EmotionRecognitionPyTorch.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "37a9dd41",
   "metadata": {},
   "source": [
    "# Emotion Recognition with CNNs\n",
    "Andrea Mansi (137857)\n",
    "\n",
    "Christian Cagnoni (137690)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "60ef9000",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import torch\n",
    "import torchvision\n",
    "from torch.utils.data import TensorDataset\n",
    "from torch.utils.data import DataLoader, RandomSampler, SequentialSampler\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "import matplotlib.pyplot as plot\n",
    "from PIL import Image\n",
    "import torchvision.transforms as T\n",
    "import time"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1be59ab4",
   "metadata": {},
   "source": [
    "## 1. Data import and preprocessing\n",
    "\n",
    "The input dataset is from FER2013 competition.\n",
    "\n",
    "Dataset URL: https://www.kaggle.com/ashishpatel26/facial-expression-recognitionferchallenge"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "67a5fceb",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape:(35887, 3)\n",
      "   emotion                                             pixels     Usage\n",
      "0        0  70 80 82 72 58 58 60 63 54 58 60 48 89 115 121...  Training\n",
      "1        0  151 150 147 155 148 133 111 140 170 174 182 15...  Training\n",
      "2        2  231 212 156 164 174 138 161 173 182 200 106 38...  Training\n",
      "3        4  24 32 36 30 32 23 19 20 30 41 21 22 32 34 21 1...  Training\n"
     ]
    }
   ],
   "source": [
    "# Data reading from csv file\n",
    "# Insert your path here:\n",
    "data_path = \"C:/Users/Mansitos_Picci/Desktop/fer2013.csv\"\n",
    "dataframe = pd.read_csv(data_path)\n",
    "\n",
    "# Printing dataset main infos\n",
    "print(\"Shape:\" + str(dataframe.shape))\n",
    "print(dataframe.head(4))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c6eddf37",
   "metadata": {},
   "source": [
    "Counting and plotting all the emotions. Dataset is not \"balanced\" some emotions are less present."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "5b92d5ee",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "emotions_labels_to_text = {0:'Anger >:(', 1:'Disgust :S', 2:'Fear :\"(', 3:'Happiness :D', 4: 'Sadness :(', 5: 'Surprise :O', 6: 'Neutral :|'}\n",
    "\n",
    "emotions_counts = dataframe['emotion'].value_counts(sort=False).reset_index()\n",
    "emotions_counts.columns = ['emotion', 'number']\n",
    "emotions_counts['emotion'] = emotions_counts['emotion'].map(emotions_labels_to_text)\n",
    "\n",
    "plot.figure(figsize=(10, 4))\n",
    "plot.bar(emotions_counts.emotion, emotions_counts.number)\n",
    "plot.title('Emotions distribution')\n",
    "plot.xlabel('\\nEmotions', fontsize=12)\n",
    "plot.ylabel('Number\\n', fontsize=12)\n",
    "plot.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c298a845",
   "metadata": {},
   "source": [
    "Processing the input data and preparing test and validation sets. Public test will be used, private test will be ignored."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "732daf85",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Taking Training data (dataset is already subdivided into training and test (private and public))\n",
    "data_train = dataframe[dataframe.Usage == \"Training\"]\n",
    "data_test  = dataframe[dataframe.Usage == \"PublicTest\"]\n",
    "\n",
    "# Processing the pixel column: converting the list (in which each pixel value is separated by a whitespace) into a list of ints\n",
    "imgs_train = pd.DataFrame(data_train.pixels.str.split(\" \").tolist())\n",
    "imgs_test  = pd.DataFrame(data_test.pixels.str.split(\" \").tolist())\n",
    "\n",
    "# Float conversion\n",
    "imgs_train = imgs_train.values.astype(float)\n",
    "imgs_test = imgs_test.values.astype(float)\n",
    "\n",
    "# reshaping to match model input\n",
    "x_train = imgs_train.reshape(-1, 48 , 48 ,1)\n",
    "x_test = imgs_test.reshape(-1, 48, 48, 1)\n",
    "\n",
    "# taking label train column and converting into monodimensional array\n",
    "y_train = data_train[\"emotion\"].values\n",
    "y_test = data_test[\"emotion\"].values"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2a59b184",
   "metadata": {},
   "source": [
    "Dataset augmentation phase. This phase enlarge the dataset by applying transformations to images. It can be skipped in order to use the raw input dataset. Enlarged dataset will require more time for training phase."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "9ec22972",
   "metadata": {},
   "outputs": [],
   "source": [
    "# set to TRUE if you want the test dataset to be enlarged by applying transformations to images\n",
    "enanche_dataset_flag = True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "5af74244",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Executing dataset augmentation... phase: 1\n",
      "Number of starting images: 28709\n",
      "Showing a random starting image:\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "if(enanche_dataset_flag):\n",
    "    print(\"Executing dataset augmentation... phase: 1\")\n",
    "    img=[]\n",
    "\n",
    "    def arrayToImage(array):\n",
    "        new_im = Image.new(\"L\", (48,48))\n",
    "\n",
    "        x_offset = 0\n",
    "        for im in array:\n",
    "          new_im.paste(im, (x_offset,0))\n",
    "          x_offset += im.size[0]\n",
    "\n",
    "        img.append((new_im.rotate(-90)).transpose(Image.FLIP_LEFT_RIGHT))\n",
    "\n",
    "    for i in range(len(x_train)):\n",
    "        tmp_img=[]\n",
    "        for j in range(len(x_train[i])):\n",
    "            tmp_img.append(Image.fromarray(x_train[i][j]))\n",
    "        arrayToImage(tmp_img)\n",
    "\n",
    "    \n",
    "    print(\"Number of starting images: \" + str(len(img)))\n",
    "    print(\"Showing a random starting image:\")\n",
    "    plot.imshow(img[0],cmap='gray')\n",
    "else:\n",
    "    print(\"Dataset augmentation skipped...\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "93720c22",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Executing dataset augmentation... phase: 2\n"
     ]
    }
   ],
   "source": [
    "if(enanche_dataset_flag):\n",
    "    print(\"Executing dataset augmentation... phase: 2\")\n",
    "    torch.manual_seed(42)\n",
    "\n",
    "    transformedImages=[]\n",
    "    transformedLabels=[]\n",
    "\n",
    "    perspective_transfromer = T.RandomPerspective(distortion_scale=0.5,p=0.5)\n",
    "    rotater = T.RandomRotation(degrees=(-90, 90))\n",
    "    \n",
    "    for i in range(len(img)):\n",
    "        transformedImages.append(perspective_transfromer(img[i]))\n",
    "        transformedLabels.append(y_train[i])\n",
    "    for i in range(len(img)):\n",
    "        transformedImages.append(rotater(img[i]))\n",
    "        transformedLabels.append(y_train[i])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "af2646d6",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Executing dataset augmentation... phase: 3\n",
      "Number of final images: 86127\n",
      "Showing a random transformed image:\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "if(enanche_dataset_flag):\n",
    "    x_train_ransformed=[]\n",
    "    print(\"Executing dataset augmentation... phase: 3\")\n",
    "\n",
    "    for i in transformedImages:\n",
    "        x_train_ransformed.append((np.asanyarray(i)).reshape(48,48,1))\n",
    "\n",
    "    x_train = np.concatenate((x_train,np.array(x_train_ransformed)))\n",
    "    y_train = np.concatenate((y_train,np.array(transformedLabels)))\n",
    "    \n",
    "    print(\"Number of final images: \" + str(len(x_train)))\n",
    "    print(\"Showing a random transformed image:\")\n",
    "    plot.imshow(transformedImages[-1],cmap='gray')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "285763b3",
   "metadata": {},
   "source": [
    "Normalizing the emotions img. count. Dataset is not balanced, some emotions have a lower abs. frequency compared to others (disgust is the extreme example of this).\n",
    "\n",
    "This phase is optionally executed. It takes the abs. frequency of the less frequent emotion and randomly discards imgs. from other classes in order to have the same amount of input data for each emotion. \n",
    "\n",
    "Note that this can improve accuracy because input data will be more balanced but at the same time some information is lost (so accuracy can decrease)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "31c035f6",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Status:1%\n",
      "Status:2%\n",
      "Status:3%\n",
      "Status:4%\n",
      "Status:5%\n",
      "Status:6%\n",
      "Status:7%\n",
      "Status:8%\n",
      "Status:9%\n",
      "Status:10%\n",
      "Status:11%\n",
      "Status:12%\n",
      "Status:13%\n",
      "Status:14%\n",
      "Status:15%\n",
      "Status:16%\n",
      "Status:17%\n",
      "Status:18%\n",
      "Status:19%\n",
      "Status:20%\n",
      "Status:21%\n",
      "Status:22%\n",
      "Status:23%\n",
      "Status:24%\n",
      "Status:25%\n",
      "Status:26%\n",
      "Status:27%\n",
      "Status:28%\n",
      "Status:29%\n",
      "Status:30%\n",
      "Status:31%\n",
      "Status:32%\n",
      "Status:33%\n",
      "Status:34%\n",
      "Status:35%\n",
      "Status:36%\n",
      "Status:37%\n",
      "Status:38%\n",
      "Status:39%\n",
      "Status:40%\n",
      "Status:41%\n",
      "Status:42%\n",
      "Status:43%\n",
      "Status:44%\n",
      "Status:45%\n",
      "Status:46%\n",
      "Status:47%\n",
      "Status:48%\n",
      "Status:49%\n",
      "Status:50%\n",
      "Status:51%\n",
      "Status:52%\n",
      "Status:53%\n",
      "Status:54%\n",
      "Status:55%\n",
      "Status:56%\n",
      "Status:57%\n",
      "Status:58%\n",
      "Status:59%\n",
      "Status:60%\n",
      "Status:61%\n",
      "Status:62%\n",
      "Status:63%\n",
      "Status:64%\n",
      "Status:65%\n",
      "Status:66%\n",
      "Status:67%\n",
      "Status:68%\n",
      "Status:69%\n",
      "Status:70%\n",
      "Status:71%\n",
      "Status:72%\n",
      "Status:73%\n",
      "Status:74%\n",
      "Status:75%\n",
      "Status:76%\n",
      "Status:77%\n",
      "Status:78%\n",
      "Status:79%\n",
      "Status:80%\n",
      "Status:81%\n",
      "Status:82%\n",
      "Status:83%\n",
      "Status:84%\n",
      "Status:85%\n",
      "Status:86%\n",
      "Status:87%\n",
      "Status:88%\n",
      "Status:89%\n",
      "Status:90%\n",
      "Status:91%\n",
      "Status:92%\n",
      "Status:93%\n",
      "Status:94%\n",
      "Status:95%\n",
      "Status:96%\n",
      "Status:97%\n",
      "Status:98%\n",
      "Status:99%\n",
      "Status: 100%\n"
     ]
    }
   ],
   "source": [
    "minAbsFreq=min(data_train[\"emotion\"].value_counts())\n",
    "originalShape=imgs_train.shape\n",
    "\n",
    "number_of_applied_transformations = 3\n",
    "originalShapeAugmented=(originalShape[0]*number_of_applied_transformations,originalShape[1])\n",
    "lowerBound=minAbsFreq*number_of_applied_transformations\n",
    "\n",
    "x_train_norm=x_train.reshape(originalShapeAugmented)\n",
    "\n",
    "tmpSer=[]\n",
    "merge=\"\"\n",
    "old_percentage = 0\n",
    "\n",
    "for i in range(len(x_train)):\n",
    "    new_percentage = int((i/len(x_train_norm))*100)\n",
    "    if(old_percentage < new_percentage):\n",
    "        print(\"Status:\" + str(new_percentage) + \"%\")\n",
    "        old_percentage = new_percentage\n",
    "    for j in range(len(x_train[i])):\n",
    "        merge=merge+\" \"+str(x_train_norm[i][j])\n",
    "    tmpSer.append(merge)\n",
    "    merge=\"\"\n",
    "print(\"Status: 100%\")\n",
    "ser=pd.Series(tmpSer)\n",
    "\n",
    "d={'emotion':y_train,'pixels':ser}\n",
    "augmentedDF=pd.DataFrame(data=d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "df00f414",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "id": "4797b48e",
   "metadata": {},
   "source": [
    "Creating model input: data is converted into tensors and then loaded into 2 dataloaders: train and test dataloaders."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "1decb3e7",
   "metadata": {},
   "outputs": [],
   "source": [
    "tensor_ids_train = torch.Tensor(x_train)\n",
    "labels_train = torch.LongTensor(y_train)\n",
    "train_dataset = TensorDataset(tensor_ids_train, labels_train)\n",
    "\n",
    "tensor_ids_test = torch.Tensor(x_test)\n",
    "labels_test = torch.LongTensor(y_test)\n",
    "test_dataset = TensorDataset(tensor_ids_test, labels_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "d4e38638",
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 128\n",
    "\n",
    "train_dataloader = DataLoader(train_dataset,\n",
    "                              sampler = RandomSampler(train_dataset),\n",
    "                              batch_size = batch_size)\n",
    "test_dataloader = DataLoader(test_dataset,\n",
    "                              sampler = RandomSampler(test_dataset),\n",
    "                              batch_size = batch_size)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d3b83e1",
   "metadata": {},
   "source": [
    "Defining useful functions for tensor printing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "e261770b",
   "metadata": {},
   "outputs": [],
   "source": [
    "def showImage(image,size,index,description):\n",
    "    img = image.reshape(size,size)\n",
    "    plot.axis(\"off\")\n",
    "    #print(description)\n",
    "    plot.figure(index)\n",
    "    plot.imshow(img, cmap ='gray')\n",
    "\n",
    "# This function can be used to see a tensor as an image\n",
    "def showTensorAsImage(tensor,size,index,description):\n",
    "      tensor_copy = tensor.to(torch.device(\"cpu\"))\n",
    "      tensor_copy = tensor_copy.detach()\n",
    "      showImage(np.array(tensor_copy[0][0]),size,index,description)\n",
    "     "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7d1528c4",
   "metadata": {},
   "source": [
    "Checking machine devices. Enable CUDA computation if available."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "cbdf8ffa",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Is GPU available?\n",
      "Yes :)\n",
      "\n",
      "Is cudnn backend enabled?\n",
      "True\n",
      "\n",
      "Training will run on device: cuda:0\n",
      "NVIDIA GeForce RTX 3070 Ti\n"
     ]
    }
   ],
   "source": [
    "print(\"Is GPU available?\")\n",
    "if(torch.cuda.is_available()):\n",
    "    print(\"Yes :)\")\n",
    "else:\n",
    "    print(\"No :(\")\n",
    "print(\"\")\n",
    "print(\"Is cudnn backend enabled?\")\n",
    "print(torch.backends.cudnn.enabled)\n",
    "print(\"\")\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(\"Training will run on device: \" + str(device))\n",
    "if(str(device)==\"cuda:0\"):print( torch.cuda.get_device_name(0))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b95f96d7",
   "metadata": {},
   "source": [
    "## 2. Defining the Model\n",
    "\n",
    "The following model is the last iteration/version. Previous versions are not included but informations are included in the PDF report."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "67901b51",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Net(\n",
      "  (conv1): Conv2d(1, 80, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))\n",
      "  (conv1_bn): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (conv2): Conv2d(80, 140, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))\n",
      "  (conv2_bn): BatchNorm2d(140, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (conv3): Conv2d(140, 200, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (conv3_bn): BatchNorm2d(200, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (conv4): Conv2d(200, 250, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (conv4_bn): BatchNorm2d(250, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (fc1): Linear(in_features=2250, out_features=160, bias=True)\n",
      "  (fc1_bn): BatchNorm1d(160, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (fc2): Linear(in_features=160, out_features=70, bias=True)\n",
      "  (fc2_bn): BatchNorm1d(70, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (fc3): Linear(in_features=70, out_features=20, bias=True)\n",
      "  (fc3_bn): BatchNorm1d(20, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (relu): ReLU()\n",
      "  (softmax): Softmax(dim=1)\n",
      "  (drop): Dropout(p=0.25, inplace=False)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "class Net(nn.Module):\n",
    "    def __init__(self):\n",
    "      super().__init__()\n",
    "      self.conv1 = nn.Conv2d(1,80,kernel_size=5,padding=2) # First convolutional layer\n",
    "      self.conv1_bn = nn.BatchNorm2d(80)\n",
    "      self.pool = nn.MaxPool2d(kernel_size=2,stride=2) # Pooling layer\n",
    "      self.conv2 = nn.Conv2d(80,140,kernel_size=5,padding=2) # Second convolutional layer\n",
    "      self.conv2_bn = nn.BatchNorm2d(140)\n",
    "      self.conv3 = nn.Conv2d(140,200,kernel_size=3,padding=1) # Third convolutional layer\n",
    "      self.conv3_bn = nn.BatchNorm2d(200)\n",
    "      self.conv4 = nn.Conv2d(200,250,kernel_size=3,padding=1) # Fourth convolutional layer\n",
    "      self.conv4_bn = nn.BatchNorm2d(250)\n",
    "      self.fc1 = nn.Linear(250*3*3,160) # First linear layer\n",
    "      self.fc1_bn = nn.BatchNorm1d(160)\n",
    "      self.fc2 = nn.Linear(160,70) # Second linear layer\n",
    "      self.fc2_bn = nn.BatchNorm1d(70)\n",
    "      self.fc3 = nn.Linear(70,20) # Third linear layer\n",
    "      self.fc3_bn = nn.BatchNorm1d(20)\n",
    "      self.relu = nn.ReLU()\n",
    "      self.softmax = nn.Softmax(dim=1)\n",
    "      self.drop = nn.Dropout(p=0.25)\n",
    "\n",
    "    # Applying step-by-step the image classification architecture to the input\n",
    "    def forward(self, x):     \n",
    "      x = self.conv1(x) \n",
    "      x = self.conv1_bn(x)\n",
    "      x = self.relu(x)\n",
    "      x = self.pool(x)\n",
    "      x = self.drop(x)\n",
    "        \n",
    "      if(net.training == False):\n",
    "        showTensorAsImage(x,24,1,description=\"Random tensor from 1st conv. layer\")\n",
    "        \n",
    "      x = self.conv2(x)  \n",
    "      x = self.conv2_bn(x)\n",
    "      x = self.relu(x)\n",
    "      x = self.pool(x)\n",
    "      x = self.drop(x)\n",
    "    \n",
    "      if(net.training == False):\n",
    "        showTensorAsImage(x,12,2,description=\"Random tensor from 2nd conv. layer\")\n",
    "    \n",
    "      x = self.conv3(x)\n",
    "      x = self.conv3_bn(x)\n",
    "      x = self.relu(x)\n",
    "      x = self.pool(x)\n",
    "      x = self.drop(x)\n",
    "        \n",
    "      if(net.training == False):\n",
    "        showTensorAsImage(x,6,3,description=\"Random tensor from 3rd conv. layer\")\n",
    "        \n",
    "      x = self.conv4(x)\n",
    "      x = self.conv4_bn(x)\n",
    "      x = self.relu(x)\n",
    "      x = self.pool(x)\n",
    "      x = self.drop(x)\n",
    "    \n",
    "      x = x.reshape(x.size(0),250*3*3) # Modifying the shape of x to make the tensor fit in the first linear\n",
    "      x = self.fc1(x)\n",
    "      x = self.fc1_bn(x)\n",
    "      x = self.relu(x) # Applying the activation function\n",
    "      x = self.drop(x)\n",
    "    \n",
    "      x = self.fc2(x)\n",
    "      x = self.fc2_bn(x)\n",
    "      x = self.relu(x) # Applying the activation function\n",
    "      x = self.drop(x)\n",
    "        \n",
    "      x = self.fc3(x)\n",
    "      x = self.fc3_bn(x)\n",
    "      x = self.softmax(x)\n",
    "      return x\n",
    "\n",
    "net = Net()\n",
    "print(net)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "82a77de1",
   "metadata": {},
   "source": [
    "#### Train function definition"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "fe847c1a",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[epoch: 1, mini-batch: 673] loss: 2.790\n",
      "[epoch: 2, mini-batch: 673] loss: 2.666\n",
      "[epoch: 3, mini-batch: 673] loss: 2.630\n",
      "[epoch: 4, mini-batch: 673] loss: 2.604\n",
      "[epoch: 5, mini-batch: 673] loss: 2.584\n",
      "[epoch: 6, mini-batch: 673] loss: 2.569\n",
      "[epoch: 7, mini-batch: 673] loss: 2.561\n",
      "[epoch: 8, mini-batch: 673] loss: 2.547\n",
      "[epoch: 9, mini-batch: 673] loss: 2.536\n",
      "[epoch: 10, mini-batch: 673] loss: 2.529\n",
      "[epoch: 11, mini-batch: 673] loss: 2.525\n",
      "[epoch: 12, mini-batch: 673] loss: 2.515\n",
      "[epoch: 13, mini-batch: 673] loss: 2.510\n",
      "[epoch: 14, mini-batch: 673] loss: 2.504\n",
      "[epoch: 15, mini-batch: 673] loss: 2.499\n",
      "[epoch: 16, mini-batch: 673] loss: 2.493\n",
      "[epoch: 17, mini-batch: 673] loss: 2.488\n",
      "[epoch: 18, mini-batch: 673] loss: 2.484\n",
      "[epoch: 19, mini-batch: 673] loss: 2.481\n",
      "[epoch: 20, mini-batch: 673] loss: 2.477\n",
      "[epoch: 21, mini-batch: 673] loss: 2.473\n",
      "[epoch: 22, mini-batch: 673] loss: 2.470\n",
      "[epoch: 23, mini-batch: 673] loss: 2.467\n",
      "[epoch: 24, mini-batch: 673] loss: 2.464\n",
      "[epoch: 25, mini-batch: 673] loss: 2.461\n",
      "[epoch: 26, mini-batch: 673] loss: 2.457\n",
      "[epoch: 27, mini-batch: 673] loss: 2.456\n",
      "[epoch: 28, mini-batch: 673] loss: 2.451\n",
      "[epoch: 29, mini-batch: 673] loss: 2.449\n",
      "[epoch: 30, mini-batch: 673] loss: 2.447\n",
      "[epoch: 31, mini-batch: 673] loss: 2.445\n",
      "[epoch: 32, mini-batch: 673] loss: 2.442\n",
      "[epoch: 33, mini-batch: 673] loss: 2.439\n",
      "[epoch: 34, mini-batch: 673] loss: 2.439\n",
      "[epoch: 35, mini-batch: 673] loss: 2.435\n",
      "[epoch: 36, mini-batch: 673] loss: 2.435\n",
      "[epoch: 37, mini-batch: 673] loss: 2.432\n",
      "[epoch: 38, mini-batch: 673] loss: 2.431\n",
      "[epoch: 39, mini-batch: 673] loss: 2.429\n",
      "[epoch: 40, mini-batch: 673] loss: 2.426\n",
      "[epoch: 41, mini-batch: 673] loss: 2.426\n",
      "[epoch: 42, mini-batch: 673] loss: 2.423\n",
      "[epoch: 43, mini-batch: 673] loss: 2.420\n",
      "[epoch: 44, mini-batch: 673] loss: 2.419\n",
      "[epoch: 45, mini-batch: 673] loss: 2.418\n",
      "[epoch: 46, mini-batch: 673] loss: 2.415\n",
      "[epoch: 47, mini-batch: 673] loss: 2.414\n",
      "[epoch: 48, mini-batch: 673] loss: 2.414\n",
      "[epoch: 49, mini-batch: 673] loss: 2.413\n",
      "[epoch: 50, mini-batch: 673] loss: 2.409\n",
      "[epoch: 51, mini-batch: 673] loss: 2.410\n",
      "[epoch: 52, mini-batch: 673] loss: 2.407\n",
      "[epoch: 53, mini-batch: 673] loss: 2.405\n",
      "[epoch: 54, mini-batch: 673] loss: 2.406\n",
      "[epoch: 55, mini-batch: 673] loss: 2.403\n",
      "[epoch: 56, mini-batch: 673] loss: 2.403\n",
      "[epoch: 57, mini-batch: 673] loss: 2.401\n",
      "[epoch: 58, mini-batch: 673] loss: 2.398\n",
      "[epoch: 59, mini-batch: 673] loss: 2.397\n",
      "[epoch: 60, mini-batch: 673] loss: 2.397\n",
      "[epoch: 61, mini-batch: 673] loss: 2.395\n",
      "[epoch: 62, mini-batch: 673] loss: 2.394\n",
      "[epoch: 63, mini-batch: 673] loss: 2.394\n",
      "[epoch: 64, mini-batch: 673] loss: 2.391\n",
      "[epoch: 65, mini-batch: 673] loss: 2.392\n",
      "[epoch: 66, mini-batch: 673] loss: 2.391\n",
      "[epoch: 67, mini-batch: 673] loss: 2.390\n",
      "[epoch: 68, mini-batch: 673] loss: 2.388\n",
      "[epoch: 69, mini-batch: 673] loss: 2.385\n",
      "[epoch: 70, mini-batch: 673] loss: 2.388\n",
      "[epoch: 71, mini-batch: 673] loss: 2.385\n",
      "[epoch: 72, mini-batch: 673] loss: 2.383\n",
      "[epoch: 73, mini-batch: 673] loss: 2.384\n",
      "[epoch: 74, mini-batch: 673] loss: 2.382\n",
      "[epoch: 75, mini-batch: 673] loss: 2.381\n",
      "[epoch: 76, mini-batch: 673] loss: 2.382\n",
      "[epoch: 77, mini-batch: 673] loss: 2.378\n",
      "[epoch: 78, mini-batch: 673] loss: 2.379\n",
      "[epoch: 79, mini-batch: 673] loss: 2.379\n",
      "[epoch: 80, mini-batch: 673] loss: 2.378\n",
      "[epoch: 81, mini-batch: 673] loss: 2.375\n",
      "[epoch: 82, mini-batch: 673] loss: 2.375\n",
      "[epoch: 83, mini-batch: 673] loss: 2.375\n",
      "[epoch: 84, mini-batch: 673] loss: 2.375\n",
      "[epoch: 85, mini-batch: 673] loss: 2.371\n",
      "[epoch: 86, mini-batch: 673] loss: 2.371\n",
      "[epoch: 87, mini-batch: 673] loss: 2.371\n",
      "[epoch: 88, mini-batch: 673] loss: 2.370\n",
      "[epoch: 89, mini-batch: 673] loss: 2.371\n",
      "[epoch: 90, mini-batch: 673] loss: 2.368\n",
      "[epoch: 91, mini-batch: 673] loss: 2.371\n",
      "[epoch: 92, mini-batch: 673] loss: 2.368\n",
      "[epoch: 93, mini-batch: 673] loss: 2.366\n",
      "[epoch: 94, mini-batch: 673] loss: 2.367\n",
      "[epoch: 95, mini-batch: 673] loss: 2.364\n",
      "[epoch: 96, mini-batch: 673] loss: 2.364\n",
      "[epoch: 97, mini-batch: 673] loss: 2.361\n",
      "[epoch: 98, mini-batch: 673] loss: 2.361\n",
      "[epoch: 99, mini-batch: 673] loss: 2.361\n",
      "[epoch: 100, mini-batch: 673] loss: 2.362\n",
      "[epoch: 101, mini-batch: 673] loss: 2.360\n",
      "[epoch: 102, mini-batch: 673] loss: 2.359\n",
      "[epoch: 103, mini-batch: 673] loss: 2.359\n",
      "[epoch: 104, mini-batch: 673] loss: 2.358\n",
      "[epoch: 105, mini-batch: 673] loss: 2.358\n",
      "[epoch: 106, mini-batch: 673] loss: 2.356\n",
      "[epoch: 107, mini-batch: 673] loss: 2.357\n",
      "[epoch: 108, mini-batch: 673] loss: 2.357\n",
      "[epoch: 109, mini-batch: 673] loss: 2.354\n",
      "[epoch: 110, mini-batch: 673] loss: 2.356\n",
      "[epoch: 111, mini-batch: 673] loss: 2.352\n",
      "[epoch: 112, mini-batch: 673] loss: 2.354\n",
      "[epoch: 113, mini-batch: 673] loss: 2.351\n",
      "[epoch: 114, mini-batch: 673] loss: 2.352\n",
      "[epoch: 115, mini-batch: 673] loss: 2.351\n",
      "[epoch: 116, mini-batch: 673] loss: 2.350\n",
      "[epoch: 117, mini-batch: 673] loss: 2.349\n",
      "[epoch: 118, mini-batch: 673] loss: 2.350\n",
      "[epoch: 119, mini-batch: 673] loss: 2.349\n",
      "[epoch: 120, mini-batch: 673] loss: 2.348\n",
      "[epoch: 121, mini-batch: 673] loss: 2.346\n",
      "[epoch: 122, mini-batch: 673] loss: 2.348\n",
      "[epoch: 123, mini-batch: 673] loss: 2.345\n",
      "[epoch: 124, mini-batch: 673] loss: 2.347\n",
      "[epoch: 125, mini-batch: 673] loss: 2.345\n",
      "[epoch: 126, mini-batch: 673] loss: 2.345\n",
      "[epoch: 127, mini-batch: 673] loss: 2.345\n",
      "[epoch: 128, mini-batch: 673] loss: 2.342\n",
      "[epoch: 129, mini-batch: 673] loss: 2.342\n",
      "[epoch: 130, mini-batch: 673] loss: 2.343\n",
      "[epoch: 131, mini-batch: 673] loss: 2.340\n",
      "[epoch: 132, mini-batch: 673] loss: 2.343\n",
      "[epoch: 133, mini-batch: 673] loss: 2.341\n",
      "[epoch: 134, mini-batch: 673] loss: 2.340\n",
      "[epoch: 135, mini-batch: 673] loss: 2.339\n",
      "[epoch: 136, mini-batch: 673] loss: 2.339\n",
      "[epoch: 137, mini-batch: 673] loss: 2.339\n",
      "[epoch: 138, mini-batch: 673] loss: 2.338\n",
      "[epoch: 139, mini-batch: 673] loss: 2.339\n",
      "[epoch: 140, mini-batch: 673] loss: 2.337\n",
      "[epoch: 141, mini-batch: 673] loss: 2.337\n",
      "[epoch: 142, mini-batch: 673] loss: 2.336\n",
      "[epoch: 143, mini-batch: 673] loss: 2.335\n",
      "[epoch: 144, mini-batch: 673] loss: 2.337\n",
      "[epoch: 145, mini-batch: 673] loss: 2.334\n",
      "[epoch: 146, mini-batch: 673] loss: 2.337\n",
      "[epoch: 147, mini-batch: 673] loss: 2.333\n",
      "[epoch: 148, mini-batch: 673] loss: 2.336\n",
      "[epoch: 149, mini-batch: 673] loss: 2.334\n",
      "[epoch: 150, mini-batch: 673] loss: 2.332\n",
      "[epoch: 151, mini-batch: 673] loss: 2.332\n",
      "[epoch: 152, mini-batch: 673] loss: 2.332\n",
      "[epoch: 153, mini-batch: 673] loss: 2.330\n",
      "[epoch: 154, mini-batch: 673] loss: 2.331\n",
      "[epoch: 155, mini-batch: 673] loss: 2.332\n",
      "[epoch: 156, mini-batch: 673] loss: 2.330\n",
      "[epoch: 157, mini-batch: 673] loss: 2.331\n",
      "[epoch: 158, mini-batch: 673] loss: 2.327\n",
      "[epoch: 159, mini-batch: 673] loss: 2.327\n",
      "[epoch: 160, mini-batch: 673] loss: 2.327\n",
      "[epoch: 161, mini-batch: 673] loss: 2.330\n",
      "[epoch: 162, mini-batch: 673] loss: 2.326\n",
      "[epoch: 163, mini-batch: 673] loss: 2.326\n",
      "[epoch: 164, mini-batch: 673] loss: 2.327\n",
      "[epoch: 165, mini-batch: 673] loss: 2.326\n",
      "[epoch: 166, mini-batch: 673] loss: 2.327\n",
      "[epoch: 167, mini-batch: 673] loss: 2.326\n",
      "[epoch: 168, mini-batch: 673] loss: 2.325\n",
      "[epoch: 169, mini-batch: 673] loss: 2.324\n",
      "[epoch: 170, mini-batch: 673] loss: 2.323\n",
      "[epoch: 171, mini-batch: 673] loss: 2.321\n",
      "[epoch: 172, mini-batch: 673] loss: 2.322\n",
      "[epoch: 173, mini-batch: 673] loss: 2.323\n",
      "[epoch: 174, mini-batch: 673] loss: 2.320\n",
      "[epoch: 175, mini-batch: 673] loss: 2.320\n",
      "[epoch: 176, mini-batch: 673] loss: 2.322\n",
      "[epoch: 177, mini-batch: 673] loss: 2.321\n",
      "[epoch: 178, mini-batch: 673] loss: 2.320\n",
      "[epoch: 179, mini-batch: 673] loss: 2.321\n",
      "[epoch: 180, mini-batch: 673] loss: 2.319\n",
      "[epoch: 181, mini-batch: 673] loss: 2.320\n",
      "[epoch: 182, mini-batch: 673] loss: 2.318\n",
      "[epoch: 183, mini-batch: 673] loss: 2.319\n",
      "[epoch: 184, mini-batch: 673] loss: 2.319\n",
      "[epoch: 185, mini-batch: 673] loss: 2.318\n",
      "[epoch: 186, mini-batch: 673] loss: 2.316\n",
      "[epoch: 187, mini-batch: 673] loss: 2.318\n",
      "[epoch: 188, mini-batch: 673] loss: 2.313\n",
      "[epoch: 189, mini-batch: 673] loss: 2.315\n",
      "[epoch: 190, mini-batch: 673] loss: 2.317\n",
      "[epoch: 191, mini-batch: 673] loss: 2.317\n",
      "[epoch: 192, mini-batch: 673] loss: 2.315\n",
      "[epoch: 193, mini-batch: 673] loss: 2.314\n",
      "[epoch: 194, mini-batch: 673] loss: 2.315\n",
      "[epoch: 195, mini-batch: 673] loss: 2.316\n",
      "[epoch: 196, mini-batch: 673] loss: 2.313\n",
      "[epoch: 197, mini-batch: 673] loss: 2.314\n",
      "[epoch: 198, mini-batch: 673] loss: 2.315\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[epoch: 199, mini-batch: 673] loss: 2.313\n",
      "[epoch: 200, mini-batch: 673] loss: 2.314\n",
      "[epoch: 201, mini-batch: 673] loss: 2.312\n",
      "[epoch: 202, mini-batch: 673] loss: 2.313\n",
      "[epoch: 203, mini-batch: 673] loss: 2.311\n",
      "[epoch: 204, mini-batch: 673] loss: 2.312\n",
      "[epoch: 205, mini-batch: 673] loss: 2.308\n",
      "[epoch: 206, mini-batch: 673] loss: 2.310\n",
      "[epoch: 207, mini-batch: 673] loss: 2.311\n",
      "[epoch: 208, mini-batch: 673] loss: 2.311\n",
      "[epoch: 209, mini-batch: 673] loss: 2.310\n",
      "[epoch: 210, mini-batch: 673] loss: 2.307\n",
      "[epoch: 211, mini-batch: 673] loss: 2.309\n",
      "[epoch: 212, mini-batch: 673] loss: 2.310\n",
      "[epoch: 213, mini-batch: 673] loss: 2.309\n",
      "[epoch: 214, mini-batch: 673] loss: 2.308\n",
      "[epoch: 215, mini-batch: 673] loss: 2.308\n",
      "[epoch: 216, mini-batch: 673] loss: 2.308\n",
      "[epoch: 217, mini-batch: 673] loss: 2.307\n",
      "[epoch: 218, mini-batch: 673] loss: 2.309\n",
      "[epoch: 219, mini-batch: 673] loss: 2.308\n",
      "[epoch: 220, mini-batch: 673] loss: 2.306\n",
      "[epoch: 221, mini-batch: 673] loss: 2.306\n",
      "[epoch: 222, mini-batch: 673] loss: 2.306\n",
      "[epoch: 223, mini-batch: 673] loss: 2.305\n",
      "[epoch: 224, mini-batch: 673] loss: 2.305\n",
      "[epoch: 225, mini-batch: 673] loss: 2.304\n",
      "[epoch: 226, mini-batch: 673] loss: 2.305\n",
      "[epoch: 227, mini-batch: 673] loss: 2.305\n",
      "[epoch: 228, mini-batch: 673] loss: 2.304\n",
      "[epoch: 229, mini-batch: 673] loss: 2.303\n",
      "[epoch: 230, mini-batch: 673] loss: 2.304\n",
      "[epoch: 231, mini-batch: 673] loss: 2.303\n",
      "[epoch: 232, mini-batch: 673] loss: 2.302\n",
      "[epoch: 233, mini-batch: 673] loss: 2.301\n",
      "[epoch: 234, mini-batch: 673] loss: 2.303\n",
      "[epoch: 235, mini-batch: 673] loss: 2.304\n",
      "[epoch: 236, mini-batch: 673] loss: 2.302\n",
      "[epoch: 237, mini-batch: 673] loss: 2.302\n",
      "[epoch: 238, mini-batch: 673] loss: 2.301\n",
      "[epoch: 239, mini-batch: 673] loss: 2.301\n",
      "[epoch: 240, mini-batch: 673] loss: 2.300\n",
      "[epoch: 241, mini-batch: 673] loss: 2.301\n",
      "[epoch: 242, mini-batch: 673] loss: 2.301\n",
      "[epoch: 243, mini-batch: 673] loss: 2.297\n",
      "[epoch: 244, mini-batch: 673] loss: 2.299\n",
      "[epoch: 245, mini-batch: 673] loss: 2.299\n",
      "[epoch: 246, mini-batch: 673] loss: 2.299\n",
      "[epoch: 247, mini-batch: 673] loss: 2.298\n",
      "[epoch: 248, mini-batch: 673] loss: 2.298\n",
      "[epoch: 249, mini-batch: 673] loss: 2.298\n",
      "[epoch: 250, mini-batch: 673] loss: 2.297\n",
      "[epoch: 251, mini-batch: 673] loss: 2.298\n",
      "[epoch: 252, mini-batch: 673] loss: 2.296\n",
      "[epoch: 253, mini-batch: 673] loss: 2.297\n",
      "[epoch: 254, mini-batch: 673] loss: 2.297\n",
      "[epoch: 255, mini-batch: 673] loss: 2.296\n",
      "[epoch: 256, mini-batch: 673] loss: 2.295\n",
      "[epoch: 257, mini-batch: 673] loss: 2.297\n",
      "[epoch: 258, mini-batch: 673] loss: 2.296\n",
      "[epoch: 259, mini-batch: 673] loss: 2.297\n",
      "[epoch: 260, mini-batch: 673] loss: 2.292\n",
      "[epoch: 261, mini-batch: 673] loss: 2.295\n",
      "[epoch: 262, mini-batch: 673] loss: 2.294\n",
      "[epoch: 263, mini-batch: 673] loss: 2.294\n",
      "[epoch: 264, mini-batch: 673] loss: 2.295\n",
      "[epoch: 265, mini-batch: 673] loss: 2.294\n",
      "[epoch: 266, mini-batch: 673] loss: 2.292\n",
      "[epoch: 267, mini-batch: 673] loss: 2.292\n",
      "[epoch: 268, mini-batch: 673] loss: 2.294\n",
      "[epoch: 269, mini-batch: 673] loss: 2.293\n",
      "[epoch: 270, mini-batch: 673] loss: 2.293\n",
      "[epoch: 271, mini-batch: 673] loss: 2.293\n",
      "[epoch: 272, mini-batch: 673] loss: 2.293\n",
      "[epoch: 273, mini-batch: 673] loss: 2.293\n",
      "[epoch: 274, mini-batch: 673] loss: 2.292\n",
      "[epoch: 275, mini-batch: 673] loss: 2.290\n",
      "[epoch: 276, mini-batch: 673] loss: 2.292\n",
      "[epoch: 277, mini-batch: 673] loss: 2.291\n",
      "[epoch: 278, mini-batch: 673] loss: 2.289\n",
      "[epoch: 279, mini-batch: 673] loss: 2.290\n",
      "[epoch: 280, mini-batch: 673] loss: 2.290\n",
      "[epoch: 281, mini-batch: 673] loss: 2.291\n",
      "[epoch: 282, mini-batch: 673] loss: 2.289\n",
      "[epoch: 283, mini-batch: 673] loss: 2.291\n",
      "[epoch: 284, mini-batch: 673] loss: 2.290\n",
      "[epoch: 285, mini-batch: 673] loss: 2.289\n",
      "[epoch: 286, mini-batch: 673] loss: 2.288\n",
      "[epoch: 287, mini-batch: 673] loss: 2.288\n",
      "[epoch: 288, mini-batch: 673] loss: 2.288\n",
      "[epoch: 289, mini-batch: 673] loss: 2.288\n",
      "[epoch: 290, mini-batch: 673] loss: 2.290\n",
      "[epoch: 291, mini-batch: 673] loss: 2.288\n",
      "[epoch: 292, mini-batch: 673] loss: 2.289\n",
      "[epoch: 293, mini-batch: 673] loss: 2.287\n",
      "[epoch: 294, mini-batch: 673] loss: 2.287\n",
      "[epoch: 295, mini-batch: 673] loss: 2.287\n",
      "[epoch: 296, mini-batch: 673] loss: 2.288\n",
      "[epoch: 297, mini-batch: 673] loss: 2.286\n",
      "[epoch: 298, mini-batch: 673] loss: 2.286\n",
      "[epoch: 299, mini-batch: 673] loss: 2.286\n",
      "[epoch: 300, mini-batch: 673] loss: 2.285\n",
      "[epoch: 301, mini-batch: 673] loss: 2.285\n",
      "[epoch: 302, mini-batch: 673] loss: 2.287\n",
      "[epoch: 303, mini-batch: 673] loss: 2.284\n",
      "[epoch: 304, mini-batch: 673] loss: 2.283\n",
      "[epoch: 305, mini-batch: 673] loss: 2.286\n",
      "[epoch: 306, mini-batch: 673] loss: 2.285\n",
      "[epoch: 307, mini-batch: 673] loss: 2.283\n",
      "[epoch: 308, mini-batch: 673] loss: 2.282\n",
      "[epoch: 309, mini-batch: 673] loss: 2.285\n",
      "[epoch: 310, mini-batch: 673] loss: 2.285\n",
      "[epoch: 311, mini-batch: 673] loss: 2.287\n",
      "[epoch: 312, mini-batch: 673] loss: 2.284\n",
      "[epoch: 313, mini-batch: 673] loss: 2.284\n",
      "[epoch: 314, mini-batch: 673] loss: 2.282\n",
      "[epoch: 315, mini-batch: 673] loss: 2.285\n",
      "[epoch: 316, mini-batch: 673] loss: 2.283\n",
      "[epoch: 317, mini-batch: 673] loss: 2.281\n",
      "[epoch: 318, mini-batch: 673] loss: 2.281\n",
      "[epoch: 319, mini-batch: 673] loss: 2.282\n",
      "[epoch: 320, mini-batch: 673] loss: 2.282\n",
      "[epoch: 321, mini-batch: 673] loss: 2.282\n",
      "[epoch: 322, mini-batch: 673] loss: 2.281\n",
      "[epoch: 323, mini-batch: 673] loss: 2.280\n",
      "[epoch: 324, mini-batch: 673] loss: 2.282\n",
      "[epoch: 325, mini-batch: 673] loss: 2.282\n",
      "[epoch: 326, mini-batch: 673] loss: 2.282\n",
      "[epoch: 327, mini-batch: 673] loss: 2.280\n",
      "[epoch: 328, mini-batch: 673] loss: 2.279\n",
      "[epoch: 329, mini-batch: 673] loss: 2.279\n",
      "[epoch: 330, mini-batch: 673] loss: 2.280\n",
      "[epoch: 331, mini-batch: 673] loss: 2.284\n",
      "[epoch: 332, mini-batch: 673] loss: 2.280\n",
      "[epoch: 333, mini-batch: 673] loss: 2.279\n",
      "[epoch: 334, mini-batch: 673] loss: 2.280\n",
      "[epoch: 335, mini-batch: 673] loss: 2.278\n",
      "[epoch: 336, mini-batch: 673] loss: 2.279\n",
      "[epoch: 337, mini-batch: 673] loss: 2.277\n",
      "[epoch: 338, mini-batch: 673] loss: 2.276\n",
      "[epoch: 339, mini-batch: 673] loss: 2.278\n",
      "[epoch: 340, mini-batch: 673] loss: 2.278\n",
      "[epoch: 341, mini-batch: 673] loss: 2.278\n",
      "[epoch: 342, mini-batch: 673] loss: 2.279\n",
      "[epoch: 343, mini-batch: 673] loss: 2.278\n",
      "[epoch: 344, mini-batch: 673] loss: 2.276\n",
      "[epoch: 345, mini-batch: 673] loss: 2.278\n",
      "[epoch: 346, mini-batch: 673] loss: 2.277\n",
      "[epoch: 347, mini-batch: 673] loss: 2.277\n",
      "[epoch: 348, mini-batch: 673] loss: 2.277\n",
      "[epoch: 349, mini-batch: 673] loss: 2.279\n",
      "[epoch: 350, mini-batch: 673] loss: 2.275\n",
      "Finished Training :)\n",
      "Training time: 19576.756263017654 seconds\n"
     ]
    }
   ],
   "source": [
    "# Moving model to device (to cuda if available)\n",
    "net = net.to(device)\n",
    "net.train()\n",
    "\n",
    "criterion = nn.CrossEntropyLoss() # Defining the criterion\n",
    "\n",
    "#optimizer = optim.SGD(net.parameters(),momentum=0.9,lr=0.001) # Defining the optimizer\n",
    "optimizer = optim.Adam(net.parameters(),lr=0.003)\n",
    "\n",
    "start_time = time.time()\n",
    "\n",
    "for epoch in range(350): #Looping over the dataset three times\n",
    "\n",
    "    running_loss = 0.0\n",
    "    for i, data in enumerate(train_dataloader, 0):\n",
    "        \n",
    "        inputs, labels = data # The input data is a list [inputs, labels]\n",
    "\n",
    "        inputs = inputs.permute(0, 3, 1, 2) # permuting the input to match the order used by pytorch\n",
    "        \n",
    "        # Moving to device (to cuda if available)\n",
    "        inputs = inputs.to(device)\n",
    "        labels = labels.to(device)\n",
    "        \n",
    "        optimizer.zero_grad() # Setting the parameter gradients to zero\n",
    "\n",
    "        outputs = net(inputs) # Forward pass\n",
    "        loss = criterion(outputs,labels) # Applying the criterion\n",
    "        loss.backward() # Backward pass\n",
    "        optimizer.step() # Optimization step\n",
    "\n",
    "        running_loss += loss.item() # Updating the running loss\n",
    "        if i % len(train_dataloader) == len(train_dataloader)-1:  # Printing the running loss\n",
    "            print('[epoch: %d, mini-batch: %d] loss: %.3f' %\n",
    "                  (epoch + 1, i + 1, running_loss / len(train_dataloader)))\n",
    "            running_loss = 0.0\n",
    "\n",
    "print('Finished Training :)')\n",
    "print(\"Training time: %s seconds\" % (time.time() - start_time))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1fec849e",
   "metadata": {},
   "source": [
    "#### Evaluating the model\n",
    "Public test set is used in order to test che model accuracy and other metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "98a61518",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of the network on the 3589 test images (Public test-set): 66 %\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Calculating the accuracy of the network on the whole dataset\n",
    "correct = 0\n",
    "total = 0\n",
    "i = 0\n",
    "\n",
    "# Setting network in evaluation/testing mode (for disabling dropout layers)\n",
    "net.eval()\n",
    "\n",
    "with torch.no_grad():\n",
    "    for data in test_dataloader:\n",
    "        \n",
    "        images, labels = data # Getting the test data\n",
    "\n",
    "        images = images.permute(0, 3, 1, 2)\n",
    "        # Moving to device (cuda if available)\n",
    "        images = images.to(device)\n",
    "        labels = labels.to(device)\n",
    "        outputs = net(images) # Getting the network output\n",
    "        _, predicted = torch.max(outputs.data, 1)\n",
    "        total += labels.size(0)\n",
    "        correct += (predicted == labels).sum().item()\n",
    "        \n",
    "accurcyMean=100 * correct / total\n",
    "              \n",
    "print('Accuracy of the network on the %d test images (Public test-set): %d %%' % (len(imgs_test), accurcyMean))\n",
    "\n",
    "# showing accumulated (convs) plots\n",
    "plot.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "b1162745",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Anger >:( (Class: 0): \t 76 %\n",
      "Accuracy of Disgust :S (Class: 1): \t  0 %\n",
      "Accuracy of Fear :\"( (Class: 2): \t 45 %\n",
      "Accuracy of Happiness :D (Class: 3): \t 97 %\n",
      "Accuracy of Sadness :( (Class: 4): \t 52 %\n",
      "Accuracy of Surprise :O (Class: 5): \t 94 %\n",
      "Accuracy of Neutral :| (Class: 6): \t 75 %\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAOcAAADnCAYAAADl9EEgAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8/fFQqAAAACXBIWXMAAAsTAAALEwEAmpwYAAAD6ElEQVR4nO3cMUpkWQCG0XpSopgZS+1BF+E2FMwNxBW5IxPB3NBAQwXx9gbKZmb69viVnBP64Odh8dUFwbuMMVZAz953vwCwnTghSpwQJU6IEidErX/3cFmWH/mn3Lu7u2lbFxcXU3aWZZmyw+4ZY2z98J2cECVOiBInRIkTosQJUeKEKHFClDghSpwQJU6IEidEiROixAlR4oQocUKUOCFKnBC1/O7e2p96EwKUuAkBdow4IUqcECVOiBInRIkTosQJUeKEKHFClDghSpwQJU6IEidEiROixAlR4oQocUKUOCFKnBAlTogSJ0SJE6LECVHihChxQpQ4IUqcECVOiBInRIkTosQJUeKEKHFClDghSpwQtf7uF/g39vbmfJd8fn5O2Slar+d8pB8fH1N2+O+cnBAlTogSJ0SJE6LECVHihChxQpQ4IUqcECVOiBInRIkTosQJUeKEKHFClDghSpwQJU6I2qlrSn7y9SKznJycTNnZbDZTdh4eHqbsvLy8TNnZJU5OiBInRIkTosQJUeKEKHFClDghSpwQJU6IEidEiROixAlR4oQocUKUOCFKnBAlTojaqZsQZjk+Pp62VfsP/WVZpuzs7+9P2Tk6OpqyU/s9/x+cnBAlTogSJ0SJE6LECVHihChxQpQ4IUqcECVOiBInRIkTosQJUeKEKHFClDghSpwQtYwxvn64LF8/hH9gs9lM2Xl6epqyUzTG2Hp9hZMTosQJUeKEKHFClDghSpwQJU6IEidEiROixAlR4oQocUKUOCFKnBAlTogSJ0SJE6LchPCHTk9Pp+zc399P2Znl/Px8ys7z8/OUnZubmyk7q9VqdXl5OW1rBjchwI4RJ0SJE6LECVHihChxQpQ4IUqcECVOiBInRIkTosQJUeKEKHFClDghSpwQJU6IWn/3C+y62g0G19fXU3aurq6m7Ly+vk7Zub29nbIz0+Hh4V/dd3JClDghSpwQJU6IEidEiROixAlR4oQocUKUOCFKnBAlTogSJ0SJE6LECVHihChxQpSbEP7QwcHBlJ339/cpO4+Pj1N2zs7Opuz8ZG9vb39138kJUeKEKHFClDghSpwQJU6IEidEiROixAlR4oQocUKUOCFKnBAlTogSJ0SJE6LECVHihKhljPH1w2X5+iEwxRhj2fZzJydEiROixAlR4oQocUKUOCFKnBAlTogSJ0SJE6LECVHihChxQpQ4IUqcECVOiBInRIkTosQJUeKEKHFClDghSpwQJU6IEidEiROixAlR4oQocUKUOCFKnBAlTogSJ0SJE6LECVHLGOO73wHYwskJUeKEKHFClDghSpwQJU6I+gU0el3PV6rJkQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "classes=len(np.unique(y_test))\n",
    "\n",
    "#Calculating the accuracy of the network on each class of images\n",
    "class_correct = list(0. for i in range(classes))\n",
    "class_total = list(0. for i in range(classes))\n",
    "\n",
    "net.eval()\n",
    "\n",
    "with torch.no_grad():\n",
    "    for data in test_dataloader:\n",
    "        images, labels = data # Getting the test data\n",
    "\n",
    "        images=images.permute(0,3,1,2)\n",
    "\n",
    "        images=images.to(device)\n",
    "        labels=labels.to(device)\n",
    "        \n",
    "        outputs = net(images) # Getting the network output\n",
    "        _, predicted = torch.max(outputs, 1)\n",
    "        c = (predicted == labels).squeeze()\n",
    "        for i in range(5):\n",
    "            label = labels[i]\n",
    "            class_correct[label] += c[i].item()\n",
    "            class_total[label] += 1\n",
    "            if(class_total[label] == 0): print(\":C\")\n",
    "\n",
    "accuracy=list(0. for i in range(classes))\n",
    "\n",
    "for i in range(classes):\n",
    "    accuracy[i] = 100 * (class_correct[i] / (class_total[i]+1))\n",
    "\n",
    "for i in range(len(np.unique(y_test))):\n",
    "    print('Accuracy of %s (Class: %s): \\t %2d %%' % (emotions_labels_to_text[i],np.unique(y_test)[i], 100 * class_correct[i] / class_total[i]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "7e8da404",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Precision of     0 : 55 %\n",
      "Precision of     1 :  0 %\n",
      "Precision of     2 : 59 %\n",
      "Precision of     3 : 83 %\n",
      "Precision of     4 : 55 %\n",
      "Precision of     5 : 81 %\n",
      "Precision of     6 : 58 %\n",
      "Recall of     0 : 61 %\n",
      "Recall of     1 :  0 %\n",
      "Recall of     2 : 45 %\n",
      "Recall of     3 : 85 %\n",
      "Recall of     4 : 60 %\n",
      "Recall of     5 : 79 %\n",
      "Recall of     6 : 63 %\n",
      "\n",
      "\n",
      "      recall  precision   accuracy\n",
      "0  61.324786  55.728155  74.074074\n",
      "1   0.000000   0.000000   0.000000\n",
      "2  45.472837  59.788360  43.478261\n",
      "3  85.714286  83.027027  94.736842\n",
      "4  60.397554  55.633803  50.000000\n",
      "5  79.807692  81.173594  90.000000\n",
      "6  63.651316  58.814590  71.428571\n",
      "Accuracy \t 66\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Calculating the accuracy of the network on each class of images\n",
    "class_true_positive = list(0. for i in range(classes))\n",
    "class_false_positive = list(0. for i in range(classes))\n",
    "\n",
    "class_true_negative = list(0. for i in range(classes))\n",
    "class_false_negative = list(0. for i in range(classes))\n",
    "\n",
    "net.eval()\n",
    "\n",
    "for i in range(classes):\n",
    "    for data in test_dataloader:\n",
    "        images, labels = data # Getting the test data\n",
    "\n",
    "        images=images.permute(0,3,1,2)\n",
    "\n",
    "        images=images.to(device)\n",
    "        labels=labels.to(device)\n",
    "        \n",
    "        outputs = net(images) # Getting the network output\n",
    "        _, predicted = torch.max(outputs, 1)\n",
    "        for j in range(len(labels)):\n",
    "            if labels[j]==i:\n",
    "                if predicted[j]==i:\n",
    "                    class_true_positive[i]+=1\n",
    "                else:\n",
    "                    class_false_negative[i]+=1\n",
    "            else:\n",
    "                if predicted[j]==i:\n",
    "                    class_false_positive[i]+=1\n",
    "                    \n",
    "precision = list(0. for i in range(classes))\n",
    "recall = list(0. for i in range(classes))\n",
    "\n",
    "for i in range(classes):\n",
    "    precision[i]= 100 * class_true_positive[i] / (class_true_positive[i] + class_false_positive[i]+1)\n",
    "\n",
    "for i in range(len(np.unique(y_test))):\n",
    "    print('Precision of %5s : %2d %%' % (\n",
    "        np.unique(y_test)[i], precision[i]))\n",
    "\n",
    "for i in range(classes):\n",
    "    recall[i]=100 * class_true_positive[i] / (class_true_positive[i] + class_false_negative[i]+1)\n",
    "\n",
    "for i in range(len(np.unique(y_test))):\n",
    "    print('Recall of %5s : %2d %%' % (\n",
    "        np.unique(y_test)[i], recall[i]))\n",
    "\n",
    "print(\"\\n\")\n",
    "\n",
    "info={'recall':recall,'precision':precision,'accuracy':accuracy}\n",
    "infoDF=pd.DataFrame(data=info)\n",
    "print(infoDF)\n",
    "print(\"Accuracy \\t %d\"%(accurcyMean))"
   ]
  }
 ],
 "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": "3.8.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}