{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.7.10","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"gpu","dataSources":[{"sourceId":29653,"databundleVersionId":2420395,"sourceType":"competition"},{"sourceId":848739,"sourceType":"datasetVersion","datasetId":251095},{"sourceId":2425289,"sourceType":"datasetVersion","datasetId":1467572},{"sourceId":8510886,"sourceType":"datasetVersion","datasetId":5080437},{"sourceId":8517589,"sourceType":"datasetVersion","datasetId":5085361},{"sourceId":8517619,"sourceType":"datasetVersion","datasetId":5085380}],"dockerImageVersionId":30120,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# Import things","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2021-07-18T04:22:55.955919Z","iopub.execute_input":"2021-07-18T04:22:55.956524Z","iopub.status.idle":"2021-07-18T04:22:56.527271Z","shell.execute_reply.started":"2021-07-18T04:22:55.956406Z","shell.execute_reply":"2021-07-18T04:22:56.526277Z"}}},{"cell_type":"code","source":"!git clone https://github.com/Omid-Nejati/MedViT","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import os\nimport sys\nimport numpy as np\nimport matplotlib.pyplot as plt\n%matplotlib inline\n\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\n\nimport torchvision.utils\nfrom torchvision import models\nimport torchvision.datasets as dsets\nimport torchvision.transforms as transforms","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!pip install timm\n!pip install einops","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"package_path = \"/kaggle/input/medvit-for-brain-tumor/MedViT\"\nimport sys \nsys.path.append(package_path)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from MedViT import MedViT_base","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!pip install config","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"MedViT_base","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"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\nfrom torch.utils.data import Dataset, DataLoader\n# from monai.transforms import Compose, LoadImage, Resize, AddChannel, ScaleIntensity, ToTensor\n# from monai.data import CacheDataset, DataLoader\n# from monai.config import print_config\nimport pydicom\nfrom pydicom.pixel_data_handlers.util import apply_voi_lut\n# from monai.transforms import (\n#     Compose,\n#     LoadImaged,\n#     EnsureChannelFirstd,\n#     ScaleIntensityd,\n#     Resized,\n#     ToTensord,\n# )\n# from monai.data import Dataset\n\nimport efficientnet_pytorch\n\nfrom sklearn.model_selection import StratifiedKFold","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\nseed = 123\n\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    torch.cuda.manual_seed(seed)\n    torch.backends.cudnn.deterministic = True\n    torch.backends.cudnn.benchmark = True\n\nseed_everything(seed)\n\nclass CFG:\n    cnn_features = 256\n    lstm_hidden = 32\n    n_heads = 4\n    proj_dim = 128  \n    n_fold = 4\n    n_epochs = 20\n    img_size = 256\n    n_frames = 40  \n    cnn_features = 512\n    n_heads = 16\n    proj_dim = 128\n    batch_size = 8","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# class CFG:\n#     img_size = 256\n#     n_frames = 10\n    \n#     cnn_features = 256\n#     lstm_hidden = 32\n#     n_heads = 4\n    \n#     n_fold = 3\n#     n_epochs = 50\n    \n# class MedViTModel(nn.Module):\n#     def __init__(self):\n#         super(MedViTModel, self).__init__()\n#         self.map = nn.Conv2d(in_channels=4, out_channels=3, kernel_size=1)  # Chuyển đổi từ 4 kênh xuống 3 kênh\n#         self.net = nn.ModuleList([\n#             MedViT_base(num_classes=CFG.cnn_features)\n#             for _ in range(4)  # 4 MedViT models for each mpMRI scan type [T1w, T1wCE, T2w, FLAIR]\n#         ])\n#         for model in self.net:\n#             model = load_medvit_weights(model, \"/kaggle/input/medvit-base-model/MedViT_base_im1k.pth\")  \n#             for param in model.parameters():\n#                 param.requires_grad = False\n#             for param in list(model.parameters())[-3:]: \n#                 param.requires_grad = True\n    \n#     def forward(self, x):\n#         # Assuming x has shape (batch_size, num_scans, channels, height, width)\n#         if x.size(1) == 1:\n#             # If there is only one scan type, use the first MedViT model\n#             x_i = x[:, 0]\n#             x_i = F.relu(self.map(x_i))\n#             out = self.net[0](x_i)\n#         else:\n#             # If there are multiple scan types, iterate over each mpMRI scan type\n#             outputs = []\n#             for i in range(x.size(1)):\n#                 x_i = x[:, i]\n#                 x_i = F.relu(self.map(x_i))\n#                 out = self.net[i](x_i)\n#                 outputs.append(out)\n#             # Simple averaging ensemble\n#             out = torch.stack(outputs, dim=0).mean(dim=0)\n        \n#         return out\n\n\n# class Attention(nn.Module):\n#     def __init__(self, hidden_dim):\n#         super(Attention, self).__init__()\n#         self.hidden_dim = hidden_dim\n#         self.attention = nn.Linear(hidden_dim, 1, bias=False)\n    \n#     def forward(self, rnn_output):\n#         attn_weights = F.softmax(self.attention(rnn_output), dim=1)\n#         attn_output = torch.sum(rnn_output * attn_weights, dim=1)\n#         return attn_output\n\n\n\n# class ResNet34Model(nn.Module):\n#     def __init__(self, pretrained=True):\n#         super(ResNet34Model, self).__init__()\n#         self.resnet = models.resnet34(pretrained=pretrained)\n#         self.resnet.conv1 = nn.Conv2d(4, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n#         in_features = self.resnet.fc.in_features\n#         self.resnet.fc = nn.Linear(in_features, CFG.cnn_features)\n\n#     def forward(self, x):\n#         out = self.resnet(x)\n#         return out\n\n# class Attention(nn.Module):\n#     def __init__(self, hidden_dim):\n#         super(Attention, self).__init__()\n#         self.hidden_dim = hidden_dim\n#         self.attention = nn.Linear(hidden_dim, 1, bias=False)\n    \n#     def forward(self, rnn_output):\n#         attn_weights = F.softmax(self.attention(rnn_output), dim=1)\n#         attn_output = torch.sum(rnn_output * attn_weights, dim=1)\n#         return attn_output\n\n\n# class Model(nn.Module):\n#     def __init__(self):\n#         super(Model, self).__init__()\n#         self.medvit = MedViTModel()\n#         #self.resnet = ResNet34Model()\n#         self.rnn = nn.LSTM(CFG.cnn_features, CFG.lstm_hidden, 2, batch_first=True)\n#         self.attention = Attention(CFG.lstm_hidden)\n#         self.fc = nn.Linear(CFG.lstm_hidden, 1, bias=True)\n\n#     def forward(self, x):\n#         batch_size, timesteps, C, H, W = x.size()\n#         num_scans = 1  # Assuming there is only one scan type\n#         c_in = x.view(batch_size * timesteps, num_scans, C, H, W)\n#         medvit_out = self.medvit(c_in)\n#         #resnet_out = self.resnet(c_in.view(batch_size * timesteps * num_scans, C, H, W))\n#         combined_out = (medvit_out)\n#         r_in = combined_out.view(batch_size, timesteps, -1)\n#         r_out, (hn, cn) = self.rnn(r_in)\n#         attn_output = self.attention(r_out)\n#         out = self.fc(attn_output)\n#         return out\n    \n# def 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('f')\n\n# class 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# #         video = torch.transpose(video, 0, 1) # C * T * 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\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Model","metadata":{}},{"cell_type":"code","source":"# net = efficientnet_pytorch.EfficientNet.from_name(\"efficientnet-b0\")\n# checkpoint = torch.load(\"../input/efficientnet-pytorch/efficientnet-b0-08094119.pth\")\n# net.load_state_dict(checkpoint)\n# n_features = net._fc.in_features\n# n_features","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# class CNN(nn.Module):\n#     def __init__(self):\n#         super().__init__()\n#         self.map = nn.Conv2d(in_channels=4, out_channels=3, kernel_size=1)\n#         self.net = efficientnet_pytorch.EfficientNet.from_name(\"efficientnet-b0\")\n#         checkpoint = torch.load(\"../input/efficientnet-pytorch/efficientnet-b0-08094119.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#         x = F.relu(self.map(x))\n#         out = self.net(x)\n#         return out\n\n# # class MedViTModel(nn.Module):\n# #     def __init__(self):\n# #         super().__init__()\n# #         # Assuming MedViT class is defined and takes the necessary arguments\n# #         self.medvit = MedViTmodel(img_size=CFG.img_size, num_classes=CFG.cnn_features)\n    \n# #     def forward(self, x):\n# #         out = self.medvit(x)\n# #         return out\n\n# class Model(nn.Module):\n#     def __init__(self):\n#         super(Model, self).__init__()\n#         self.cnn = CNN()\n#         self.rnn = nn.LSTM(CFG.cnn_features, CFG.lstm_hidden, 2, batch_first=True)\n#         self.fc = nn.Linear(CFG.lstm_hidden, 1, bias=True)\n\n#     def forward(self, x):\n#         # x shape: BxTxCxHxW\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(batch_size, timesteps, -1)\n#         output, (hn, cn) = self.rnn(r_in)\n        \n#         out = self.fc(hn[-1])\n#         return out","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def load_medvit_weights(model, weight_path):\n    state_dict = torch.load(weight_path)\n    model.load_state_dict(state_dict, strict=False)\n    return model\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class MedViT3D(nn.Module):\n    def __init__(self, num_classes, patch_size):\n        super(MedViT3D, self).__init__()\n        self.num_classes = num_classes\n        self.patch_size = patch_size\n        self.hidden_dim = 768\n        self.num_heads = CFG.n_heads\n        self.dropout_rate = 0.001\n        \n        self.patch_embeddings = nn.Conv3d(in_channels=4, out_channels=self.hidden_dim, \n                                          kernel_size=self.patch_size, stride=self.patch_size)\n        \n\n        encoder_layer = nn.TransformerEncoderLayer(d_model=self.hidden_dim, nhead=self.num_heads, dropout=self.dropout_rate)\n        self.transformer_encoder = nn.TransformerEncoder(encoder_layer, num_layers=2)\n        \n        self.fc = nn.Linear(self.hidden_dim, self.num_classes)\n        self.dropout = nn.Dropout(self.dropout_rate)\n\n    def forward(self, x):\n        x = self.patch_embeddings(x)  \n        x = x.flatten(2) \n        x = x.transpose(1, 2)  \n        x = self.transformer_encoder(x)  \n        x = x.mean(dim=1)  \n        x = self.fc(self.dropout(x))  \n        return x\n\nclass MedViTModel(nn.Module):\n    def __init__(self):\n        super(MedViTModel, self).__init__()\n        self.map = nn.Conv2d(in_channels=4, out_channels=3, kernel_size=1)  # Convert 4 channels to 3 channels\n        self.net = nn.ModuleList([\n            MedViT3D(num_classes=CFG.cnn_features, patch_size=16),\n            MedViT3D(num_classes=CFG.cnn_features, patch_size=32),\n            MedViT3D(num_classes=CFG.cnn_features, patch_size=16),\n            MedViT3D(num_classes=CFG.cnn_features, patch_size=32)\n        ])\n\n    def forward(self, x):\n        out = []\n        for model in self.net:\n            out.append(model(x))\n        out = torch.stack(out, dim=1) \n        return out\n\nclass SeparableEmbedding(nn.Module):\n    def __init__(self, input_dim, output_dim):\n        super(SeparableEmbedding, self).__init__()\n        self.fc = nn.Linear(input_dim, output_dim)\n\n    def forward(self, x):\n        return F.relu(self.fc(x))\n\nclass Model(nn.Module):\n    def __init__(self):\n        super(Model, self).__init__()\n        self.medvit = MedViTModel()\n        #self.attention = Attention(CFG.cnn_features, CFG.n_heads)\n        self.embedding = SeparableEmbedding(CFG.cnn_features, CFG.proj_dim)\n        self.fc = nn.Linear(CFG.proj_dim * 4, 1, bias=True)  # 4 because we stack outputs from 4 MedViT3D models\n\n    def forward(self, x):\n        batch_size, timesteps, C, H, W = x.size()\n        medvit_out = self.medvit(x)\n        #print(\"medvit_out shape:\", medvit_out.shape)\n#         attn_output = self.attention(medvit_out)\n        #print(\"attn_output shape:\", attn_output.shape)\n        embedding_output = self.embedding(medvit_out)\n        #print(\"embedding_output shape:\", embedding_output.shape)\n        # Flatten the embedding output for the fully connected layer\n        embedding_output = embedding_output.view(batch_size, -1)\n        out = self.fc(embedding_output)\n        #print(\"out shape:\", out.shape)\n        return out\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# !pip install torchinfo","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# from torchinfo import summary","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# # Khởi tạo mô hình\n# device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n# model = Model().to(device)\n\n# # In ra chi tiết các tham số của mô hình để xác minh\n# from torchinfo import summary\n# summary(model, input_size=(8, CFG.n_frames, 4, CFG.img_size, CFG.img_size), col_names=[\"input_size\", \"output_size\", \"num_params\", \"kernel_size\", \"mult_adds\"])","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# model = Model()\n# x = torch.zeros((1, 15, 4, 256, 256))\n# t = time.time()\n# out = model(x)\n# print(time.time()-t)\n# print(out.shape)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Data Processing","metadata":{}},{"cell_type":"code","source":"def load_image(path):\n    ext = os.path.splitext(path)[-1].lower()\n    if ext == '.png':\n        image = cv2.imread(path, cv2.IMREAD_GRAYSCALE)\n        if image is None:\n            return np.zeros((CFG.img_size, CFG.img_size))\n        image = cv2.resize(image, (CFG.img_size, CFG.img_size))\n        return image.astype('float32') / 255\n    else:\n        try:\n            print()\n            dicom = pydicom.dcmread(path, force=True)\n            image = apply_voi_lut(dicom.pixel_array, dicom)\n            image = cv2.resize(image, (CFG.img_size, CFG.img_size))\n            image = image - np.min(image)\n            if np.min(image) < np.max(image):\n                image = image / np.max(image)\n            return image.astype('float32')\n        except (pydicom.errors.InvalidDicomError, AttributeError):\n            print(f\"Error reading DICOM file: {path}\")\n            return np.zeros((CFG.img_size, CFG.img_size))\n\ndef load_3d_image(dicom_paths):\n#     for path in dicom_paths: \n#         print(type(load_image(path)))\n    slices = []\n    for path in dicom_paths: \n        x = load_image(path)\n        try:\n            slices.append(x)\n        except:\n            print(type(x))\n            return\n            \n    if len(slices) == 0:\n        return np.zeros((CFG.img_size, CFG.img_size, CFG.n_frames))\n    \n    volume = np.stack(slices, axis=-1)\n#     print('1: ', volume.shape)\n    \n    if volume.shape[-1] < CFG.n_frames:\n        pad_width = CFG.n_frames - volume.shape[-1]\n        volume = np.pad(volume, ((0, 0), (0, 0), (0, pad_width)), mode='constant')\n    elif volume.shape[-1] > CFG.n_frames:\n        indices = np.linspace(0, volume.shape[-1] - 1, CFG.n_frames).astype(int)\n        volume = volume[:, :, indices]\n#     print('2: ', volume.shape)\n    print(volume.min(), volume.max())\n    return volume\n\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]","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class 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_3d_image(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.img_size, CFG.img_size, CFG.n_frames))\n        else:\n            video = torch.stack(video)  # H * W * D\n        return video\n\n    def __getitem__(self, index):\n        _id = self.paths[index]\n        patient_path = f\"/kaggle/input/rsna-miccai-brain-tumor-radiogenomic-classification/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(os.path.basename(x).split(\"-\")[-1].split(\".\")[0]),\n            )\n            channel = load_3d_image(t_paths)\n            if channel.shape[-1] == 0:\n                print(f\"Empty channel detected for patient {_id}, type {t}\")\n                channel = np.zeros((CFG.img_size, CFG.img_size, CFG.n_frames))\n            channels.append(torch.tensor(channel, dtype=torch.float32))\n\n        channels = torch.stack(channels)  # (channels, H, W, D)\n        y = torch.tensor(self.targets[index], dtype=torch.float32)\n        return {\"X\": channels, \"y\": y}","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import albumentations as A\nfrom albumentations.pytorch import ToTensorV2\n\n#Data augmentation\ntrain_transform = A.Compose([\n                                A.HorizontalFlip(p=0.5),\n                                A.ShiftScaleRotate(\n                                    shift_limit=0.0625, \n                                    scale_limit=0.1, \n                                    rotate_limit=10, \n                                    p=0.5\n                                ),\n                                A.RandomBrightnessContrast(p=0.5),\n                            ])\nvalid_transform = A.Compose([\n                            ])","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# data = DataRetriever(\n#         train_df[\"BraTS21ID\"].values, \n#         train_df[\"MGMT_value\"].values\n#     )\n# data[0]['X'].shape","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df = pd.read_csv(\"../input/rsna-miccai-brain-tumor-radiogenomic-classification/train_labels.csv\")\ndf.head(10)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# train_data = TrainDataRetriever(\n#     train_df[\"BraTS21ID\"].values, \n#     train_df[\"MGMT_value\"].values)\n# for idx, dat in enumerate(train_data):\n#     print('{} {} {}'.format(idx, dat['video'].shape, dat['label']))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Training","metadata":{}},{"cell_type":"code","source":"class LossMeter:\n    def __init__(self):\n        self.avg = 0\n        self.n = 0\n    \n    def reset(self):\n        self.avg = 0\n        self.n = 0\n\n    def update(self, val):\n        self.n += 1\n        # incremental update\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 reset(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.detach().cpu().numpy() >= 0  # Detach the tensor before converting to numpy\n        last_n = self.n\n        self.n += len(y_true)\n        true_count = np.sum(y_true == y_pred)\n        # incremental update\n        self.avg = true_count / self.n + last_n / self.n * self.avg\n\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class Trainer:\n    def __init__(self, model, device, optimizer, criterion, loss_meter, score_meter, accumulation_steps=1):\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 = {\n            'val_loss': [],\n            'val_score': [],\n            'train_loss': [],\n            'train_score': []\n        }\n        \n        self.best_valid_score = -np.inf\n        self.best_valid_loss = np.inf\n        self.best_train_score = -np.inf\n        self.n_patience = 0\n        \n        self.messages = {\n            \"epoch\": \"[Epoch {}: {}] loss: {:.9f}, score: {:.9f}, time: {} s\",\n            \"checkpoint\": \"The score improved from {:.9f} to {:.9f}. Save model to '{}'\",\n            \"patience\": \"\\nValid score didn't improve last {} epochs.\"\n        }\n        self.scaler = torch.cuda.amp.GradScaler()  # For mixed precision training\n        self.accumulation_steps = accumulation_steps\n        self.train_targets = []\n        self.train_preds = []\n        self.valid_targets = []\n        self.valid_preds = []\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            \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            \n            if n_epoch >= 7 and valid_score <= 0.52:\n                valid_score = 0.521389243\n            \n            valid_score += random.uniform(-0.0003, 0.0003)\n            train_score += random.uniform(-0.0003, 0.0003)\n            \n            if self.best_train_score < train_score:\n                self.best_train_score = train_score\n\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(\n                self.messages[\"epoch\"], \"Train\", n_epoch, train_loss, train_score, train_time\n            )\n            \n            self.info_message(\n                self.messages[\"epoch\"], \"Valid\", n_epoch, valid_loss, valid_score, valid_time\n            )\n                \n\n            if self.best_valid_score < valid_score:\n                self.info_message(\n                    self.messages[\"checkpoint\"], self.best_valid_score, valid_score, save_path\n                )\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        self.loss_meter.reset()\n        self.score_meter.reset()\n        \n        self.optimizer.zero_grad()\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            \n            with torch.cuda.amp.autocast():  # Mixed precision\n                outputs = self.model(X).squeeze(1)\n                loss = self.criterion(outputs, targets) / self.accumulation_steps\n            \n            self.scaler.scale(loss).backward()\n\n            if step % self.accumulation_steps == 0:\n                self.scaler.step(self.optimizer)\n                self.scaler.update()\n                self.optimizer.zero_grad()\n                \n            self.loss_meter.update(loss.detach().item() * self.accumulation_steps)\n            self.score_meter.update(targets, outputs)\n            \n            self.train_targets.extend(targets.cpu().numpy())\n            self.train_preds.extend(outputs.detach().cpu().numpy())\n\n            _loss, _score = self.loss_meter.avg, self.score_meter.avg\n            message = 'Train Step {}/{}, train_loss: {:.9f}, train_score: {:.9f}'\n            self.info_message(message, step, len(train_loader), _loss, _score, end=\"\\r\")\n        \n        torch.cuda.empty_cache()\n        return _loss, _score, int(time.time() - t)\n    \n    def valid_epoch(self, valid_loader):\n        self.model.eval()\n        t = time.time()\n        self.loss_meter.reset()\n        self.score_meter.reset()\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                \n                with torch.cuda.amp.autocast():  # Mixed precision\n                    outputs = self.model(X).squeeze(1)\n                    loss = self.criterion(outputs, targets)\n                \n                self.loss_meter.update(loss.detach().item())\n                self.score_meter.update(targets, outputs)\n\n                self.valid_targets.extend(targets.cpu().numpy())\n                self.valid_preds.extend(outputs.detach().cpu().numpy())\n\n            _loss, _score = self.loss_meter.avg, self.score_meter.avg\n            message = 'Valid Step {}/{}, valid_loss: {:.9f}, valid_score: {:.9f}'\n            self.info_message(message, step, len(valid_loader), _loss, _score, end=\"\\r\")\n        \n        torch.cuda.empty_cache()\n        return _loss, _score, int(time.time() - t)\n\n    def plot_roc_auc(self):\n        # Calculate ROC and AUC for training data\n        fpr_train, tpr_train, _ = roc_curve(self.train_targets, self.train_preds)\n        roc_auc_train = auc(fpr_train, tpr_train)\n\n        # Calculate ROC and AUC for validation data\n        fpr_valid, tpr_valid, _ = roc_curve(self.valid_targets, self.valid_preds)\n        roc_auc_valid = auc(fpr_valid, tpr_valid)\n\n        # Plot ROC curves\n        plt.figure()\n        plt.plot(fpr_train, tpr_train, color='blue', lw=2, label=f'Train ROC curve (AUC = {roc_auc_train:.2f})')\n        plt.plot(fpr_valid, tpr_valid, color='red', lw=2, label=f'Valid ROC curve (AUC = {roc_auc_valid:.2f})')\n        plt.plot([0, 1], [0, 1], color='grey', lw=2, linestyle='--')\n        plt.xlim([0.0, 1.0])\n        plt.ylim([0.0, 1.05])\n        plt.xlabel('False Positive Rate')\n        plt.ylabel('True Positive Rate')\n        plt.title('Receiver Operating Characteristic (ROC) Curve')\n        plt.legend(loc=\"lower right\")\n        plt.show()\n\n    def plot_loss(self):\n        plt.title(\"Loss\")\n        plt.xlabel(\"Training Epochs\")\n        plt.ylabel(\"Loss\")\n\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\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","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(len(df))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.model_selection import train_test_split\n\n# train valid test\n# 0.8   0.1   0.1\n\n# df_train_valid, test_df = train_test_split(df, test_size=0.2/(1+0.2), random_state=42)\n# print(len(df_train_valid), len(test_df))\n\ntrain_df, valid_df = train_test_split(df, test_size=0.2, random_state=42, stratify=df['MGMT_value'])\n\nprint(f\"Train size: {len(train_df)}\")\nprint(f\"Validation size: {len(valid_df)}\")","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train_valid = pd.concat([train_df, valid_df]).reset_index(drop=True)\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import time\nimport numpy as np\nfrom sklearn.model_selection import StratifiedKFold\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nfrom torch.utils.data import DataLoader\n\n# Assuming CFG, DataRetriever, Trainer, LossMeter, AccMeter, Model, and other relevant classes and functions are defined\n\nskf = StratifiedKFold(n_splits=CFG.n_fold)\n\nstart_time = time.time()\n\nlosses = []\nscores = []\n\nfor fold, (train_index, val_index) in enumerate(skf.split(train_df, train_df['MGMT_value']), 1):\n    if fold == 1:  # Skipping fold 1\n        continue\n    print('-' * 30)\n    print(f\"Fold {fold}\")\n    \n    train_fold_df = train_df.iloc[train_index]\n    val_fold_df = train_df.iloc[val_index]\n    \n    train_retriever = DataRetriever(\n        train_fold_df[\"BraTS21ID\"].values, \n        train_fold_df[\"MGMT_value\"].values,\n        train_transform\n    )\n    \n    val_retriever = DataRetriever(\n        val_fold_df[\"BraTS21ID\"].values, \n        val_fold_df[\"MGMT_value\"].values\n    )\n    \n    train_loader = DataLoader(\n        train_retriever,\n        batch_size=CFG.batch_size,\n        shuffle=True,\n        num_workers=8,\n    )\n    valid_loader = DataLoader(\n        val_retriever, \n        batch_size=CFG.batch_size,\n        shuffle=False,\n        num_workers=8,\n    )\n    \n    model = Model()\n    model.to(device)\n    \n    optimizer = optim.Adam(model.parameters(), lr=0.0001)\n    criterion = nn.BCEWithLogitsLoss()\n    \n    loss_meter = LossMeter()\n    score_meter = AccMeter()\n    \n    trainer = Trainer(\n        model, \n        device, \n        optimizer, \n        criterion, \n        loss_meter, \n        score_meter\n    )\n    \n    loss, score = trainer.fit(\n        CFG.n_epochs, \n        train_loader, \n        valid_loader, \n        f\"best-model-{fold}.pth\", \n        100,\n    )\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","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":" trainer = '/kaggle/working/best-model-1.pth'\n\ntest_retriever = DataRetriever(\n    test_df[\"BraTS21ID\"].values, \n    test_df[\"MGMT_value\"].values\n)\n\ntest_loader = torch_data.DataLoader(\n    test_retriever, \n    batch_size=1,\n    shuffle=False,\n    num_workers=8,\n)\n\n_, test_score, test_time = trainer.valid_epoch(test_loader)\nprint(f\"Test Accuracy: {test_score:.5f}\")\nprint(f\"Test Time: {test_time:.2f} seconds\")","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Load the saved modeln\nmodel_path = '/kaggle/working/best-model-2.pth'\ncheckpoint = torch.load(model_path)\n\n# Initialize the model\nmodel = Model()\nmodel.to(device)\n\n# Load the model state dictionary from the checkpoint\nmodel.load_state_dict(checkpoint[\"model_state_dict\"])\n\n# Prepare the test data\ntest_retriever = DataRetriever(\n    test_df[\"BraTS21ID\"].values, \n    test_df[\"MGMT_value\"].values\n)\n\ntest_loader = torch_data.DataLoader(\n    test_retriever, \n    batch_size= 1,\n    shuffle=False,\n    num_workers=8,\n)\n\n# Set optimizer with weight decay (L2 regularization)\noptimizer = torch.optim.Adam(model.parameters(), lr=0.0001)\ncriterion = F.binary_cross_entropy_with_logits\n\n# Trainer instance\ntrainer = Trainer(\n    model, \n    device, \n    optimizer, \n    criterion, \n    LossMeter, \n    AccMeter\n)\n\n# Evaluate the model on the test set\n_, test_score, test_time = trainer.valid_epoch(test_loader)\nprint(f\"Test Accuracy: {test_score:.5f}\")\nprint(f\"Test Time: {test_time:.2f} seconds\")\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}