MNIST CNN.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "eb8a2ff3",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from PIL import Image\n",
    "import pandas as pd\n",
    "\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "679a731c",
   "metadata": {},
   "outputs": [],
   "source": [
    "img = Image.open(\"datasets/images/street.jpg\").convert('RGB')\n",
    "img"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "e1d20b5a",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from torchvision import transforms as tf\n",
    "import torchvision\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2cd416d2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Transform images to tensors using transforms\n",
    "transforms = tf.Compose([tf.Resize(400), \n",
    "                        tf.ToTensor()])\n",
    "img_tensor = transforms(img)\n",
    "img_tensor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "aeac1e4b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>label</th>\n",
       "      <th>1x1</th>\n",
       "      <th>1x2</th>\n",
       "      <th>1x3</th>\n",
       "      <th>1x4</th>\n",
       "      <th>1x5</th>\n",
       "      <th>1x6</th>\n",
       "      <th>1x7</th>\n",
       "      <th>1x8</th>\n",
       "      <th>1x9</th>\n",
       "      <th>...</th>\n",
       "      <th>28x19</th>\n",
       "      <th>28x20</th>\n",
       "      <th>28x21</th>\n",
       "      <th>28x22</th>\n",
       "      <th>28x23</th>\n",
       "      <th>28x24</th>\n",
       "      <th>28x25</th>\n",
       "      <th>28x26</th>\n",
       "      <th>28x27</th>\n",
       "      <th>28x28</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>9</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 785 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   label  1x1  1x2  1x3  1x4  1x5  1x6  1x7  1x8  1x9  ...  28x19  28x20  \\\n",
       "0      5    0    0    0    0    0    0    0    0    0  ...      0      0   \n",
       "1      0    0    0    0    0    0    0    0    0    0  ...      0      0   \n",
       "2      4    0    0    0    0    0    0    0    0    0  ...      0      0   \n",
       "3      1    0    0    0    0    0    0    0    0    0  ...      0      0   \n",
       "4      9    0    0    0    0    0    0    0    0    0  ...      0      0   \n",
       "\n",
       "   28x21  28x22  28x23  28x24  28x25  28x26  28x27  28x28  \n",
       "0      0      0      0      0      0      0      0      0  \n",
       "1      0      0      0      0      0      0      0      0  \n",
       "2      0      0      0      0      0      0      0      0  \n",
       "3      0      0      0      0      0      0      0      0  \n",
       "4      0      0      0      0      0      0      0      0  \n",
       "\n",
       "[5 rows x 785 columns]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\"\"\" Dimensions of Convolution output\n",
    "Output = ((Input - kernel_dim + 2 * pad)/ Stride_size) +1\n",
    "\"\"\"\n",
    "mnist_train = pd.read_csv('datasets/mnist-in-csv/mnist_train.csv')\n",
    "mnist_test = pd.read_csv('datasets/mnist-in-csv/mnist_test.csv')\n",
    "mnist_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "f4d5794f",
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_train = mnist_train.dropna()\n",
    "mnist_test = mnist_test.dropna()\n",
    "\n",
    "# Display Some\n",
    "random_sel = mnist_train.sample(8)\n",
    "image_features = random_sel.drop('label', axis =1)\n",
    "image_batch = (torch.Tensor(image_features.values / 255.)).reshape((-1, 28, 28))\n",
    "grid = torchvision.utils.make_grid(image_batch.unsqueeze(1), nrow=8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "1afd198e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(-0.5, 241.5, 31.5, -0.5)"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x864 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure (figsize = (12, 12))\n",
    "\n",
    "plt.imshow(grid.numpy().transpose((1, 2, 0)))\n",
    "\n",
    "plt.axis('off')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "ca49195e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([60000, 784])\n",
      "torch.Size([60000])\n",
      "torch.Size([10000, 784])\n",
      "torch.Size([10000])\n"
     ]
    }
   ],
   "source": [
    "# Create training and test set tensors\n",
    "mnist_train_features = mnist_train.drop('label', axis =1)\n",
    "mnist_train_target = mnist_train['label']\n",
    "\n",
    "mnist_test_features = mnist_test.drop('label', axis =1)\n",
    "mnist_test_target = mnist_test['label']\n",
    "\n",
    "X_train_tensor = torch.tensor(mnist_train_features.values, dtype=torch.float)\n",
    "x_test_tensor  = torch.tensor(mnist_test_features.values, dtype=torch.float) \n",
    "\n",
    "Y_train_tensor = torch.tensor(mnist_train_target.values, dtype=torch.long)\n",
    "y_test_tensor  = torch.tensor(mnist_test_target.values, dtype=torch.long)\n",
    "\n",
    "print(X_train_tensor.shape)\n",
    "print(Y_train_tensor.shape)\n",
    "print(x_test_tensor.shape)\n",
    "print(y_test_tensor.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "d6f65240",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([60000, 1, 28, 28])\n",
      "torch.Size([60000])\n",
      "torch.Size([10000, 1, 28, 28])\n",
      "torch.Size([10000])\n"
     ]
    }
   ],
   "source": [
    "# Reshape (batch, RGB or grayscale , height, width)/ It is pytorch's shape\n",
    "X_train_tensor = X_train_tensor.reshape(-1, 1, 28, 28)\n",
    "x_test_tensor = x_test_tensor.reshape(-1, 1, 28, 28)\n",
    "print(X_train_tensor.shape)\n",
    "print(Y_train_tensor.shape)\n",
    "print(x_test_tensor.shape)\n",
    "print(y_test_tensor.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "86ef8737",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Parameters\n",
    "in_size = 1\n",
    "hid1_size = 16 #Re-run for 32\n",
    "hid2_size = 32 #Re-run for 64\n",
    "out_size = 10\n",
    "k_conv_size = 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "6aceb2ae",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ConvNet(nn.Module):\n",
    "    \n",
    "    def __init__(self):\n",
    "        super(ConvNet, self).__init__()\n",
    "        \n",
    "        self.layer1 = nn.Sequential(\n",
    "            nn.Conv2d(in_size, hid1_size, k_conv_size),\n",
    "            nn.BatchNorm2d(hid1_size),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(kernel_size=2))\n",
    "        \n",
    "        self.layer2 = nn.Sequential(\n",
    "            nn.Conv2d(hid1_size, hid2_size, k_conv_size),\n",
    "            nn.BatchNorm2d(hid2_size),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(kernel_size=2))\n",
    "        \n",
    "        self.fc = nn.Linear(512, out_size)\n",
    "        \n",
    " \n",
    "    def forward(self, x):\n",
    "        out = self.layer1(x)\n",
    "        #print(out.shape)\n",
    "        \n",
    "        out = self.layer2(out)\n",
    "        #print(out.shape)\n",
    "        \n",
    "        out = out.reshape(out.size(0), -1)\n",
    "        #print(out.shape)\n",
    "        \n",
    "        out = self.fc(out)\n",
    "        #print(out.shape)\n",
    "        \n",
    "        ## F.log_softmax(out, dim=-1)\n",
    "        \n",
    "        return out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "45ae9fdb",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cpu\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "ConvNet(\n",
       "  (layer1): Sequential(\n",
       "    (0): Conv2d(1, 16, kernel_size=(5, 5), stride=(1, 1))\n",
       "    (1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU()\n",
       "    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  )\n",
       "  (layer2): Sequential(\n",
       "    (0): Conv2d(16, 32, kernel_size=(5, 5), stride=(1, 1))\n",
       "    (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU()\n",
       "    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  )\n",
       "  (fc): Linear(in_features=512, out_features=10, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = ConvNet()\n",
    "\n",
    "# Set pytorch device\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(device)\n",
    "\n",
    "model.to(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "2dbd3dc7",
   "metadata": {},
   "outputs": [],
   "source": [
    "# All needed tensors for training must be in the device\n",
    "X_train_tensor = X_train_tensor.to(device)\n",
    "x_test_tensor  = x_test_tensor.to(device) \n",
    "\n",
    "Y_train_tensor = Y_train_tensor.to(device)\n",
    "y_test_tensor  = y_test_tensor.to(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "b239ce61",
   "metadata": {},
   "outputs": [],
   "source": [
    "learning_rate = 0.001 \n",
    "\n",
    "#Loss function\n",
    "criterion = nn.CrossEntropyLoss() \n",
    "#nn.NLLLoss() \n",
    "\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate) \n",
    "#optimizer =torch.optim.SGD(model.parameters(), lr=learning_rate, momentum=0.9) \n",
    "                    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ec724d0f",
   "metadata": {},
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "fc57eca6",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 1, loss - 2.49831 \n",
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 2, loss - 2.11666 \n",
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 3, loss - 1.87246 \n",
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 4, loss - 1.68898 \n",
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 5, loss - 1.53617 \n",
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 6, loss - 1.40442 \n",
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 7, loss - 1.29018 \n",
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 8, loss - 1.19029 \n",
      "torch.Size([60000, 16, 12, 12])\n",
      "torch.Size([60000, 32, 4, 4])\n",
      "torch.Size([60000, 512])\n",
      "torch.Size([60000, 10])\n",
      "Epoch - 9, loss - 1.10193 \n"
     ]
    }
   ],
   "source": [
    "num_epochs = 10\n",
    "loss_values = list()\n",
    "\n",
    "for epoch in range(1, num_epochs):\n",
    "        \n",
    "        outputs = model(X_train_tensor)\n",
    "        loss = criterion(outputs,Y_train_tensor)\n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "    \n",
    "        print('Epoch - %d, loss - %0.5f '%(epoch, loss.item()))\n",
    "        loss_values.append(loss.item())\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fd9eecae",
   "metadata": {},
   "source": [
    "# Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "21457c96",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, 'Loss')"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = (range(0, 9))\n",
    "\n",
    "plt.figure(figsize = (8, 8))\n",
    "plt.plot(x, loss_values)\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Loss')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "595be591",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ConvNet(\n",
       "  (layer1): Sequential(\n",
       "    (0): Conv2d(1, 16, kernel_size=(5, 5), stride=(1, 1))\n",
       "    (1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU()\n",
       "    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  )\n",
       "  (layer2): Sequential(\n",
       "    (0): Conv2d(16, 32, kernel_size=(5, 5), stride=(1, 1))\n",
       "    (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU()\n",
       "    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  )\n",
       "  (fc): Linear(in_features=512, out_features=10, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.eval() #also turns off batch norm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "25a4a692",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import accuracy_score, precision_score, recall_score\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "9300627d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([10000, 16, 12, 12])\n",
      "torch.Size([10000, 32, 4, 4])\n",
      "torch.Size([10000, 512])\n",
      "torch.Size([10000, 10])\n",
      "Accuracy:  0.7838\n",
      "Precision:  0.8102704377352453\n",
      "Recall:  0.7838\n"
     ]
    }
   ],
   "source": [
    "with torch.no_grad():\n",
    "    \n",
    "    correct = 0\n",
    "    total = 0\n",
    "    \n",
    "    outputs = model(x_test_tensor)\n",
    "    _, predicted = torch.max(outputs.data, 1)\n",
    "    \n",
    "    y_test = y_test_tensor.cpu().numpy()\n",
    "    predicted = predicted.cpu()\n",
    "    \n",
    "    print(\"Accuracy: \", accuracy_score(predicted, y_test))\n",
    "    print(\"Precision: \", precision_score(predicted, y_test, average='weighted'))\n",
    "    print(\"Recall: \", recall_score(predicted, y_test, average='weighted'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bb1028d9",
   "metadata": {},
   "source": [
    "# Future Predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "a87aced1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sample target data =  9\n"
     ]
    }
   ],
   "source": [
    "print(\"sample target data = \", mnist_test_target.values[1005])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "693dbb15",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x267d892cdf0>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWgAAAFlCAYAAADGe3ILAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAAQvUlEQVR4nO3dbYxc5XnG8etivTapX1pTwBhjXkJNFRqKIVunAZQ6pRCw1Bik0GAV6kipTFWQoKVtEB8aFKmUpoFETVoiI1xMSqBRwWCpKMVyIDRx4rKmLthsKS41ZLFjhwCCFLC967sf9rhaO7vseXZmPLdn/j/J2tkz9555DgN/jsczx44IAQDyOardCwAAjI1AA0BSBBoAkiLQAJAUgQaApAg0ACQ15XA+2FRPi6M1/XA+JACk9q7+V3tjj8e677AG+mhN14d94eF8SABIbWOsH/e+hl7isH2J7edtb7N9UyP7AgAcbNKBtt0j6W8lXSrpTEnLbJ/ZrIUBQLdr5Ax6kaRtEfFiROyV9ICkpc1ZFgCgkUDPk/TDUd8PVtsOYnuF7X7b/fu0p4GHA4Du0kigx/pTx5+58lJErIyIvojo69W0Bh4OALpLI4EelDR/1PcnSdrR2HIAAAc0EuinJC2wfZrtqZKulLS2OcsCAEz6fdARMWT7Okn/IqlH0qqI2Nq0lQFAl2vogyoR8aikR5u0FgDAKFyLAwCSItAAkBSBBoCkCDQAJEWgASApAg0ASRFoAEiKQANAUgQaAJIi0ACQFIEGgKQINAAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUgQaApAg0ACRFoAEgKQINAEkRaABIikADQFIEGgCSItAAkBSBBoCkCDQAJEWgASApAg0ASRFoAEiKQANAUgQaAJIi0ACQFIEGgKQINAAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUgQaApAg0ACRFoAEgKQINAEkRaABIikADQFIEGgCSmtLID9veLuktScOShiKirxmLAgA0GOjKxyLi1SbsBwAwCi9xAEBSjQY6JD1me5PtFc1YEABgRKMvcZwfETtsHy9pne3/jIgnRw9U4V4hSUfr5xp8OADoHg2dQUfEjurrbklrJC0aY2ZlRPRFRF+vpjXycADQVSYdaNvTbc88cFvSxZK2NGthANDtGnmJY46kNbYP7OcbEfGtpqwKADD5QEfEi5LObuJaAACjNON90EAK/tCv1J79n8tmFe37zz65pmj+C89cXHv25L/rKdr3Ud/596J5HLl4HzQAJEWgASApAg0ASRFoAEiKQANAUgQaAJIi0ACQFIEGgKQINAAkRaABICkCDQBJcS0OHDY9H1hQND/wR79QNH//b32t9uyvTXPRvkt9+oJ7as/+5Lx3ivZ91af+sGjeG/6jaB55cAYNAEkRaABIikADQFIEGgCSItAAkBSBBoCkCDQAJEWgASApAg0ASRFoAEiKj3qjIT1nnF579tZH/6Fo32dN7S1cTf2Pb9/4o0VFe35sTdn83jPfrj279TfuKtr31Ft3F83vW1w0jkQ4gwaApAg0ACRFoAEgKQINAEkRaABIikADQFIEGgCSItAAkBSBBoCkCDQAJEWgASAprsWBg5RcW0OSvrzu3tqzp095X9G+n3i37Focf37T79eenblmU9G+5w9tKJov8Zvf+lTR/OcXPFI0f/uxi4vmh1/9SdE8WoczaABIikADQFIEGgCSItAAkBSBBoCkCDQAJEWgASApAg0ASRFoAEiKQANAUgQaAJLiWhw4yMCNxxTNl1xf4/JtS4r2PfQHM4vmZwxsrD0bRXuWjpo+vWj+pRvOrj171Ynri/a9+Oh9RfN/fcoJRfPiWhxpcAYNAEkRaABIasJA215le7ftLaO2HWN7ne0Xqq+zW7tMAOg+dc6g75F0ySHbbpK0PiIWSFpffQ8AaKIJAx0RT0p67ZDNSyWtrm6vlnRZc5cFAJjsa9BzImKnJFVfjx9v0PYK2/22+/dpzyQfDgC6T8v/kDAiVkZEX0T09Wpaqx8OADrGZAO9y/ZcSaq+7m7ekgAA0uQDvVbS8ur2ckllf4slAGBCdd5md7+k70v6ZduDtj8j6TZJF9l+QdJF1fcAgCaa8KPeEbFsnLsubPJakMDvfvgHRfOv73+n9uzQ0vqzkjT8xo+K5lvp+VvPKpp/4ZNfbdFKpJt3n1s0763/XTRf+jF4tA6fJASApAg0ACRFoAEgKQINAEkRaABIikADQFIEGgCSItAAkBSBBoCkCDQAJEWgASCpCa/FgSNbz+yyvy7yghkbiuZ/e8vv1Z6d9UbZNSFaacrcE4rmH/jEVwofoadwvr5H1p5XNH/Ku2XPKfLgDBoAkiLQAJAUgQaApAg0ACRFoAEgKQINAEkRaABIikADQFIEGgCSItAAkBSBBoCkuBZHhxt+/fWi+Zf3/WLR/FtPzKk9O0utvRZHz6xZtWd//p/2FO37Q1Nbd22NIQ0XzZ/4vb0tWgmy4QwaAJIi0ACQFIEGgKQINAAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUH/XGQZ57+8Si+auvWld79tt/Nb1o3z2zZxfND/zlgtqz2079WtG+W+lPd55XNN/7WH+LVoJsOIMGgKQINAAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUgQaApAg0ACRFoAEgKa7FgYM8sXpR0fxTn/1K7dmPvNhTtO9Z3lM0f9bU9bVnX9//TtG+/3jw0qL5vz/5idqz3375jKJ9z9PWonkcuTiDBoCkCDQAJDVhoG2vsr3b9pZR226x/YrtzdWvJa1dJgB0nzpn0PdIumSM7V+KiIXVr0ebuywAwISBjognJb12GNYCABilkdegr7P9TPUSyLh/9YXtFbb7bffvU9mfygNAN5tsoO+UdLqkhZJ2Srp9vMGIWBkRfRHR16tpk3w4AOg+kwp0ROyKiOGI2C/pLkllb54FAExoUoG2PXfUt5dL2jLeLABgcib8JKHt+yUtlnSs7UFJn5O02PZCSSFpu6RrWrdEAOhOEwY6IpaNsfnuFqwFADAK1+LAQeb8zYai+cVLrqg9+71ffaho38PRWzRfcn2N877+J0X7Hpqxv2i+55Qna8++PTijaN/oHnzUGwCSItAAkBSBBoCkCDQAJEWgASApAg0ASRFoAEiKQANAUgQaAJIi0ACQFIEGgKS4FgcaMmPJS7VnP/qJFUX7HvxY2fnDL/1j/WtxnLbh+0X7/uCmsrUMR/1rd8zewnkSxsa/GQCQFIEGgKQINAAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUgQaApPioNxqzf7j26Pse/reiXS94uHAtQIfhDBoAkiLQAJAUgQaApAg0ACRFoAEgKQINAEkRaABIikADQFIEGgCSItAAkBSBBoCkuBYHutJRM2cWzZ8747kWrQQYH2fQAJAUgQaApAg0ACRFoAEgKQINAEkRaABIikADQFIEGgCSItAAkBSBBoCkCDQAJMW1ONCd3n9S0fiVM77TooUA4+MMGgCSmjDQtufbftz2gO2ttq+vth9je53tF6qvs1u/XADoHnXOoIck3RgRH5D065KutX2mpJskrY+IBZLWV98DAJpkwkBHxM6IeLq6/ZakAUnzJC2VtLoaWy3pshatEQC6UtFr0LZPlXSOpI2S5kTETmkk4pKOH+dnVtjut92/T3saXC4AdI/agbY9Q9KDkm6IiDfr/lxErIyIvojo69W0yawRALpSrUDb7tVInO+LiIeqzbtsz63unytpd2uWCADdqc67OCzpbkkDEXHHqLvWSlpe3V4u6ZHmLw8AuledD6qcL+lqSc/a3lxtu1nSbZK+afszkl6WdEVLVggAXWrCQEfEdyV5nLsvbO5yAAAH8FFvdKXBj+f5XNUJ//xS0fxQi9aBfPioNwAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUgQaApAg0ACRFoAEgKQINAElxLQ50peM27233Ev7f/tffaPcSkBRn0ACQFIEGgKQINAAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUgQaApAg0ACTFtTjQlX68cGq7lwBMiDNoAEiKQANAUgQaAJIi0ACQFIEGgKQINAAkRaABICkCDQBJEWgASIpAA0BSfNQbXWnqG9HuJQAT4gwaAJIi0ACQFIEGgKQINAAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUgQaApLgWB7rScaueKpr//LVnFc3f2/+R2rNnvLOpaN/oHpxBA0BSEwba9nzbj9sesL3V9vXV9ltsv2J7c/VrSeuXCwDdo85LHEOSboyIp23PlLTJ9rrqvi9FxBdbtzwA6F4TBjoidkraWd1+y/aApHmtXhgAdLui16BtnyrpHEkbq03X2X7G9irbs5u9OADoZrUDbXuGpAcl3RARb0q6U9LpkhZq5Az79nF+boXtftv9+7Sn8RUDQJeoFWjbvRqJ830R8ZAkRcSuiBiOiP2S7pK0aKyfjYiVEdEXEX29mtasdQNAx6vzLg5LulvSQETcMWr73FFjl0va0vzlAUD3qvMujvMlXS3pWdubq203S1pme6GkkLRd0jUtWB8AdK067+L4riSPcdejzV8OAOAAPkkIAElxLQ50pRgaKpr/wdm9RfNnqL9oHhgLZ9AAkBSBBoCkCDQAJEWgASApAg0ASRFoAEiKQANAUgQaAJIi0ACQFIEGgKQINAAkRaABICkCDQBJEWgASIpAA0BSBBoAkiLQAJAUgQaApAg0ACRFoAEgKQINAEkRaABIyhFx+B7M/rGkl8a461hJrx62hbQPx9l5uuVYOc7WOSUijhvrjsMa6PHY7o+Ivnavo9U4zs7TLcfKcbYHL3EAQFIEGgCSyhLole1ewGHCcXaebjlWjrMNUrwGDQD4WVnOoAEAh2hroG1fYvt529ts39TOtbSa7e22n7W92XZ/u9fTLLZX2d5te8uobcfYXmf7herr7HausRnGOc5bbL9SPaebbS9p5xqbwfZ824/bHrC91fb11faOek7f4zhTPadte4nDdo+k/5J0kaRBSU9JWhYRz7VlQS1me7ukvojoqPeS2v6opJ9KujciPlht+4Kk1yLitup/vLMj4rPtXGejxjnOWyT9NCK+2M61NZPtuZLmRsTTtmdK2iTpMkmfVgc9p+9xnL+jRM9pO8+gF0naFhEvRsReSQ9IWtrG9WASIuJJSa8dsnmppNXV7dUa+Rf/iDbOcXaciNgZEU9Xt9+SNCBpnjrsOX2P40ylnYGeJ+mHo74fVMJ/QE0Ukh6zvcn2inYvpsXmRMROaeQ/BEnHt3k9rXSd7Weql0CO6N/2H8r2qZLOkbRRHfycHnKcUqLntJ2B9hjbOvktJedHxLmSLpV0bfVbZhzZ7pR0uqSFknZKur2tq2ki2zMkPSjphoh4s93raZUxjjPVc9rOQA9Kmj/q+5Mk7WjTWlouInZUX3dLWqORl3g61a7qNb4Dr/XtbvN6WiIidkXEcETsl3SXOuQ5td2rkWjdFxEPVZs77jkd6zizPaftDPRTkhbYPs32VElXSlrbxvW0jO3p1R9EyPZ0SRdL2vLeP3VEWytpeXV7uaRH2riWljkQrMrl6oDn1LYl3S1pICLuGHVXRz2n4x1ntue0rR9Uqd7C8mVJPZJWRcRftG0xLWT7/Ro5a5akKZK+0SnHavt+SYs1chWwXZI+J+lhSd+UdLKklyVdERFH9B+wjXOcizXyW+GQtF3SNQdepz1S2b5A0r9KelbS/mrzzRp5fbZjntP3OM5lSvSc8klCAEiKTxICQFIEGgCSItAAkBSBBoCkCDQAJEWgASApAg0ASRFoAEjq/wCMka2ppH9UdAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sample_img = mnist_test_features.values[1005]\n",
    "sample_img = sample_img.reshape(1, 28, 28)\n",
    "\n",
    "sample_img = sample_img[0, :, :]\n",
    "\n",
    "plt.figure(figsize =(6, 6))\n",
    "plt.imshow(sample_img)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "bdcd0015",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([1, 16, 12, 12])\n",
      "torch.Size([1, 32, 4, 4])\n",
      "torch.Size([1, 512])\n",
      "torch.Size([1, 10])\n",
      "tensor([[-0.4880, -0.9562,  0.1608, -0.5884,  1.0530, -0.0247, -0.9057,  0.2945,\n",
      "          1.9283,  2.4567]], grad_fn=<AddmmBackward0>)\n",
      " The predicted label is :  9\n"
     ]
    }
   ],
   "source": [
    "sample = np.array(mnist_test_features.values[1005]) \n",
    "\n",
    "# create tensor and send it to device\n",
    "sample_tensor = torch.from_numpy(sample).float()\n",
    "sample_tensor = sample_tensor.reshape(-1, 1, 28, 28)\n",
    "sample_tensor = sample_tensor.to(device) \n",
    "\n",
    "y_pred = model(sample_tensor)\n",
    "print(y_pred)\n",
    "\n",
    "_, predicted = torch.max(y_pred.data, -1)\n",
    "\n",
    "print (\" The predicted label is : \", predicted.item())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bc524e20",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}