size=128), device)\n",
        "result = evaluate(model, test_loader)\n",
        "result"
      ],
      "execution_count": 120,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'val_acc': 0.9428201913833618, 'val_loss': 0.18934908509254456}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 120
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LYiumYJs3NAk",
        "colab_type": "text"
      },
      "source": [
        "#### The final accuracy of the model on the test dataset is 94.28% using a convolutional neural network."
      ]
    }
  ]
}