CNN/MNIST-digit-recognition-using-CNN.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "d9538537",
   "metadata": {},
   "source": [
    "## Architecture\n",
    "- Input : 28 * 28 * 1\n",
    "- Convolution layer 1 with 32 filters - Output: 28 * 28 * 32 (32 filter maps)\n",
    "- Max pooling with 2 * 2 matrix - Output: 14 * 14 * 32 (Shrinked)\n",
    "- Convolution layer 2 with 64 filters of depth 32 - Output: 14 * 14 * 64\n",
    "- Max pooling with 2 * 2 matrix - Output: 7 * 7 * 64\n",
    "- FC1 \n",
    "- FC2 \n",
    "- Softmax"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "54cbc0a8",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/muhammadwaseem/miniconda3/envs/torch/lib/python3.9/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: dlopen(/Users/muhammadwaseem/miniconda3/envs/torch/lib/python3.9/site-packages/torchvision/image.so, 0x0006): Symbol not found: __ZN2at4_ops19empty_memory_format4callEN3c108ArrayRefIxEENS2_8optionalINS2_10ScalarTypeEEENS5_INS2_6LayoutEEENS5_INS2_6DeviceEEENS5_IbEENS5_INS2_12MemoryFormatEEE\n",
      "  Referenced from: <CAE66874-17C2-35C9-9C4D-6BA9770AF17F> /Users/muhammadwaseem/miniconda3/envs/torch/lib/python3.9/site-packages/torchvision/image.so\n",
      "  Expected in:     <459875AA-DE2C-366B-9C44-90D4B3887080> /Users/muhammadwaseem/miniconda3/envs/torch/lib/python3.9/site-packages/torch/lib/libtorch_cpu.dylib\n",
      "  warn(f\"Failed to load image Python extension: {e}\")\n"
     ]
    }
   ],
   "source": [
    "import torch \n",
    "import torch.nn as nn\n",
    "import torchvision\n",
    "from torchvision import transforms\n",
    "from torch.utils.data import Subset\n",
    "from torch.utils.data import DataLoader, TensorDataset\n",
    "from torch.optim import Adam"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "90394714",
   "metadata": {},
   "outputs": [],
   "source": [
    "mps_device = torch.device(\"mps\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0cda754e",
   "metadata": {},
   "source": [
    "## Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "f09196ec",
   "metadata": {},
   "outputs": [],
   "source": [
    "image_path = './'\n",
    "mnist_data = torchvision.datasets.MNIST(root = image_path,\n",
    "                                        train = True,\n",
    "                                        transform = transforms.ToTensor(),\n",
    "                                        download = True)\n",
    "mnist_test_data = torchvision.datasets.MNIST(root = image_path, \n",
    "                                             train = False,\n",
    "                                             transform = transforms.ToTensor(),\n",
    "                                             download = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 223,
   "id": "bdf38163",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.uint8"
      ]
     },
     "execution_count": 223,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mnist_data.data.dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "a391c636",
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_valid_dataset = Subset(mnist_data, torch.arange(10000)) \n",
    "mnist_train_dataset = Subset(mnist_data, torch.arange(10000, len(mnist_data)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 221,
   "id": "19dc898d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.float32"
      ]
     },
     "execution_count": 221,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mnist_valid_dataset[1][0].dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "6d01cac8",
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 64\n",
    "torch.manual_seed(1) # To produce same set of 64 batch on each time when we run \n",
    "data = DataLoader(mnist_train_dataset, batch_size=batch_size, shuffle=True)\n",
    "\n",
    "# Why to shuffle validation?? What is the difference if I dont batch and calculate validation performance\n",
    "# entirely on the validation set after each epoch ?\n",
    "data_vl = DataLoader(mnist_valid_dataset, batch_size=batch_size, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "b62bc869",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([64, 1, 28, 28])"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "next(iter(data))[0].shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8637465b",
   "metadata": {},
   "source": [
    "## Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "d1af417a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (conv1): Conv2d(1, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))\n",
       "  (relu1): ReLU()\n",
       "  (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  (conv2): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))\n",
       "  (relu2): ReLU()\n",
       "  (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  (flatten): Flatten(start_dim=1, end_dim=-1)\n",
       "  (fc1): Linear(in_features=3136, out_features=1024, bias=True)\n",
       "  (relu3): ReLU()\n",
       "  (dropout): Dropout(p=0.5, inplace=False)\n",
       "  (fc2): Linear(in_features=1024, out_features=10, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = nn.Sequential()\n",
    "model.add_module('conv1', \n",
    "                 nn.Conv2d( in_channels=1, out_channels=32,\n",
    "                 kernel_size=5, padding=2)) # Padding=2 which makes it as `same mode`\n",
    "model.add_module('relu1', nn.ReLU())\n",
    "model.add_module('pool1', nn.MaxPool2d(kernel_size=2)) # default: stride=kernel_size=2\n",
    "model.add_module('conv2',\n",
    "                 nn.Conv2d( in_channels=32, out_channels=64,\n",
    "                 kernel_size=5, padding=2))\n",
    "model.add_module('relu2', nn.ReLU())\n",
    "model.add_module('pool2', nn.MaxPool2d(kernel_size=2))\n",
    "model.add_module('flatten', nn.Flatten(1))\n",
    "model.add_module('fc1', nn.Linear(3136, 1024)) # 3136 is determined by inspecting the output of prev layer :)\n",
    "model.add_module('relu3', nn.ReLU())\n",
    "# DROPOUT: leave half of the neuron units while training on each iteration. But batch wise or epoch wise?\n",
    "model.add_module('dropout', nn.Dropout(p=0.5))  # Inverse dropout will be carried out during training phase\n",
    "model.add_module('fc2', nn.Linear(1024, 10))\n",
    "#model.add_module('relu4', nn.ReLU()) ## Mistakely added this lead to accuracy very low\n",
    "model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "9da84926",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = model.to(mps_device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "873a2fc0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([64, 10])"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x = torch.ones((64,1, 28, 28), device=mps_device)\n",
    "model(x).shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bc170603",
   "metadata": {},
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "57112a8b",
   "metadata": {},
   "outputs": [],
   "source": [
    "loss_fn = nn.CrossEntropyLoss()\n",
    "optimizer = Adam(model.parameters(), lr=0.001)\n",
    "\n",
    "def train(model, epoch, data, data_vl):\n",
    "    train_loss = [0] * epoch\n",
    "    train_accuracy = [0] * epoch\n",
    "    validation_loss = [0] * epoch\n",
    "    validation_accuracy = [0] * epoch\n",
    "    for i in range(epoch):\n",
    "        model.train() # to ENABLE dropout while training\n",
    "        for x_batch, y_batch in data:\n",
    "            x_batch = x_batch.to(mps_device)\n",
    "            y_batch = y_batch.to(mps_device)\n",
    "            pred = model(x_batch)\n",
    "            loss = loss_fn(pred, y_batch)\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "            optimizer.zero_grad()\n",
    "            train_loss[i] += loss.item() * x_batch.size(0)\n",
    "            soft_pred = torch.softmax(pred, axis=1) # why is it not done?\n",
    "            \n",
    "            is_crt = (torch.argmax(soft_pred, dim=1) == y_batch).to(torch.float32).sum().cpu()\n",
    "            train_accuracy[i] += is_crt\n",
    "\n",
    "        train_loss[i] /= len(data.dataset)\n",
    "        train_accuracy[i] /= len(data.dataset)\n",
    "        \n",
    "        model.eval() # to AVOID dropout while inference\n",
    "        \n",
    "        # Why dont we calculate the validation error/accuracy after an entire epoch? Is performance a reason?\n",
    "        with torch.no_grad():\n",
    "            for x_batch, y_batch in data_vl:\n",
    "                x_batch = x_batch.to(mps_device)\n",
    "                y_batch = y_batch.to(mps_device)\n",
    "                pred = model(x_batch)\n",
    "                loss = loss_fn(pred, y_batch)\n",
    "                validation_loss[i] += loss.item() * x_batch.size(0)\n",
    "                soft_pred = torch.softmax(pred, axis=1) # why is it not done?\n",
    "                is_crt = (torch.argmax(soft_pred, dim=1) == y_batch).sum().to(torch.float32).cpu()\n",
    "                validation_accuracy[i] += is_crt\n",
    "            validation_loss[i] /= len(data_vl.dataset)\n",
    "            validation_accuracy[i] /= len(data_vl.dataset)\n",
    "        print(f\"Epoch: {i+1}, Accuracy - train: {train_accuracy[i]}, validation: {validation_accuracy[i]}\")\n",
    "    return train_loss, train_accuracy, validation_loss, validation_accuracy\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "31502581",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 1, Accuracy - train: 0.9506199955940247, validation: 0.9811999797821045\n",
      "Epoch: 2, Accuracy - train: 0.9853000044822693, validation: 0.9851999878883362\n",
      "Epoch: 3, Accuracy - train: 0.9902200102806091, validation: 0.982699990272522\n",
      "Epoch: 4, Accuracy - train: 0.9920200109481812, validation: 0.989300012588501\n",
      "Epoch: 5, Accuracy - train: 0.9929999709129333, validation: 0.9883000254631042\n",
      "Epoch: 6, Accuracy - train: 0.9944599866867065, validation: 0.987500011920929\n",
      "Epoch: 7, Accuracy - train: 0.9960600137710571, validation: 0.9896000027656555\n",
      "Epoch: 8, Accuracy - train: 0.9959999918937683, validation: 0.9878000020980835\n",
      "Epoch: 9, Accuracy - train: 0.9963799715042114, validation: 0.9883999824523926\n",
      "Epoch: 10, Accuracy - train: 0.9964600205421448, validation: 0.9912999868392944\n",
      "Epoch: 11, Accuracy - train: 0.9975200295448303, validation: 0.991100013256073\n",
      "Epoch: 12, Accuracy - train: 0.9973800182342529, validation: 0.9900000095367432\n",
      "Epoch: 13, Accuracy - train: 0.9977800250053406, validation: 0.9909999966621399\n",
      "Epoch: 14, Accuracy - train: 0.9973400235176086, validation: 0.9897000193595886\n",
      "Epoch: 15, Accuracy - train: 0.9981399774551392, validation: 0.9919999837875366\n",
      "Epoch: 16, Accuracy - train: 0.9974799752235413, validation: 0.9912999868392944\n",
      "Epoch: 17, Accuracy - train: 0.9979199767112732, validation: 0.9918000102043152\n",
      "Epoch: 18, Accuracy - train: 0.9987800121307373, validation: 0.9916999936103821\n",
      "Epoch: 19, Accuracy - train: 0.998520016670227, validation: 0.9926999807357788\n",
      "Epoch: 20, Accuracy - train: 0.999019980430603, validation: 0.9912999868392944\n"
     ]
    }
   ],
   "source": [
    "torch.manual_seed(1)\n",
    "epoch = 20\n",
    "train_loss, train_accuracy, validation_loss, validation_accuracy = train(model, epoch, data, data_vl)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "f4a7a897",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, '$Accuracy$')"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1200x500 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "plt.figure(figsize=(12,5))\n",
    "plt.subplot(1,2,1)\n",
    "plt.plot(np.arange(epoch), train_loss, '-o', label=\"Train Loss\")\n",
    "plt.plot(np.arange(epoch), validation_loss, '--<', label=\"Validation Loss\")\n",
    "plt.xlabel(\"$Epoch$\")\n",
    "plt.ylabel(\"$Loss$\")\n",
    "\n",
    "plt.subplot(1,2,2)\n",
    "plt.plot(np.arange(epoch), train_accuracy, '-o', label=\"Train Accuracy\")\n",
    "plt.plot(np.arange(epoch), validation_accuracy, '--<', label=\"Validation Accuracy\")\n",
    "plt.xlabel(\"$Epoch$\")\n",
    "plt.ylabel(\"$Accuracy$\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b4ee82f7",
   "metadata": {},
   "source": [
    "## Test data accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 259,
   "id": "107e851c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.uint8"
      ]
     },
     "execution_count": 259,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mnist_test_data.data[0].dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 260,
   "id": "7e6d9bf0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.float32"
      ]
     },
     "execution_count": 260,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mnist_test_data[0][0].dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "7d6bfec5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10000"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(mnist_test_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "id": "7a8ad49f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.9932\n"
     ]
    }
   ],
   "source": [
    "data_test = DataLoader(mnist_test_data, 64, shuffle=True)\n",
    "\n",
    "total_crt = 0\n",
    "with torch.no_grad():\n",
    "    for x_batch, y_batch in data_test:\n",
    "        x_batch = x_batch.to(mps_device)\n",
    "        y_batch = y_batch.to(mps_device)\n",
    "        pred = model(x_batch)\n",
    "        loss = loss_fn(pred, y_batch)\n",
    "        soft_pred = torch.softmax(pred, axis=1)\n",
    "        is_crt = (torch.argmax(soft_pred, axis=1) == y_batch).float().sum()\n",
    "        total_crt += is_crt\n",
    "    print(f\"Accuracy: {total_crt/len(data_test.dataset):.4f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "id": "859708b3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(9932., device='mps:0')"
      ]
     },
     "execution_count": 85,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "total_crt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "id": "a560e914",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([10000, 1, 28, 28])\n",
      "tensor([[-37.6836, -19.1679, -17.8474,  ...,  42.5817, -36.9515, -18.8134],\n",
      "        [-10.7249, -13.9031,  35.2150,  ..., -31.3361, -15.8908, -30.5881],\n",
      "        [-26.1027,  32.3364, -27.1926,  ..., -17.1892, -11.1988, -25.5374],\n",
      "        ...,\n",
      "        [-43.9979, -17.6347, -33.0978,  ..., -25.9862, -14.4071, -18.7628],\n",
      "        [-25.4065, -27.3832, -34.0300,  ..., -22.3598,  -1.8935, -17.1660],\n",
      "        [-10.2998, -21.0205, -21.6235,  ..., -33.4993, -13.6961, -20.9885]],\n",
      "       device='mps:0')\n",
      "Accuracy: 0.9932\n"
     ]
    }
   ],
   "source": [
    "model = model.to(mps_device)\n",
    "data_test = DataLoader(mnist_test_data, 10000, shuffle=False)\n",
    "\n",
    "total_crt = 0\n",
    "with torch.no_grad():\n",
    "    for x_batch, y_batch in data_test:\n",
    "        print(x_batch.shape)\n",
    "        x_batch = x_batch.to(mps_device)\n",
    "        y_batch = y_batch.to(mps_device)\n",
    "        pred = model(x_batch)\n",
    "        print(pred)\n",
    "        loss = loss_fn(pred, y_batch)\n",
    "        soft_pred = torch.softmax(pred, axis=1)\n",
    "        is_crt = (torch.argmax(soft_pred, axis=1) == y_batch).float().sum()\n",
    "        total_crt += is_crt\n",
    "    print(f\"Accuracy: {total_crt/len(data_test.dataset):.4f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 239,
   "id": "d60bc1b1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-37.6836, -19.1679, -17.8474,  ...,  42.5817, -36.9514, -18.8134],\n",
       "        [-10.7249, -13.9031,  35.2149,  ..., -31.3361, -15.8908, -30.5881],\n",
       "        [-26.1026,  32.3364, -27.1926,  ..., -17.1892, -11.1988, -25.5374],\n",
       "        ...,\n",
       "        [-43.9979, -17.6347, -33.0978,  ..., -25.9862, -14.4071, -18.7628],\n",
       "        [-25.4065, -27.3831, -34.0300,  ..., -22.3598,  -1.8935, -17.1660],\n",
       "        [-10.2998, -21.0205, -21.6235,  ..., -33.4993, -13.6961, -20.9885]],\n",
       "       grad_fn=<AddmmBackward0>)"
      ]
     },
     "execution_count": 239,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = model.to('cpu')\n",
    "testing = mnist_test_data.data.unsqueeze(1).to(torch.float32)/255\n",
    "proba= model(testing)\n",
    "proba"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "12d7b7dd",
   "metadata": {},
   "source": [
    "### Why only 9894 are correct, wheras in earlier batch wise calculation 9932 are correct ?\n",
    "- Reason : For some reason, the mnist_test_data is not scaled to [0-1] because of which there was issue. After dividing by 255, the range is 0-1 (which is used during training), the issue solved\n",
    "- But why it is not scaled to 0-1 despite scaled in the batchwise method above? \n",
    "    - Look above - when indexing directly it is scaled, but not when accessing via `.data`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 261,
   "id": "4db9e165",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[1.3843e-35, 1.5222e-27, 5.7014e-27,  ..., 1.0000e+00, 2.8787e-35,\n",
       "         2.1700e-27],\n",
       "        [1.1183e-20, 4.6593e-22, 1.0000e+00,  ..., 1.2509e-29, 6.3835e-23,\n",
       "         2.6431e-29],\n",
       "        [4.1709e-26, 1.0000e+00, 1.4024e-26,  ..., 3.0994e-22, 1.2385e-19,\n",
       "         7.3405e-26],\n",
       "        ...,\n",
       "        [4.1752e-39, 1.1750e-27, 2.2622e-34,  ..., 2.7736e-31, 2.9634e-26,\n",
       "         3.8031e-28],\n",
       "        [1.0559e-24, 1.4628e-25, 1.8989e-28,  ..., 2.2222e-23, 1.7187e-14,\n",
       "         4.0034e-21],\n",
       "        [1.5485e-18, 3.4196e-23, 1.8712e-23,  ..., 1.3017e-28, 5.1870e-20,\n",
       "         3.5307e-23]], grad_fn=<SoftmaxBackward0>)"
      ]
     },
     "execution_count": 261,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.softmax(proba, dim=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 241,
   "id": "b67b235e",
   "metadata": {},
   "outputs": [],
   "source": [
    "pred = torch.argmax(nn.Softmax( dim=1)(proba), axis=1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 230,
   "id": "4f33efcd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(0.9932)"
      ]
     },
     "execution_count": 230,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(pred == mnist_test_data.targets).float().mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 243,
   "id": "03188ba2",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import  classification_report"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 244,
   "id": "12d3535f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       1.00      1.00      1.00       980\n",
      "           1       1.00      0.99      1.00      1142\n",
      "           2       0.99      0.99      0.99      1029\n",
      "           3       0.99      1.00      0.99      1007\n",
      "           4       0.99      0.99      0.99       984\n",
      "           5       0.99      0.99      0.99       894\n",
      "           6       0.99      1.00      0.99       954\n",
      "           7       0.99      0.99      0.99      1034\n",
      "           8       1.00      1.00      1.00       974\n",
      "           9       0.99      0.99      0.99      1002\n",
      "\n",
      "    accuracy                           0.99     10000\n",
      "   macro avg       0.99      0.99      0.99     10000\n",
      "weighted avg       0.99      0.99      0.99     10000\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(classification_report(pred, mnist_test_data.targets))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 271,
   "id": "2744f3e3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(0.9932)"
      ]
     },
     "execution_count": 271,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from torchmetrics.classification import Recall, Precision\n",
    "recall = Recall(task=\"multiclass\", average='macro', num_classes=10)\n",
    "recall(pred, mnist_test_data.targets)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 270,
   "id": "6de50892",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(0.9932)"
      ]
     },
     "execution_count": 270,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "precision = Precision(task=\"multiclass\", average='macro', num_classes=10)\n",
    "precision(pred, mnist_test_data.targets)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "06c339ae",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.9 (pytorch)",
   "language": "python",
   "name": "pytorch"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}