{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":29653,"databundleVersionId":2420395,"sourceType":"competition"},{"sourceId":848739,"sourceType":"datasetVersion","datasetId":251095},{"sourceId":2132855,"sourceType":"datasetVersion","datasetId":900016}],"dockerImageVersionId":30698,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"package_path = \"./input/efficientnet-pytorch/EfficientNet-PyTorch/EfficientNet-PyTorch-master/\"\nimport sys \nsys.path.append(package_path)\n\nimport os\nimport glob\nimport time\nimport random\n\nimport numpy as np\nimport pandas as pd\n\nimport pydicom\nfrom pydicom.pixel_data_handlers.util import apply_voi_lut\nimport cv2\nimport matplotlib.pyplot as plt\n\nimport torch\nfrom torch import nn\nfrom torch.utils import data as torch_data\nfrom torch.nn import functional as F\nfrom torch.utils.data import Dataset, DataLoader\n\nimport efficientnet_pytorch\n\nfrom sklearn.model_selection import StratifiedKFold\n\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")  # Prioritize GPU (cuda) if available, otherwise use CPU\nseed = 123  # Set a fixed seed for random number generators\n\n# Ensure reproducibility by setting seeds for random number generators\ndef seed_everything(seed):\n    random.seed(seed)\n    os.environ['PYTHONHASHSEED'] = str(seed)\n    np.random.seed(seed)\n    torch.manual_seed(seed)\n    if torch.cuda.is_available():\n        torch.cuda.manual_seed(seed)\n        torch.backends.cudnn.deterministic = True\n        torch.backends.cudnn.benchmark = False\n\nseed_everything(seed)\n\nclass CFG:\n    img_size = 256\n    n_frames = 10\n    \n    cnn_features = 256\n    transformer_hidden = 64\n    n_heads = 4\n    n_layers = 2\n    \n    n_fold = 5\n    n_epochs = 15\n    learning_rate = 1e-4\n    batch_size = 8\n\nclass CNN(nn.Module):\n    def __init__(self):\n        super().__init__()\n        self.map = nn.Conv2d(in_channels=4, out_channels=1, kernel_size=1)\n        self.map1 = nn.Conv2d(in_channels=4, out_channels=1, kernel_size=1)\n        self.map2 = nn.Conv2d(in_channels=4, out_channels=1, kernel_size=1)\n        self.map3 = nn.Conv2d(in_channels=4, out_channels=1, kernel_size=1)\n        self.FT = nn.Conv2d(in_channels=4, out_channels=3, kernel_size=1)\n        self.net = efficientnet_pytorch.EfficientNet.from_name(\"efficientnet-b3\")  # Using a more powerful EfficientNet variant\n        checkpoint = torch.load(\"../input/efficientnet-pytorch/efficientnet-b3-c8376fa2.pth\")\n        self.net.load_state_dict(checkpoint)\n        \n        n_features = self.net._fc.in_features\n        self.net._fc = nn.Linear(in_features=n_features, out_features=CFG.cnn_features, bias=True)\n    \n    def forward(self, x):\n        x1 = F.relu(self.map(x))\n        x2 = F.relu(self.map1(x))\n        x3 = F.relu(self.map2(x))\n        x4 = F.relu(self.map3(x))\n        x_cat = torch.cat((x1, x2, x3, x4), dim=1)\n        net_in = F.relu(self.FT(x_cat))\n        out = self.net(net_in)\n        return out\n\nclass Model(nn.Module):\n    def __init__(self):\n        super(Model, self).__init__()\n        self.cnn = CNN()\n        self.transformer = nn.Transformer(\n            d_model=CFG.cnn_features, \n            nhead=CFG.n_heads, \n            num_encoder_layers=CFG.n_layers,\n            num_decoder_layers=CFG.n_layers\n        )\n        self.fc = nn.Linear(CFG.cnn_features, 1, bias=True)\n\n    def forward(self, x):\n        batch_size, timesteps, C, H, W = x.size()\n        c_in = x.view(batch_size * timesteps, C, H, W)\n        c_out = self.cnn(c_in)\n        r_in = c_out.view(timesteps, batch_size, -1)\n        r_out = self.transformer(r_in, r_in)\n        out = self.fc(r_out[-1])\n        return out\n\ndef load_image(path):\n    image = cv2.imread(path, 0)\n    if image is None:\n        return np.zeros((CFG.img_size, CFG.img_size))\n    \n    image = cv2.resize(image, (CFG.img_size, CFG.img_size)) / 255\n    return image.astype('float32')\n\ndef uniform_temporal_subsample(x, num_samples):\n    t = len(x)\n    indices = torch.linspace(0, t - 1, num_samples)\n    indices = torch.clamp(indices, 0, t - 1).long()\n    return [x[i] for i in indices]\n\ndf = pd.read_csv(\"../input/rsna-miccai-brain-tumor-radiogenomic-classification/train_labels.csv\")\nprint(df)\n\nimport albumentations as A\nfrom albumentations.pytorch import ToTensorV2\n\ntrain_transform = A.Compose([\n    A.HorizontalFlip(p=0.5),\n    A.ShiftScaleRotate(shift_limit=0.0625, scale_limit=0.1, rotate_limit=10, p=0.5),\n    A.RandomBrightnessContrast(p=0.5),\n    A.Normalize(mean=(0.5), std=(0.5)),  # Normalizing images\n    ToTensorV2()\n])\nvalid_transform = A.Compose([\n    A.Normalize(mean=(0.5), std=(0.5)),  # Normalizing images\n    ToTensorV2()\n])\n\nclass DataRetriever(Dataset):\n    def __init__(self, paths, targets, transform=None):\n        self.paths = paths\n        self.targets = targets\n        self.transform = transform\n          \n    def __len__(self):\n        return len(self.paths)\n    \n    def read_video(self, vid_paths):\n        video = [load_image(path) for path in vid_paths]\n        if self.transform:\n            seed = random.randint(0, 99999)\n            for i in range(len(video)):\n                random.seed(seed)\n                video[i] = self.transform(image=video[i])[\"image\"]\n        \n        video = [torch.tensor(frame, dtype=torch.float32) for frame in video]\n        if len(video) == 0:\n            video = torch.zeros(CFG.n_frames, CFG.img_size, CFG.img_size)\n        else:\n            video = torch.stack(video)  # T * C * H * W\n        return video\n    \n    def __getitem__(self, index):\n        _id = self.paths[index]\n        patient_path = f\"./input/rsna-miccai-png/train/{str(_id).zfill(5)}/\"\n        channels = []\n        for t in [\"FLAIR\", \"T1w\", \"T1wCE\", \"T2w\"]:\n            t_paths = sorted(\n                glob.glob(os.path.join(patient_path, t, \"*\")), \n                key=lambda x: int(x[:-4].split(\"-\")[-1]),\n            )\n            num_samples = CFG.n_frames\n            if len(t_paths) < num_samples:\n                in_frames_path = t_paths\n            else:\n                in_frames_path = uniform_temporal_subsample(t_paths, num_samples)\n            \n            channel = self.read_video(in_frames_path)\n            if channel.shape[0] == 0:\n                print(\"1 channel empty\")\n                channel = torch.zeros(num_samples, CFG.img_size, CFG.img_size)\n            channels.append(channel)\n            \n        channels = torch.stack(channels).transpose(0, 1)\n        \n        y = torch.tensor(self.targets[index], dtype=torch.float)\n        return {\"X\": channels.float(), \"y\": y}\n\nclass LossMeter:\n    def __init__(self):\n        self.avg = 0\n        self.n = 0\n\n    def update(self, val):\n        self.n += 1\n        self.avg = val / self.n + (self.n - 1) / self.n * self.avg\n\nclass AccMeter:\n    def __init__(self):\n        self.avg = 0\n        self.n = 0\n        \n    def update(self, y_true, y_pred):\n        y_true = y_true.cpu().numpy().astype(int)\n        y_pred = y_pred.cpu().numpy() >= 0\n        last_n = self.n\n        self.n += len(y_true)\n        true_count = np.sum(y_true == y_pred)\n        self.avg = true_count / self.n + last_n / self.n * self.avg\n\nclass Trainer:\n    def __init__(self, model, device, optimizer, criterion, loss_meter, score_meter):\n        self.model = model\n        self.device = device\n        self.optimizer = optimizer\n        self.criterion = criterion\n        self.loss_meter = loss_meter\n        self.score_meter = score_meter\n        self.hist = {'val_loss':[], 'val_score':[], 'train_loss':[], 'train_score':[]}\n        self.best_valid_score = -np.inf\n        self.best_valid_loss = np.inf\n        self.n_patience = 0\n        self.messages = {\n            \"epoch\": \"[Epoch {}: {}] loss: {:.5f}, score: {:.5f}, time: {} s\",\n            \"checkpoint\": \"The score improved from {:.5f} to {:.5f}. Save model to '{}'\",\n            \"patience\": \"\\nValid score didn't improve last {} epochs.\"\n        }\n    \n    def fit(self, epochs, train_loader, valid_loader, save_path, patience):        \n        for n_epoch in range(1, epochs + 1):\n            self.info_message(\"EPOCH: {}\", n_epoch)\n            train_loss, train_score, train_time = self.train_epoch(train_loader)\n            valid_loss, valid_score, valid_time = self.valid_epoch(valid_loader)\n            self.hist['val_loss'].append(valid_loss)\n            self.hist['train_loss'].append(train_loss)\n            self.hist['val_score'].append(valid_score)\n            self.hist['train_score'].append(train_score)\n            \n            self.info_message(self.messages[\"epoch\"], \"Train\", n_epoch, train_loss, train_score, train_time)\n            self.info_message(self.messages[\"epoch\"], \"Valid\", n_epoch, valid_loss, valid_score, valid_time)\n\n            if self.best_valid_score < valid_score:\n                self.info_message(self.messages[\"checkpoint\"], self.best_valid_score, valid_score, save_path)\n                self.best_valid_score = valid_score\n                self.best_valid_loss = valid_loss\n                self.save_model(n_epoch, save_path)\n                self.n_patience = 0\n            else:\n                self.n_patience += 1\n            \n            if self.n_patience >= patience:\n                self.info_message(self.messages[\"patience\"], patience)\n                break\n                \n        return self.best_valid_loss, self.best_valid_score\n            \n    def train_epoch(self, train_loader):\n        self.model.train()\n        t = time.time()\n        train_loss = self.loss_meter()\n        train_score = self.score_meter()\n        \n        for step, batch in enumerate(train_loader, 1):\n            X = batch[\"X\"].to(self.device)\n            targets = batch[\"y\"].to(self.device)\n            self.optimizer.zero_grad()\n            outputs = self.model(X).squeeze(1)\n            loss = self.criterion(outputs, targets)\n            loss.backward()\n            train_loss.update(loss.detach().item())\n            train_score.update(targets, outputs.detach())\n            self.optimizer.step()\n            _loss, _score = train_loss.avg, train_score.avg\n            message = 'Train Step {}/{}, train_loss: {:.5f}, train_score: {:.5f}'\n            self.info_message(message, step, len(train_loader), _loss, _score, end=\"\\r\")\n        \n        return train_loss.avg, train_score.avg, int(time.time() - t)\n    \n    def valid_epoch(self, valid_loader):\n        self.model.eval()\n        t = time.time()\n        valid_loss = self.loss_meter()\n        valid_score = self.score_meter()\n\n        for step, batch in enumerate(valid_loader, 1):\n            with torch.no_grad():\n                X = batch[\"X\"].to(self.device)\n                targets = batch[\"y\"].to(self.device)\n                outputs = self.model(X).squeeze(1)\n                loss = self.criterion(outputs, targets)\n                valid_loss.update(loss.detach().item())\n                valid_score.update(targets, outputs)\n            _loss, _score = valid_loss.avg, valid_score.avg\n            message = 'Valid Step {}/{}, valid_loss: {:.5f}, valid_score: {:.5f}'\n            self.info_message(message, step, len(valid_loader), _loss, _score, end=\"\\r\")\n        \n        return valid_loss.avg, valid_score.avg, int(time.time() - t)\n    \n    def plot_loss(self):\n        plt.title(\"Loss\")\n        plt.xlabel(\"Training Epochs\")\n        plt.ylabel(\"Loss\")\n        plt.plot(self.hist['train_loss'], label=\"Train\")\n        plt.plot(self.hist['val_loss'], label=\"Validation\")\n        plt.legend()\n        plt.show()\n    \n    def plot_score(self):\n        plt.title(\"Score\")\n        plt.xlabel(\"Training Epochs\")\n        plt.ylabel(\"Acc\")\n        plt.plot(self.hist['train_score'], label=\"Train\")\n        plt.plot(self.hist['val_score'], label=\"Validation\")\n        plt.legend()\n        plt.show()\n    \n    def save_model(self, n_epoch, save_path):\n        torch.save(\n            {\n                \"model_state_dict\": self.model.state_dict(),\n                \"optimizer_state_dict\": self.optimizer.state_dict(),\n                \"best_valid_score\": self.best_valid_score,\n                \"n_epoch\": n_epoch,\n            },\n            save_path,\n        )\n    \n    @staticmethod\n    def info_message(message, *args, end=\"\\n\"):\n        print(message.format(*args), end=end)\n\nskf = StratifiedKFold(n_splits=CFG.n_fold)\nt = df['MGMT_value']\n\nstart_time = time.time()\n\nlosses = []\nscores = []\n\nfor fold, (train_index, val_index) in enumerate(skf.split(np.zeros(len(t)), t), 1):\n    print('-' * 30)\n    print(f\"Fold {fold}\")\n    \n    train_df = df.loc[train_index]\n    val_df = df.loc[val_index]\n    train_retriever = DataRetriever(\n        train_df[\"BraTS21ID\"].values, \n        train_df[\"MGMT_value\"].values,\n        train_transform\n    )\n    val_retriever = DataRetriever(\n        val_df[\"BraTS21ID\"].values, \n        val_df[\"MGMT_value\"].values,\n        valid_transform\n    )\n    train_loader = torch_data.DataLoader(\n        train_retriever,\n        batch_size=CFG.batch_size,\n        shuffle=True,\n        num_workers=4,\n    )\n    valid_loader = torch_data.DataLoader(\n        val_retriever, \n        batch_size=CFG.batch_size,\n        shuffle=False,\n        num_workers=4,\n    )\n    \n    model = Model()\n    model.to(device)\n    \n    optimizer = torch.optim.Adam(model.parameters(), lr=CFG.learning_rate)\n    criterion = F.binary_cross_entropy_with_logits\n    \n    trainer = Trainer(\n        model, \n        device, \n        optimizer, \n        criterion, \n        LossMeter, \n        AccMeter\n    )\n    loss, score = trainer.fit(\n        CFG.n_epochs, \n        train_loader, \n        valid_loader, \n        f\"best-model-{fold}.pth\", \n        5,  # Early stopping patience\n    )\n    losses.append(loss)\n    scores.append(score)\n    \n    trainer.plot_loss()\n    trainer.plot_score()\n    \nelapsed_time = time.time() - start_time\nprint('\\nTraining complete in {:.0f}m {:.0f}s'.format(elapsed_time // 60, elapsed_time % 60))\nprint('Avg loss {}'.format(np.mean(losses)))\nprint('Avg score {}'.format(np.mean(scores)))\n\n\n\n# Replaced LSTM with Transformer: Transformers have shown superior performance in various tasks due to their ability to model long-range dependencies better than RNNs.\n# Improved Data Augmentation: Added more data augmentation techniques to increase the diversity of the training set and help the model generalize better.\n# Hyperparameter Tuning: Made some adjustments to hyperparameters like learning rate.\n# Model Changes: Used a more powerful variant of EfficientNet.\n\n\n# Key Changes:\n# Transformer Integration: The LSTM in the Model class has been replaced with a Transformer to handle temporal data more effectively.\n# EfficientNet Variant: Upgraded from efficientnet-b0 to efficientnet-b3 for a potentially stronger feature extractor.\n# Normalization: Added normalization to the data augmentation pipeline.\n# Data Augmentation: Enhanced data augmentation strategies to include normalization.\n# Hyperparameters: Adjusted learning rate and batch size for potential improvements.\n# Early Stopping: Implemented early stopping with patience to prevent overfitting.\n# These improvements should help in achieving better accuracy by leveraging the strengths of Transformers, enhanced data preprocessing, and better hyperparameter settings.\n\n\n","metadata":{"_uuid":"390cc731-21b3-4a43-9a40-f71affce4617","_cell_guid":"32c8a2f5-596c-4d73-b531-25ac2b2ae18a","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2024-05-20T04:02:20.006922Z","iopub.execute_input":"2024-05-20T04:02:20.007354Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!pip install early-stopping-pytorch\n","metadata":{"execution":{"iopub.status.busy":"2024-05-17T11:56:19.089797Z","iopub.execute_input":"2024-05-17T11:56:19.090141Z","iopub.status.idle":"2024-05-17T11:56:20.67114Z","shell.execute_reply.started":"2024-05-17T11:56:19.090116Z","shell.execute_reply":"2024-05-17T11:56:20.670148Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!pip install efficientnet_pytorch\n!pip install cuba","metadata":{"execution":{"iopub.status.busy":"2024-05-20T04:00:01.943424Z","iopub.execute_input":"2024-05-20T04:00:01.94381Z","iopub.status.idle":"2024-05-20T04:00:25.635213Z","shell.execute_reply.started":"2024-05-20T04:00:01.943782Z","shell.execute_reply":"2024-05-20T04:00:25.633723Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}