{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.7.6","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":13836,"databundleVersionId":1718836,"isSourceIdPinned":false,"sourceType":"competition"},{"sourceId":2660070,"sourceType":"datasetVersion","datasetId":686792},{"sourceId":228299282,"sourceType":"kernelVersion"}],"dockerImageVersionId":30034,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"package_paths = [\n    \"../input/pytorch-image-models/pytorch-image-models-master\",\n    \"/kaggle/input/cassavasimplebaseline\",\n]\nimport sys\nfor pth in package_paths:\n    sys.path.append(pth)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.178753Z","iopub.execute_input":"2025-03-19T05:13:31.179286Z","iopub.status.idle":"2025-03-19T05:13:31.184778Z","shell.execute_reply.started":"2025-03-19T05:13:31.179240Z","shell.execute_reply":"2025-03-19T05:13:31.184038Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from glob import glob\nfrom sklearn.model_selection import KFold\nimport cv2\nfrom skimage import io\nimport torch\nfrom torch import nn\nimport os\nfrom datetime import datetime\nimport time\nimport random\nimport cv2\nimport torchvision\nfrom torchvision import transforms\nimport pandas as pd\nimport numpy as np\nfrom tqdm import tqdm\n\nimport matplotlib.pyplot as plt\nfrom torch.utils.data import Dataset,DataLoader\nfrom torch.utils.data.sampler import SequentialSampler, RandomSampler\nfrom torch.cuda.amp import autocast, GradScaler\nfrom torch.nn.modules.loss import _WeightedLoss\nimport torch.nn.functional as F\n\nimport timm\n\nimport sklearn\nimport warnings\nimport joblib\nfrom sklearn.metrics import roc_auc_score, log_loss\nfrom sklearn import metrics\nimport warnings\nimport cv2\nimport pydicom\n#from efficientnet_pytorch import EfficientNet\nfrom scipy.ndimage.interpolation import zoom\n","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.186501Z","iopub.execute_input":"2025-03-19T05:13:31.186732Z","iopub.status.idle":"2025-03-19T05:13:31.199220Z","shell.execute_reply.started":"2025-03-19T05:13:31.186694Z","shell.execute_reply":"2025-03-19T05:13:31.198552Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# 今回の実験の設定 (configration)\nCFG = {\n    'fold_num': 5,\n    'seed': 719,\n    'model_arch': 'simple_cnn',\n    'model_weight_dir': '/kaggle/input/cassavasimplebaseline',\n    'img_size': 224,\n    'epochs': 1,\n    'train_bs': 32,\n    'valid_bs': 32,\n    'lr': 1e-4,\n    'dropout_rate': 0.0,\n    'num_workers': 4,\n    'accum_iter': 1, # suppoprt to do batch accumulation for backprop with effectively larger batch size\n    'verbose_step': 1,\n    'device': 'cuda:0',\n    'tta': 1,\n    'used_epochs': [0],\n    'weights': [1]\n}","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.200190Z","iopub.execute_input":"2025-03-19T05:13:31.200412Z","iopub.status.idle":"2025-03-19T05:13:31.214260Z","shell.execute_reply.started":"2025-03-19T05:13:31.200390Z","shell.execute_reply":"2025-03-19T05:13:31.213622Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train = pd.read_csv('../input/cassava-leaf-disease-classification/train.csv')\ntrain.head()","metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.215537Z","iopub.execute_input":"2025-03-19T05:13:31.215889Z","iopub.status.idle":"2025-03-19T05:13:31.244615Z","shell.execute_reply.started":"2025-03-19T05:13:31.215855Z","shell.execute_reply":"2025-03-19T05:13:31.244004Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train.label.value_counts()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.246682Z","iopub.execute_input":"2025-03-19T05:13:31.247020Z","iopub.status.idle":"2025-03-19T05:13:31.253820Z","shell.execute_reply.started":"2025-03-19T05:13:31.246988Z","shell.execute_reply":"2025-03-19T05:13:31.253154Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"> We could do stratified validation split in each fold to make each fold's train and validation set looks like the whole train set in target distributions.","metadata":{}},{"cell_type":"code","source":"submission = pd.read_csv('../input/cassava-leaf-disease-classification/sample_submission.csv')\nsubmission.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.255221Z","iopub.execute_input":"2025-03-19T05:13:31.255434Z","iopub.status.idle":"2025-03-19T05:13:31.268422Z","shell.execute_reply.started":"2025-03-19T05:13:31.255408Z","shell.execute_reply":"2025-03-19T05:13:31.267754Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Helper Functions","metadata":{}},{"cell_type":"code","source":"def 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    \ndef get_img(path):\n    im_bgr = cv2.imread(path)\n    im_rgb = im_bgr[:, :, ::-1]\n    #print(im_rgb)\n    return im_rgb\n\nimg = get_img('../input/cassava-leaf-disease-classification/train_images/1000015157.jpg')\nplt.imshow(img)\nplt.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.269524Z","iopub.execute_input":"2025-03-19T05:13:31.269722Z","iopub.status.idle":"2025-03-19T05:13:31.477904Z","shell.execute_reply.started":"2025-03-19T05:13:31.269703Z","shell.execute_reply":"2025-03-19T05:13:31.477233Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Dataset","metadata":{}},{"cell_type":"code","source":"class CassavaDataset(Dataset):\n    def __init__(\n        self, df, data_root, transforms=None, output_label=True\n    ):\n        \n        super().__init__()\n        self.df = df.reset_index(drop=True).copy()\n        self.transforms = transforms\n        self.data_root = data_root\n        self.output_label = output_label\n    \n    def __len__(self):\n        return self.df.shape[0]\n    \n    def __getitem__(self, index: int):\n        \n        # get labels\n        if self.output_label:\n            target = self.df.iloc[index]['label']\n          \n        path = \"{}/{}\".format(self.data_root, self.df.iloc[index]['image_id'])\n        \n        img  = get_img(path)\n        \n        if self.transforms:\n            img = self.transforms(image=img)['image']\n        if self.output_label == True:\n            return img, target\n        else:\n            return img","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.478929Z","iopub.execute_input":"2025-03-19T05:13:31.479146Z","iopub.status.idle":"2025-03-19T05:13:31.486003Z","shell.execute_reply.started":"2025-03-19T05:13:31.479126Z","shell.execute_reply":"2025-03-19T05:13:31.485263Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Define Train\\Validation Image Augmentations","metadata":{}},{"cell_type":"code","source":"from albumentations import (\n    RandomResizedCrop,Compose, Normalize, Resize, CenterCrop\n)\n\nfrom albumentations.pytorch import ToTensorV2\n\ndef get_train_transforms():\n    return Compose([\n            RandomResizedCrop(CFG['img_size'], CFG['img_size']),\n            Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225], max_pixel_value=255.0, p=1.0),\n            ToTensorV2(p=1.0),\n        ], p=1.)\n  \n        \ndef get_valid_transforms():\n    return Compose([\n            CenterCrop(CFG['img_size'], CFG['img_size'], p=1.),\n            Resize(CFG['img_size'], CFG['img_size']),\n            Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225], max_pixel_value=255.0, p=1.0),\n            ToTensorV2(p=1.0),\n        ], p=1.)\n\n\ndef get_inference_transforms():\n    return Compose([\n            CenterCrop(CFG['img_size'], CFG['img_size'], p=1.),\n            Resize(CFG['img_size'], CFG['img_size']),\n            Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225], max_pixel_value=255.0, p=1.0),\n            ToTensorV2(p=1.0),\n        ], p=1.)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.487264Z","iopub.execute_input":"2025-03-19T05:13:31.487584Z","iopub.status.idle":"2025-03-19T05:13:31.501469Z","shell.execute_reply.started":"2025-03-19T05:13:31.487551Z","shell.execute_reply":"2025-03-19T05:13:31.500851Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Model","metadata":{}},{"cell_type":"code","source":"class CassvaImgClassifier(nn.Module):\n    def __init__(self, n_class):\n        super().__init__()\n        # 第1層: Conv2d → ReLU\n        self.conv1 = nn.Conv2d(in_channels=3, \n                               out_channels=16, \n                               kernel_size=3, \n                               stride=1, \n                               padding=1)\n        self.relu1 = nn.ReLU()\n\n        self.pool_1 = nn.MaxPool2d(kernel_size=2, stride=2)\n\n        # 第2層: Conv2d → ReLU\n        self.conv2 = nn.Conv2d(in_channels=16, \n                               out_channels=32, \n                               kernel_size=3, \n                               stride=1, \n                               padding=1)\n        self.relu2 = nn.ReLU()\n\n        # 第2層: Conv2d → ReLU\n        self.conv3 = nn.Conv2d(in_channels=32, \n                               out_channels=128, \n                               kernel_size=3, \n                               stride=1, \n                               padding=1)\n        self.relu3 = nn.ReLU()\n\n        # Global Average Pooling (空間方向を1×1に潰す)\n        self.pool_2 = nn.AdaptiveAvgPool2d((1, 1))\n\n        # 全結合層（出力次元 = n_class）\n        self.fc = nn.Linear(in_features=128, out_features=n_class)\n\n    def forward(self, x):\n        # Conv1 → ReLU\n        x = self.conv1(x)\n        x = self.relu1(x)\n        x = self.pool_1(x)\n\n        # Conv2 → ReLU\n        x = self.conv2(x)\n        x = self.relu2(x)\n        x = self.pool_1(x)\n\n        x = self.conv3(x)\n        x = self.relu3(x)\n\n        # Global Average Poolingで空間次元を1×1に\n        x = self.pool_2(x)\n\n        # バッチ次元を残してFlatten\n        x = x.view(x.size(0), -1)\n\n        # 全結合層を通して最終出力(n_class次元)\n        x = self.fc(x)\n        return x","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.503622Z","iopub.execute_input":"2025-03-19T05:13:31.503893Z","iopub.status.idle":"2025-03-19T05:13:31.517200Z","shell.execute_reply.started":"2025-03-19T05:13:31.503857Z","shell.execute_reply":"2025-03-19T05:13:31.516310Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Main Loop","metadata":{}},{"cell_type":"code","source":"def inference_one_epoch(model, data_loader, device):\n    model.eval()\n\n    image_preds_all = []\n    \n    pbar = tqdm(enumerate(data_loader), total=len(data_loader))\n    for step, (imgs) in pbar:\n        imgs = imgs.to(device).float()\n        \n        image_preds = model(imgs)   #output = model(input)\n        image_preds_all += [torch.softmax(image_preds, 1).detach().cpu().numpy()]\n        \n    \n    image_preds_all = np.concatenate(image_preds_all, axis=0)\n    return image_preds_all","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.518286Z","iopub.execute_input":"2025-03-19T05:13:31.518535Z","iopub.status.idle":"2025-03-19T05:13:31.574130Z","shell.execute_reply.started":"2025-03-19T05:13:31.518496Z","shell.execute_reply":"2025-03-19T05:13:31.573355Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"if __name__ == '__main__':\n     # for training only, need nightly build pytorch\n\n    seed_everything(CFG['seed'])\n    \n    folds = KFold(n_splits=CFG['fold_num']).split(np.arange(train.shape[0]), train.label.values)\n    \n    for fold, (trn_idx, val_idx) in enumerate(folds):\n        # we'll train fold 0 first\n        if fold > 0:\n            break \n\n        print('Inference fold {} started'.format(fold))\n\n        valid_ = train.loc[val_idx,:].reset_index(drop=True)\n        valid_ds = CassavaDataset(valid_, '../input/cassava-leaf-disease-classification/train_images/', transforms=get_inference_transforms(), output_label=False)\n        \n        test = pd.DataFrame()\n        test['image_id'] = list(os.listdir('../input/cassava-leaf-disease-classification/test_images/'))\n        test_ds = CassavaDataset(test, '../input/cassava-leaf-disease-classification/test_images/', transforms=get_inference_transforms(), output_label=False)\n        \n        val_loader = torch.utils.data.DataLoader(\n            valid_ds, \n            batch_size=CFG['valid_bs'],\n            num_workers=CFG['num_workers'],\n            shuffle=False,\n            pin_memory=False,\n        )\n        \n        tst_loader = torch.utils.data.DataLoader(\n            test_ds, \n            batch_size=CFG['valid_bs'],\n            num_workers=CFG['num_workers'],\n            shuffle=False,\n            pin_memory=False,\n        )\n\n        device = torch.device(CFG['device'])\n        model = CassvaImgClassifier(train.label.nunique()).to(device)\n        \n        val_preds = []\n        tst_preds = []\n        \n        for i, epoch in enumerate(CFG['used_epochs']):    \n            model.load_state_dict(torch.load('{}/{}_fold_{}_{}'.format(CFG['model_weight_dir'], CFG['model_arch'], fold, epoch)))\n            model.eval()\n            \n            with torch.no_grad():\n                for _ in range(CFG['tta']):\n                    val_preds += [CFG['weights'][i]/sum(CFG['weights'])/CFG['tta']*inference_one_epoch(model, val_loader, device)]\n                    tst_preds += [CFG['weights'][i]/sum(CFG['weights'])/CFG['tta']*inference_one_epoch(model, tst_loader, device)]\n\n        val_preds = np.mean(val_preds, axis=0) \n        tst_preds = np.mean(tst_preds, axis=0) \n        \n        print('fold {} validation loss = {:.5f}'.format(fold, log_loss(valid_.label.values, val_preds)))\n        print('fold {} validation accuracy = {:.5f}'.format(fold, (valid_.label.values==np.argmax(val_preds, axis=1)).mean()))\n        \n        del model\n        torch.cuda.empty_cache()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:31.575159Z","iopub.execute_input":"2025-03-19T05:13:31.575380Z","iopub.status.idle":"2025-03-19T05:13:48.621682Z","shell.execute_reply.started":"2025-03-19T05:13:31.575359Z","shell.execute_reply":"2025-03-19T05:13:48.620735Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"test['label'] = np.argmax(tst_preds, axis=1)\ntest.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:48.623590Z","iopub.execute_input":"2025-03-19T05:13:48.623992Z","iopub.status.idle":"2025-03-19T05:13:48.635597Z","shell.execute_reply.started":"2025-03-19T05:13:48.623949Z","shell.execute_reply":"2025-03-19T05:13:48.634835Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"test.to_csv('submission.csv', index=False)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-03-19T05:13:48.636614Z","iopub.execute_input":"2025-03-19T05:13:48.636893Z","iopub.status.idle":"2025-03-19T05:13:48.648642Z","shell.execute_reply.started":"2025-03-19T05:13:48.636854Z","shell.execute_reply":"2025-03-19T05:13:48.647982Z"}},"outputs":[],"execution_count":null}]}