{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# **Mohammed Bilal Shakeel**\n# **ID:- 210051726**","metadata":{"papermill":{"duration":0.018867,"end_time":"2023-10-07T08:29:02.095914","exception":false,"start_time":"2023-10-07T08:29:02.077047","status":"completed"},"tags":[]}},{"cell_type":"code","source":"!pip install py7zr","metadata":{"_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_kg_hide-output":true,"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","papermill":{"duration":12.09222,"end_time":"2023-10-07T08:29:14.206293","exception":false,"start_time":"2023-10-07T08:29:02.114073","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:10:18.078750Z","iopub.execute_input":"2023-10-17T14:10:18.079127Z","iopub.status.idle":"2023-10-17T14:10:26.323038Z","shell.execute_reply.started":"2023-10-17T14:10:18.079097Z","shell.execute_reply":"2023-10-17T14:10:26.321856Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np\nimport pandas as pd\nimport os\nimport copy\nimport matplotlib.pyplot as plt\nfrom PIL import Image\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nimport torchvision.transforms as transforms\nfrom torchvision import models\nimport torchvision\nfrom torch.utils.data import DataLoader, Dataset, Subset\nfrom sklearn.model_selection import train_test_split\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\nimport warnings\nwarnings.filterwarnings(\"ignore\")\nfrom py7zr import unpack_7zarchive\nimport shutil\nshutil.register_unpack_format('7zip', ['.7z'], unpack_7zarchive)","metadata":{"papermill":{"duration":2.537902,"end_time":"2023-10-07T08:29:16.766436","exception":false,"start_time":"2023-10-07T08:29:14.228534","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:10:32.503688Z","iopub.execute_input":"2023-10-17T14:10:32.504774Z","iopub.status.idle":"2023-10-17T14:10:34.804519Z","shell.execute_reply.started":"2023-10-17T14:10:32.504728Z","shell.execute_reply":"2023-10-17T14:10:34.803542Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#HYPERPARAMS \nnum_epochs = 3\nbatch_size = 32\nlr = 0.01","metadata":{"papermill":{"duration":0.027516,"end_time":"2023-10-07T08:29:16.815729","exception":false,"start_time":"2023-10-07T08:29:16.788213","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:10:34.806145Z","iopub.execute_input":"2023-10-17T14:10:34.806596Z","iopub.status.idle":"2023-10-17T14:10:34.811661Z","shell.execute_reply.started":"2023-10-17T14:10:34.806563Z","shell.execute_reply":"2023-10-17T14:10:34.810386Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class CifarLoader(Dataset):\n  def __init__(self, root_dir, csv_file = None, transform = None):\n\n    # images directory \n    self.root_dir = root_dir\n    # transformations if any\n    self.transform = transform\n\n    if csv_file:\n      # read csv file\n      self.df = pd.read_csv(csv_file)\n\n      # {class: lbl}\n      self.annot_dict = {}\n\n      for i, lbl in enumerate(self.df['label'].unique()):\n        self.annot_dict[lbl] = i\n\n      # {lbl: class}\n      self.annot_dict_reversed = {v:k for k,v in self.annot_dict.items()}\n\n      # add a column with decoded labels\n      self.df['encoded'] = self.df['label'].map(self.annot_dict)\n\n    else: \n      # if no csv file provided -> create a df, fill labels with 0's\n      self.df = pd.DataFrame({\n          'id': list(range(1, len(os.listdir(self.root_dir)) + 1)),\n          'label': np.zeros(len(os.listdir(self.root_dir))),\n          'encoded': np.zeros(len(os.listdir(self.root_dir)))\n      })\n\n\n  def __len__(self):\n    # num of records in the df\n    return len(self.df)\n\n  def __getitem__(self, index):\n    # path to the image\n    img_path = os.path.join(self.root_dir, str(self.df.iloc[index, 0]) + '.png')\n    # read image\n    image = np.array(Image.open(img_path).convert('RGB'))\n    # get label\n    y_label = torch.tensor(int(self.df.iloc[index, 2]))\n\n    # apply transformations\n    if self.transform:\n      image = self.transform(image)\n    \n    return (image, y_label)\n","metadata":{"papermill":{"duration":0.036812,"end_time":"2023-10-07T08:29:16.916431","exception":false,"start_time":"2023-10-07T08:29:16.879619","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:10:40.974599Z","iopub.execute_input":"2023-10-17T14:10:40.974979Z","iopub.status.idle":"2023-10-17T14:10:40.984256Z","shell.execute_reply.started":"2023-10-17T14:10:40.974938Z","shell.execute_reply":"2023-10-17T14:10:40.983286Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# transform tensors to normalized range form [0, 1] to [-1, 1]\nmean = np.array([0.5, 0.5, 0.5])\nstd = np.array([0.25, 0.25, 0.25])\n\ntr = transforms.Compose([\n        transforms.ToPILImage(),\n        transforms.Resize((224, 224)),  \n        transforms.ToTensor(),        \n        transforms.Normalize(mean, std)\n    ])","metadata":{"papermill":{"duration":0.029111,"end_time":"2023-10-07T08:29:16.968023","exception":false,"start_time":"2023-10-07T08:29:16.938912","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:10:42.104270Z","iopub.execute_input":"2023-10-17T14:10:42.104616Z","iopub.status.idle":"2023-10-17T14:10:42.110133Z","shell.execute_reply.started":"2023-10-17T14:10:42.104589Z","shell.execute_reply":"2023-10-17T14:10:42.108708Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# unzip train and test data\nshutil.unpack_archive('../input/cifar-10/train.7z')","metadata":{"papermill":{"duration":45.808791,"end_time":"2023-10-07T08:30:02.798151","exception":false,"start_time":"2023-10-07T08:29:16.989360","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:11:26.714586Z","iopub.execute_input":"2023-10-17T14:11:26.714950Z","iopub.status.idle":"2023-10-17T14:12:11.615985Z","shell.execute_reply.started":"2023-10-17T14:11:26.714919Z","shell.execute_reply":"2023-10-17T14:12:11.614939Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"dataset = CifarLoader(csv_file = '../input/cifar-10/trainLabels.csv', \n                      root_dir = './train',\n                      transform = tr)\n\nprint(dataset.df.head())","metadata":{"papermill":{"duration":0.072447,"end_time":"2023-10-07T08:30:02.894539","exception":false,"start_time":"2023-10-07T08:30:02.822092","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:12:11.617450Z","iopub.execute_input":"2023-10-17T14:12:11.617774Z","iopub.status.idle":"2023-10-17T14:12:11.649353Z","shell.execute_reply.started":"2023-10-17T14:12:11.617745Z","shell.execute_reply":"2023-10-17T14:12:11.648378Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"############################### SPLIT THE DATA #################################\n\nindexes = list(range(len(dataset)))\n\ntrain_indexes, val_indexes = train_test_split(indexes, test_size=0.2)\n\ndata = {}\ndata['train'] = Subset(dataset, train_indexes)\ndata['val'] = Subset(dataset, val_indexes)\n\ndataloader_train = DataLoader(dataset = data['train'], batch_size = batch_size, shuffle=False)\ndataloader_val = DataLoader(dataset = data['val'], batch_size = batch_size, shuffle=False)","metadata":{"papermill":{"duration":0.045648,"end_time":"2023-10-07T08:30:02.962372","exception":false,"start_time":"2023-10-07T08:30:02.916724","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:12:11.651628Z","iopub.execute_input":"2023-10-17T14:12:11.652204Z","iopub.status.idle":"2023-10-17T14:12:11.666496Z","shell.execute_reply.started":"2023-10-17T14:12:11.652153Z","shell.execute_reply":"2023-10-17T14:12:11.665528Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Load pretrained Resnet-152.\n\nmodel = models.resnet152(pretrained=True)\n\n# reset final fully connected layer\nnum_ftrs = model.fc.in_features\n\nmodel.classifier = nn.Sequential(\n                        nn.Linear(num_ftrs, 256),  \n                        nn.ReLU(), \n                        nn.Dropout(0.3),\n                        nn.Linear(256, 10))\n\n# copy weights for futher retraining on full train dataset\nmodel_wts = copy.deepcopy(model.state_dict())\n\n# move model to a device\nmodel = model.to(device)\n\n# loss\ncriterion = nn.CrossEntropyLoss()\n\n# all parameters are being optimized\noptimizer = optim.SGD(model.parameters(), lr=lr)","metadata":{"papermill":{"duration":6.915398,"end_time":"2023-10-07T08:30:11.457883","exception":false,"start_time":"2023-10-07T08:30:04.542485","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:16:17.097950Z","iopub.execute_input":"2023-10-17T14:16:17.098313Z","iopub.status.idle":"2023-10-17T14:16:22.381568Z","shell.execute_reply.started":"2023-10-17T14:16:17.098279Z","shell.execute_reply":"2023-10-17T14:16:22.380554Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Training\n\ndef train(model, criterion, optimizer, num_epochs, \n          dataloader_train, dataloader_val = None):\n\n  history = {\n        'loss': [],\n        'accuracy': [],\n        'val_loss': [],\n        'val_accuracy': [],\n        'best train acc': (0, 0)\n        }\n\n  best_model_wts = copy.deepcopy(model.state_dict())\n  best_acc = 0.0\n\n\n  for epoch in range(num_epochs):\n    print('Epoch {}/{}'.format(epoch + 1, num_epochs))\n\n    # set model to the training mode\n    model.train()\n\n    n_samples = 0\n    correct_train = 0\n    epoch_loss_train = 0\n    epoch_loss_val = 0\n\n\n    for i, (images, labels) in enumerate(dataloader_train):\n      # origin shape: [4, 3, 32, 32] = 4, 3, 1024\n    \n      images = images.to(device)\n      labels = labels.to(device)\n\n      # forward pass (pred)\n      pred = model(images)\n\n      # loss\n      loss = criterion(pred, labels)\n\n      # mean loss * num samples in batch\n      epoch_loss_train += labels.shape[0] * loss.item()\n\n      # empty gradients\n      optimizer.zero_grad()\n\n      # gradient (backpropagation)\n      loss.backward()\n\n      # update weights\n      optimizer.step()\n\n      # values, indexes\n      value, predicted = torch.max(pred, 1)\n\n      # += batch_size\n      n_samples += labels.shape[0]\n      # num of correctly predicted in this batch\n      correct_train += (predicted==labels).sum().item()\n\n    # train acc per epoch\n    train_acc = 100 * correct_train / n_samples\n    # train loss per epoch\n    epoch_loss_train = epoch_loss_train / n_samples\n\n    print(f'Train accuracy: {train_acc:.2f}, loss: {epoch_loss_train:.2f}')\n    #print(f'Best accuracy: {best_acc:.2f}')\n\n    history['accuracy'].append(train_acc)\n    history['loss'].append(epoch_loss_train)\n\n    # find best accuracy on training data\n    if train_acc > best_acc:\n\n      #print('*New best accuracy*')\n      best_acc = train_acc\n\n      # copy current model weights\n      best_model_wts = copy.deepcopy(model.state_dict())\n      history['best train acc'] = (epoch, best_acc)\n\n\n\n    if dataloader_val:\n\n      # evaluation mode\n      model.eval()\n\n      with torch.no_grad():\n        correct_val = 0\n        n_samples = 0\n\n        for images, labels in dataloader_val: \n          images = images.to(device)\n          labels = labels.to(device)\n\n          outputs = model(images)\n\n          loss = criterion(outputs, labels)\n          epoch_loss_val += labels.shape[0] * loss.item()\n\n          # value, index\n          _, pred = torch.max(outputs, 1)\n          n_samples += labels.shape[0]\n          correct_val += (pred==labels).sum().item()\n\n        # calculate validation accuracy and loss for each epoch\n        val_acc = 100 * correct_val / len(data['val'])\n        epoch_loss_val = epoch_loss_val / len(data['val'])\n\n        print(f'Val accuracy: {val_acc:.2f}, loss: {epoch_loss_val:.2f}')\n\n        history['val_accuracy'].append(val_acc)\n        history['val_loss'].append(epoch_loss_val)\n      \n    print('=' * 80)\n\n\n  print('Best train acc: {:2f}'.format(best_acc))\n\n  # load best weights\n  model.load_state_dict(best_model_wts)\n\n  return model, history","metadata":{"papermill":{"duration":0.118706,"end_time":"2023-10-07T08:30:11.642833","exception":false,"start_time":"2023-10-07T08:30:11.524127","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:16:22.383481Z","iopub.execute_input":"2023-10-17T14:16:22.383819Z","iopub.status.idle":"2023-10-17T14:16:22.395081Z","shell.execute_reply.started":"2023-10-17T14:16:22.383770Z","shell.execute_reply":"2023-10-17T14:16:22.394200Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model, history = train(model, criterion, optimizer, num_epochs, \n                       dataloader_train, dataloader_val)","metadata":{"papermill":{"duration":1722.843117,"end_time":"2023-10-07T08:58:54.533978","exception":false,"start_time":"2023-10-07T08:30:11.690861","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:16:24.023571Z","iopub.execute_input":"2023-10-17T14:16:24.023936Z","iopub.status.idle":"2023-10-17T14:45:43.577700Z","shell.execute_reply.started":"2023-10-17T14:16:24.023908Z","shell.execute_reply":"2023-10-17T14:45:43.576586Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Evaluation\ndef evaluate(model, dataloader):\n\n  with torch.no_grad():\n    model.eval()\n    n_correct = 0\n    n_samples = 0\n\n    for images, labels in dataloader: \n      images = images.to(device)\n      labels = labels.to(device)\n\n      outputs = model(images)\n\n      # value, index\n      v, pred = torch.max(outputs, 1)\n\n      n_samples += labels.shape[0]\n      n_correct += (pred==labels).sum().item()\n\n    acc = 100.0 * n_correct / n_samples\n    print(acc)","metadata":{"papermill":{"duration":0.039181,"end_time":"2023-10-07T08:58:55.218572","exception":false,"start_time":"2023-10-07T08:58:55.179391","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:45:44.113880Z","iopub.execute_input":"2023-10-17T14:45:44.114443Z","iopub.status.idle":"2023-10-17T14:45:44.121033Z","shell.execute_reply.started":"2023-10-17T14:45:44.114410Z","shell.execute_reply":"2023-10-17T14:45:44.120093Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"evaluate(model, dataloader_val)","metadata":{"papermill":{"duration":52.605587,"end_time":"2023-10-07T08:59:47.854832","exception":false,"start_time":"2023-10-07T08:58:55.249245","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:45:44.122499Z","iopub.execute_input":"2023-10-17T14:45:44.123115Z","iopub.status.idle":"2023-10-17T14:46:36.973385Z","shell.execute_reply.started":"2023-10-17T14:45:44.123082Z","shell.execute_reply":"2023-10-17T14:46:36.972311Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# COMPLETE DATASET \ndataloader = DataLoader(dataset = dataset, batch_size = batch_size, shuffle=False)","metadata":{"papermill":{"duration":0.038263,"end_time":"2023-10-07T08:59:47.989923","exception":false,"start_time":"2023-10-07T08:59:47.951660","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:46:36.975070Z","iopub.execute_input":"2023-10-17T14:46:36.975425Z","iopub.status.idle":"2023-10-17T14:46:36.980716Z","shell.execute_reply.started":"2023-10-17T14:46:36.975392Z","shell.execute_reply":"2023-10-17T14:46:36.979454Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"shutil.unpack_archive('../input/cifar-10/test.7z')\ndataset_test = CifarLoader(root_dir = './test',\n                      transform = tr)\n\ndataloader_test = DataLoader(dataset = dataset_test, \n                             batch_size = batch_size, shuffle=False)","metadata":{"papermill":{"duration":823.768927,"end_time":"2023-10-07T09:13:31.791209","exception":false,"start_time":"2023-10-07T08:59:48.022282","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:46:36.982390Z","iopub.execute_input":"2023-10-17T14:46:36.983125Z","iopub.status.idle":"2023-10-17T14:59:36.275822Z","shell.execute_reply.started":"2023-10-17T14:46:36.983093Z","shell.execute_reply":"2023-10-17T14:59:36.274740Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# RETRAINING MODEL \n\nmodel.load_state_dict(model_wts)\n\n# loss\ncriterion = nn.CrossEntropyLoss()\n\n# all parameters are being optimized\noptimizer = optim.SGD(model.parameters(), lr=lr)","metadata":{"papermill":{"duration":0.175783,"end_time":"2023-10-07T09:13:31.999582","exception":false,"start_time":"2023-10-07T09:13:31.823799","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:59:36.277309Z","iopub.execute_input":"2023-10-17T14:59:36.278362Z","iopub.status.idle":"2023-10-17T14:59:36.387351Z","shell.execute_reply.started":"2023-10-17T14:59:36.278327Z","shell.execute_reply":"2023-10-17T14:59:36.386422Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model, history = train(model, criterion, optimizer, 3, dataloader)","metadata":{"papermill":{"duration":1941.951723,"end_time":"2023-10-07T09:45:53.983369","exception":false,"start_time":"2023-10-07T09:13:32.031646","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T14:59:36.388647Z","iopub.execute_input":"2023-10-17T14:59:36.389199Z","iopub.status.idle":"2023-10-17T15:32:44.675658Z","shell.execute_reply.started":"2023-10-17T14:59:36.389166Z","shell.execute_reply":"2023-10-17T15:32:44.674663Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Prediction\n\npredictions = np.array([])\n\nwith torch.no_grad():\n    model.eval()\n\n    for images, _ in dataloader_test:\n      images = images.to(device)\n\n      outputs = model(images)\n\n      # value, index\n      v, pred = torch.max(outputs, 1)\n\n      pred = pred.cpu().numpy()\n\n      predictions = np.concatenate((predictions, pred), axis = None)","metadata":{"papermill":{"duration":1581.602113,"end_time":"2023-10-07T10:12:15.691800","exception":false,"start_time":"2023-10-07T09:45:54.089687","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T15:32:44.687240Z","iopub.execute_input":"2023-10-17T15:32:44.688072Z","iopub.status.idle":"2023-10-17T15:59:19.233914Z","shell.execute_reply.started":"2023-10-17T15:32:44.688041Z","shell.execute_reply":"2023-10-17T15:59:19.232859Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"dataset_test.df['encoded'] = predictions.astype(int)\ndataset_test.df['label'] = dataset_test.df['encoded'].map(dataset.annot_dict_reversed)\ndataset_test.df['label'].fillna('cat', inplace=True)\ndataset_test.df.drop(columns='encoded', inplace=True)\ndataset_test.df.to_csv('submission.csv', index=False)","metadata":{"papermill":{"duration":0.370897,"end_time":"2023-10-07T10:12:17.620640","exception":false,"start_time":"2023-10-07T10:12:17.249743","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-10-17T16:09:49.390967Z","iopub.execute_input":"2023-10-17T16:09:49.391357Z","iopub.status.idle":"2023-10-17T16:09:49.727236Z","shell.execute_reply.started":"2023-10-17T16:09:49.391329Z","shell.execute_reply":"2023-10-17T16:09:49.726020Z"},"trusted":true},"execution_count":null,"outputs":[]}]}